As AI agents evolve beyond text and into real-world workflows, the ability to understand information and reason across multiple modalities is becoming essential. From video and audio to documents and UI screens, modern agentic systems require models that can reason across modalities with accuracy, efficiency, and enterprise readiness. 

Today, we’re delighted to announce that NVIDIA Nemotron™ 3 Nano Omni is available at launch on our Bitdeer AI Model Studio. As part of the broader NVIDIA Nemotron family, it represents a new step forward in open, production-ready multimodal reasoning models.

What is NVIDIA Nemotron 3 Nano Omni

NVIDIA Nemotron 3 Nano Omni is an open multimodal foundation model designed for production agentic AI. It unifies reasoning across video, audio, images, documents, charts, and text in a single model—eliminating the need for fragmented multimodal pipelines.

By consolidating perception and reasoning into one system, Nemotron 3 Nano Omni simplifies agent development, reduces orchestration complexity, improves efficiency and scalability, and delivers leading multimodal accuracy. Built with open weights, datasets, and recipes, it enables developers and enterprises to customize, deploy, and operate multimodal agents with full control and flexibility.

Key Specifications:

●  Model Size: 30B A3B

●  Modalities: Input (text, image, video, audio), Output (text).

●  Architecture: Hybrid Mixture of Experts (MoE) with Transformer-Mamba design

●  VIsion Encoder: CRADIO v4-H

●  Audio ENcoder: Parakeet

●  Context Length: 256k

●      Optimizations:

• Conv3D fortemporal video reasoning
• Efficient Video Sampling (EVS) for lower inference cost

●      Quantization: FP8, NVFP4

Why a Unified Multimodal Model Matters

Many enterprise AI systems still rely on stitched-together pipelines across vision, speech, OCR, and reasoning models. This approach introduces higher latency from repeated inference passes, increased operational complexity, and fragmented context across modalities.

Nemotron 3 Nano Omni addresses this by acting as a multimodal perception and reasoning layer within agent systems—enabling a unified perception → reasoning → action loop.

Enterprise use cases

Customer Service Agent: A customer service agent operates in a highly multimodal environment. It analyzes recorded customer interactions, including audio and speech transcriptions. It reasons over screen recordings of customer sessions and images such as screenshots of errors or invoices. At the same time, it reads documents like knowledge‑base articles, policies, and CRM history. Nemotron 3 Nano Omni unifies all of these signals so the agent understands not just what the customer said, but what they experienced and what the business rules allow—enabling accurate, context‑aware resolution.

Financial Analyst Agent: Financial analysis depends on more than text alone. This agent reasons across documents like financial filings and earnings transcripts, images such as charts and scanned reports, audio and speech from earnings calls, and video from investor presentations. Nemotron 3 Nano Omni ties together what executives say, how numbers are presented visually, and what the underlying documents show—producing grounded insights rather than surface‑level summaries.

Computer Use Agent: The computer use agent is one of the clearest demonstrations of unified multimodality. It analyzes video and images from screen recordings to understand UI state over time, interprets instructions and system audio cues, and reads documents like task instructions and validation policies. Nemotron 3 Nano Omni enables the agent to see the interface, understand intent, read constraints, and take the correct action—all within one reasoning loop. This collapses when perception and decision‑making are split across models.

Media and Entertainment Agent: Media workflows depend on more than transcripts alone. This agent reasons across video content, dialogue, on-screen text, and visual scene changes to support richer video and speech analysis. Nemotron 3 Nano Omni can generate dense captions that capture not just what is said, but what appears and happens on screen, while also improving video search and summarization across large content libraries. This helps media teams turn raw footage into searchable, contextualized, and production-ready assets more efficiently.

Run Nemotron 3 Nano Omni Via API on Bitdeer AI Model Studio

You can run Nemotron 3 Nano Omni on Bitdeer AI Model Studio, our serverless inference platform designed to make access to advanced foundation models simple and scalable. With a straightforward API, Model Studio allows developers and enterprises to start using models quickly without managing underlying infrastructure, reducing deployment complexity and time to value.

This makes it easier to integrate multimodal reasoning capabilities into applications and agentic workflows, while benefiting from a more flexible and efficient path from experimentation to production in just a few steps.

Get Started

1. Log in to Bitdeer AI Model Studio
2. Locate Nemotron 3 Nano Omni in the model list
3. Generate API key and start making API calls

This streamlined workflow enables rapid integration of multimodal reasoning into applications and agent systems.

Conclusion

NVIDIA Nemotron 3 Nano Omni makes multimodal AI more practical for real-world deployments. Instead of managing multiple models and pipelines, teams can focus on building agentic applications, automating workflows, and delivering better user experiences.

With availability on Bitdeer AI Model Studio, organizations can move from experiment to production faster—turning multimodal AI into measurable business impact.

OpenClaw has surpassed 290,000 stars on GitHub, making it one of the most popular open-source AI Agent frameworks of 2026. A growing number of developers and tech enthusiasts are experimenting with deploying their own AI assistants locally.

However, local deployment presents significant challenges in practice: service interruptions when the machine shuts down, complex environment configuration, API key security risks, and more — all of which severely limit the continuous availability of an AI Agent.

Deploying OpenClaw to the cloud addresses these issues at their root. This article examines the key advantages of cloud deployment, provides guidance on selecting the right large language model, and demonstrates real-world use cases for OpenClaw.

What Is OpenClaw?

OpenClaw is an open-source personal AI Agent framework built with TypeScript. Its core capabilities include:

• 24/7 Autonomous Operation: A built-in heartbeat mechanism proactively monitors tasks and takes action without requiring continuous user intervention
• 50+ Platform Integrations: Supports WhatsApp, Telegram, Discord, Slack, Gmail, GitHub, and other major platforms for unified cross-platform management
• Model Flexibility: Compatible with cloud-based LLMs such as Claude, GPT, and Grok, as well as local models via Ollama / vLLM
• Extensible Skills System: Skills are defined using Markdown + YAML, with 100+ pre-built skills available on ClawHub
• Privacy-First Design: All data is stored locally by default, with the memory system based on local Markdown files

Unlike traditional chatbots, OpenClaw features contextual memory, proactive behavior, and tool invocation capabilities — functioning more as a persistent AI collaborator than a simple Q&A interface.

Cloud Deployment vs. Local Deployment

While OpenClaw supports local installation, cloud deployment offers clear advantages for users who require stability and continuous availability.

Local deployment is suitable for short-term testing, while cloud deployment is what enables an AI Agent to deliver continuous, reliable service. An assistant that needs to be always available requires an environment that is always online.

Deployment Cost

Running OpenClaw does not require GPU resources. A basic CPU virtual machine instance (2 cores / 4GB RAM / 20GB SSD) is sufficient for stable operation. At this configuration, Bitdeer AI Cloud's on-demand pricing is approximately $0.0363/hour (around $26/month) — providing a 24/7 online AI assistant environment at a fraction of typical infrastructure costs.

Bitdeer AI Cloud, as an NVIDIA Preferred Cloud Service Provider, offers compute resources spanning multiple generations of NVIDIA GPUs including H100, H200, B200, and GB200. For Agent deployment scenarios like OpenClaw, a CPU-only instance is all that's needed. For advanced use cases such as model fine-tuning or local inference, GPU instances are available for seamless scaling. Virtual machine instances are billed on-demand — for detailed pricing, refer to the GPU Compute Pricing page.

Model inference is handled via API calls and billed per token. Bitdeer AI Cloud provides multi-vendor inference services covering text generation and other task types. For detailed inference pricing, refer to the AI Models Pricing page.

Model Selection Guide

One of OpenClaw's core strengths is model flexibility — it is not locked into any single LLM provider. Bitdeer AI Cloud currently offers 40+ models from 10+ providers, spanning text generation, visual understanding, reasoning, and image generation. All APIs are OpenAI-compatible, making model switching nearly effortless.

Below are representative text generation models suitable for OpenClaw, with pricing reference:

Scenario-Based Recommendations

• Best overall value — DeepSeek-V3.2, with the lowest combined input/output cost among flagship models and industry-leading code and reasoning capabilities
• Long-context input scenarios (e.g., document analysis, email summarization) — Qwen3-235B-A22B, with input pricing as low as $0.22 per million tokens
• Deep conversational AI — Kimi-K2.5, with excellent language comprehension for high-quality conversational output
• Ultra-low cost — The platform also offers lightweight models such as Gemma and Ministral, with input pricing starting at $0.02 per million tokens for high-frequency, low-complexity tasks

For the full model catalog and interactive demos, visit Bitdeer Model Explore.

OpenClaw integrates with Bitdeer's model service entirely through its configuration file — users simply enter their API key and model name in openclaw.json, and OpenClaw handles all API communication automatically with no additional code required. All Bitdeer model APIs are OpenAI REST-compatible, allowing OpenClaw to recognize and invoke them directly.

API keys can be generated in the Bitdeer AI Cloud console under API Keys.

Deployment Guide

Bitdeer AI has published a comprehensive deployment tutorial covering everything from creating a cloud instance and installing OpenClaw to configuring model connections and integrating with Telegram. For the full step-by-step guide, refer to:

Installing and Configuring OpenClaw on Bitdeer AI Cloud

The deployment process can be summarized in five steps:

1. Create a Cloud Instance — Provision a virtual machine on Bitdeer AI Cloud
2. Install OpenClaw — Run the official one-line installation script
3. Configure Models — Enter your Bitdeer AI API key and specify the LLM
4. Connect Telegram — Pair with a Telegram Bot for mobile access
5. Run in Background — The service runs continuously 24/7 with no manual maintenance required

Real-World Use Cases

Once deployed, OpenClaw delivers practical value across multiple dimensions. Below are several representative use cases:

Personal Productivity

• Connect to Gmail to automatically read emails, extract action items, and send daily digests on a set schedule
• Integrate with Google Calendar to create and manage events directly through conversation
• Perform real-time multilingual translation across English, Chinese, Japanese, Korean, and more

Research and Analysis

• Execute web searches, read page content, and generate structured summaries
• Extract key points and summaries from any given URL
• Conduct multi-step research with comparative analysis and structured report output

Code Generation and Execution

• Generate and execute code in a sandboxed environment, returning results directly
• Support for Python, JavaScript, and other mainstream languages for data processing and scripting
• Combine web search capabilities with code generation to automatically reference documentation and produce runnable solutions

These capabilities are powered by OpenClaw's skills system. Over 100 pre-built skills are available on ClawHub for immediate use. For custom skill development, refer to the OpenClaw Skills Documentation.

Conclusion

As a fully featured open-source AI Agent framework, OpenClaw — combined with Bitdeer AI Cloud's infrastructure and model inference services — provides a stable, secure, and always-online AI assistant solution.

Get Staarted

• Sign up for Bitdeer AI Cloud
• View the full deployment tutorial
• Visit the OpenClaw GitHub repository
• Bitdeer AI Documentation Center

Note:Pricing for models and GPU resources is subject to change. Please refer to the platform for the most up-to-date pricing.

Deploy secure, production-ready AI agents in minutes with NVIDIA NemoClaw on Bitdeer AI Cloud. In this guide, you’ll learn how to quickly set up a fully sandboxed environment and run powerful models like Moonshot AI’s Kimi K2.5 through Bitdeer AI’s high-performance inference endpoints.

What You'll Build

By the end of this guide, you'll have:

• A secure, sandboxed AI agent running on Bitdeer AI Cloud
• Access to powerful models like Kimi K2.5 through Bitdeer AI's inference API
• Full network policy controls and monitoring capabilities

Why NemoClaw on Bitdeer AI?

NemoClaw on Bitdeer AI Cloud brings together secure agent execution with scalable infrastructure and high-performance inference. While sandboxed execution with Landlock, seccomp, and network isolation is handled within the NemoClaw and OpenShell framework, Bitdeer AI Cloud can provide the underlying compute environment, enabling on-demand scaling of CPU and GPU resources with usage-based pricing. Our platform also delivers access to optimized inference endpoints for running leading models, along with network-level controls to help manage how agents interact with external services, ensuring both flexibility and operational control.

Prerequisites

NemoClaw System Requirements Minimum: 4+ vCPU, 16 GB RAM, 40 GB of free disk space, and Ubuntu 22.04 LTS.

One Bitdeer AI Cloud Requirements

• Active Bitdeer AI Cloud account
• API key from Bitdeer AI Cloud (for inference)

Bitdeer AI Instance: Select g4a.xlarge (4 vCPU, 16 GB RAM)

Important: The sandbox image is approximately 2.4 GB compressed. For instances with 16 GB RAM, you'll have sufficient memory for smooth operation.

Software Dependencies

• Node.js 20 or later
• npm 10 or later
• Docker (installed and running)
• OpenShell CLI (installed by NemoClaw)

1.1 Create Virtual Machine

1. Log in to your Bitdeer AI Cloud account

2. Navigate to Virtual Machine services

3. Click Create Instance and select:

Instance Type: g4a.xlarge (4 vCPU, 16 GB RAM)

Storage: 40 GB SSD minimum

OS: Ubuntu 22.04 LTS

Region: Choose nearest to your location

4. Complete the order and wait for instance provisioning

5. Once running, note the public IP address

1.2 Access Your Instance

Connect via SSH:

ssh ubuntu@<your-instance-ip>

Step 2: Install Docker

NemoClaw requires Docker as the container runtime:

# Update package index
sudo apt-get update

# Install Docker (service starts automatically)
sudo apt-get install -y docker.io

# Add user to docker group
sudo usermod -aG docker $USER

Important: Log out and log back in for group changes to take effect:

exit
ssh ubuntu@<your-instance-ip>

Verify Docker installation:

docker --version
docker run hello-world

Step 3: Get Bitdeer AI API Key

Before installing NemoClaw, obtain your Bitdeer AI API key:

1. Log in to Bitdeer AI Cloud

2. Navigate to Models → API Keys

3. Click Generate API Key

4. Copy the generated key and save it securely

⚠ Security Warning: Never commit API keys to version control or share them publicly.

Step 4: Install and Configure NemoClaw

Download and run the NemoClaw installer, which includes an interactive wizard for configuring your Bitdeer AI inference endpoint:

curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash

During the Installation Wizard

When prompted, configure Bitdeer AI as your inference provider:

[2/7] Configuring inference (NIM)
──────────────────────────────────────────────────

Inference options:
  1) NVIDIA Endpoints (recommended)
  2) OpenAI
  3) Other OpenAI-compatible endpoint  ← Select this
  4) Anthropic
  5) Other Anthropic-compatible endpoint
  6) Google Gemini

Choose [1]: 3

OpenAI-compatible base URL: https://api-inference.bitdeer.ai/v1
Other OpenAI-compatible endpoint API key: <paste-your-bitdeer-api-key>

Other OpenAI-compatible endpoint model []: nvidia/NVIDIA-Nemotron-3-Super-120B-A12B
# or: moonshotai/Kimi-K2.5

The wizard will:

• Install Node.js 20+ if not present
• Install npm packages
• Set up the NemoClaw CLI
• Configure Bitdeer AI as your inference provider
• Save your API key securely to ~/.nemoclaw/credentials.json
• Create your first sandbox

Post-Installation Summary

After installation completes, you should see:

──────────────────────────────────────────────────
Sandbox      my-bitdeer-agent (Landlock + seccomp + netns)
Model        nvidia/NVIDIA-Nemotron-3-Super-120B-A12B (Bitdeer AI Endpoint)
──────────────────────────────────────────────────
Run:         nemoclaw my-bitdeer-agent connect
Status:      nemoclaw my-bitdeer-agent status
Logs:        nemoclaw my-bitdeer-agent logs --follow
──────────────────────────────────────────────────

[INFO]  === Installation complete ===

If nemoclaw is not found after installation, run:

source ~/.bashrc
# or for zsh users:
source ~/.zshrc

Verify installation:

nemoclaw --version

Step 5: Connect and Chat with Your Agent

5.1 Connect to the Sandbox

nemoclaw my-bitdeer-agent connect

You'll enter the sandbox shell:

sandbox@my-bitdeer-agent:~$

5.2 Interactive TUI Mode

Launch the OpenClaw TUI for interactive chat:

openclaw tui

Send test messages and verify responses.

5.3 CLI Mode

For programmatic access or long outputs, use the CLI:

openclaw agent --agent main --local -m "Hello, how can you help me today?" --session-id test

To exit the sandbox shell:

exit

What's Next?

Now that you have your NemoClaw agent running on Bitdeer AI Cloud, here are some ways to extend it:

• Connect to Telegram

Enable mobile access to your agent:

nemoclaw my-bitdeer-agent telegram-setup

• Try Different Models

Experiment with other available models through Bitdeer AI Model Studio:

openshell inference set --provider openai --model zai-org/GLM-5

• Monitor Activity

Launch the OpenShell TUI to watch your agent in action:

openshell tui

• Customize Security

Define custom network policies to control what your agent can access.

Conclusion

With NemoClaw running on Bitdeer AI Cloud, deploying secure, production-ready AI agents becomes both practical and scalable. By combining NemoClaw’s sandboxed execution and policy framework with Bitdeer AI’s flexible infrastructure and optimized inference endpoints, you can move from setup to real-world usage in just a few steps. Whether you are experimenting with AI agents or building toward production, Bitdeer AI Cloud offers a reliable environment to run, scale, and manage your workloads with confidence.

Resources

Documentation:

• NemoClaw Docs
• OpenShell Docs
• Bitdeer AI Cloud

Community & Support:

• NemoClaw GitHub
• NemoClaw Discord
• Bitdeer AI Support

Recently, the emergence of OpenClaw drew attention to a growing demand for a new class of AI systems. Often referred to as “claws,” these systems represent long-running, autonomous, self-evolving agents that can plan and execute multi-step workflows with limited user intervention. Unlike traditional AI assistants, claws are not limited to single-turn interactions. They can access local files, interact with applications and external tools, and dynamically orchestrate sub-agents to break down complex tasks. More importantly, they are able to continuously refine strategies, decide where tasks should be executed, and optimize outcomes over time.

At the same time, this evolution introduces new challenges. As agents are granted broader access to data, tools, and infrastructure, concerns around security, privacy, and governance become significantly more critical. Questions around access control, data flow, model usage, and execution boundaries are no longer theoretical, but essential for deploying these systems safely in production.

To address the growing need for safer and more controllable agent execution, at NVIDIA GTC 2026, NVIDIA introduced NVIDIA NemoClaw for OpenClaw, an open source stack that adds privacy and security controls to long-running, autonomous agents. Built on NVIDIA Agent Toolkit, NemoClaw serves as an integration layer that enables OpenClaw agents to run within the newly introduced NVIDIA OpenShell runtime, combining open models such as NVIDIA Nemotron with controlled execution environments to make always-on agents safer and more practical to deploy.

NVIDIA OpenShell: Powering Safer Agent Execution for NemoClaw

NVIDIA OpenShell is an open-source runtime designed to make autonomous AI agents safer and more practical to deploy.  It addresses this by acting as a control layer between the agent and the underlying infrastructure, governing how the agent executes, what it can access, and where inference is routed.

Instead of relying solely on prompts or built-in guardrails inside the agent, OpenShell enforces policies at the runtime level. It helps bridge the gap between experimentation and production by allowing developers to deploy autonomous, self-evolving agents with clearer boundaries around privacy, security, and infrastructure access.

How OpenShell Is Built for Safer Agents

OpenShell uses a gateway to manage one or more isolated sandboxes where agents run. To understand how it works, it helps to look at it from two complementary angles: core components and protection layers. The components describe how the system is structured and managed, while the protection layers describe where policy enforcement is applied.

• The gateway acts as the control plane, coordinating sandbox lifecycle, authentication, providers, and policy management.
• The sandbox is the isolated runtime where the agent runs. The policy engine enforces constraints across filesystem, network, and process behavior.
• The privacy router governs inference routing, helping keep sensitive context under tighter control and directing model traffic according to defined privacy and cost policies.

These components enforce control across four protection layers: filesystem, network, process, and inference. Filesystem and process controls are locked when the sandbox is created, while network and inference controls can be updated at runtime. Together, this gives OpenShell a defense-in-depth model that supports long-running agents while maintaining clearer privacy, security, and governance boundaries.

To better understand how OpenShell enables safer autonomous agents, it helps to look at how its architecture is structured. The diagram below illustrates the key components that work together to govern execution, enforce policy, and control inference. 

OpenShell’s architecture for safer autonomous agents, illustrating the core components: the sandbox, the policy engine, and the privacy router

Deploy AI Agent Workloads with Bitdeer AI Cloud

Bitdeer AI Cloud is among the first NVIDIA Cloud Partners (NCPs) in Asia Pacific available as a model provider in the region, helping developers support the model deployment and inference behind OpenShell-based agent workloads

Bitdeer AI Cloud provides access to a wide range of high-performance NVIDIA AI infrastructure, including NVIDIA GB200 NVL72, NVIDIA HGX H200. Workloads can be deployed across multiple GPU configurations, with built-in support for load balancing and dynamic scaling to optimize performance, cost, and availability.

This enables developers to power the model deployment and inference required by agent workloads with the performance and reliability needed for production.

How to Get Started

Running models with Bitdeer AI Cloud is straightforward.  Start building on build.nvidia.com, deploy to Bitdeer AI Cloud:

First, select a model from the NVIDIA model catalog, such as Kimi K2.5 or Nemotron 3 Super, and navigate to the Deploy section.

Then, under Partner Endpoints, choose Bitdeer AI as your model provider. This allows the underlying model deployment and inference for agent workloads to be powered by Bitdeer AI Cloud, backed by scalable, production-grade NVIDIA AI infrastructure.

Once selected, you can proceed with deployment using serverless endpoints or dedicated GPU-backed infrastructure, depending on your workload requirements.

From there, OpenShell-based agents can leverage this setup to execute tasks, call tools, and operate within controlled environments, while running on scalable, production-grade compute.

By combining NVIDIA OpenShell’s secure runtime with Bitdeer AI’s infrastructure, developers can:

• Run long-running agents on dedicated, high-performance GPU clusters
• Scale workloads without managing underlying infrastructure
• Maintain stronger control over privacy, security, and inference routing

This integration bridges the gap between agent development and real-world deployment, making it easier to move from experimentation to scalable production systems.

Closing Thoughts

What stood out at NVIDIA GTC 2026 is that AI is shifting from prompt-based interactions to long-running, autonomous systems. As token consumption continues to grow, it is increasingly driven by agents themselves, reasoning over longer contexts, coordinating tools, and executing tasks over time.

This evolution brings new requirements beyond model performance. It becomes equally important how agents are executed, how data is governed, and how systems remain controlled in production environments. NVIDIA OpenShell reflects this shift by introducing a runtime layer designed for safer and more structured agent execution.

While OpenShell introduces a safer runtime layer for agent execution, Bitdeer AI Cloud serves as a model provider, delivering the scalable compute foundation needed to support real-world deployment of these systems. By providing scalable, high-performance AI compute, we aim to support developers in building and operating long-running agents within environments that prioritize privacy, security, and reliability.

Resource:

1. NVIDIA Developer Guide:  https://docs.nvidia.com/openshell/latest/about/overview.html
2. NVIDIA blog: https://developer.nvidia.com/blog/run-autonomous-self-evolving-agents-more-safely-with-nvidia-openshell/

• The Inference Inflection: GTC keynote explained massive shift in compute demand as AI moves from pre-training to real-time agentic inference.
• The NVIDIA Vera Rubin Platform: A "generational leap" featuring seven new chips (including NVIDIA Vera CPU and NVIDIA Rubin GPU) and five specialized racks designed to power the world’s largest AI factories.
• Disaggregated Inference: The newly integrated NVIDIA Groq 3 LPX rack unites with the Vera Rubin platform to deliver up to 35x higher inference throughput per megawatt.

The "Moment of Claws": NVIDIA is driving the next AI inflection point through NVIDIA OpenShell, an open-source runtime that enables autonomous agents to reason and act safely within isolated sandboxes.

The $1 Trillion Inference Inflection

At GTC 2026, the keynote underscored the massive shift from pre-training models to real-time execution, Huang revealed a staggering jump in the demand for AI infrastructure

The doubling of demand is driven by the era of agentic AI—systems that plan tasks, run tools, and validate results. Computing is no longer retrieval-based; it is growing from generative, to reasoning-oriented and finally agentic. Now, as AI Agents begin to perform different tasks through iterative processes of planning, breaking down tasks, reasoning and reflecting, the burst of inference compute demand has grown by roughly 1,000,000x in just two years.

Vera Rubin: The Foundation of the AI Factory

NVIDIA launched the Vera Rubin platform, a supercomputer where multiple racks work together as one massive, coherent system to maximize tokens per watt.

• NVIDIA Groq 3 LPX Rack: Marks a milestone in accelerated computing by enabling disaggregated inference. While the Rubin GPU handles the prefill, decode attention, and generating massive KV cache, the NVIDIA Groq LPU manages the low-latency decode phase, providing up to 10x more revenue opportunity for trillion-parameter models.
• Vera CPU Rack: The world’s first processor purpose-built for agentic AI. It delivers results twice as efficiently and 50% faster than traditional CPUs. With 88 custom Olympus cores and 1.5TB of LPDDR5X memory, it provides the single-threaded performance required for agents to run code and validate results.
• NVIDIA  BlueField-4 STX Rack: An AI-native storage infrastructure that treats data as "Context Memory." Using the NVIDIA DOCA Memos™ framework, it boosts inference throughput by up to 5x for long-context reasoning

NVIDIA OpenShell: Enabling the "Moment of Claws"

In early 2026, the emergence of OpenClaw drew attention to the demand for autonomous agents capable of executing multi-step workflows. By enabling agents to interact with local files and external tools, frameworks like OpenClaw illustrated a shift toward persistent, execution-oriented systems.

To support the safe deployment of these SOTA agents, NVIDIA is celebrating the “moment of claws” with NVIDIA OpenShell, an open-source runtime designed to govern how agents execute.

We’re also working with NVIDIA on NVIDIA NemoClaw — an open source stack that simplifies running OpenClaw always-on assistants, more safely, with a single command. As part of the NVIDIA Agent Toolkit, it installs the NVIDIA OpenShell runtime—a secure environment for running autonomous agents, and open source models like NVIDIA Nemotron.

NVIDIA OpenShell sits between the agent and the infrastructure, running any coding agent—OpenClaw, Claude Code, Cursor, or Codex—in an isolated sandbox with zero code changes. Every action is policy-enforced at the infrastructure layer, ensuring privacy and security for long-running agentic workflows.

Source: NVIDIA GTC 2026 Keynote

In tandem with these global advancements, Bitdeer AI is leading the charge for secure agentic infrastructure in the Asia Pacific region. Developers and enterprises can now find Bitdeer AI featured on build.nvidia.com as a provider for the infrastructure necessary to run the underlying "brains" of these agents.

We are also proud to highlight that Nemotron 3 Super is available on our platform. As the best-performing open model for OpenClaw, Nemotron 3 Super provides the high-level reasoning and planning capabilities required to fuel complex autonomous workflows.

Start building your own safe autonomous agent today directly from build.nvidia.com:

1. Access NVIDIA OpenShell on build.nvidia.com.
2. Seamlessly integrate frontier open model endpoints from Bitdeer AI to power your long-running agents.

Powered by NVIDIA GB200 NVL72 systems deployed in our state-of-the-art AI data centers, Bitdeer AI Cloud delivers the high-performance infrastructure required to run autonomous, self-evolving agents in NVIDIA OpenShell at scale. By combining secure runtime governance with high-performance AI cloud, we empower developers to build the next generation of AI with stronger privacy and infrastructure reliability

Source: https://blogs.nvidia.com/blog/gtc-2026-news/

Bitdeer AI Model Studio now supports NVIDIA-Nemotron-3-Super-120B-A12B, bringing a new open model for advanced reasoning, long-context understanding, and agentic AI development to the platform.

Designed for complex multi-agent applications, Nemotron 3 Super combines strong reasoning accuracy with high compute efficiency. Developers can now explore and run this model on Bitdeer AI’s high performance NVIDIA GPU infrastructure, making it easier to experiment, deploy, and scale demanding AI workloads.

What is NVIDIA Nemotron 3 Super

NVIDIA Nemotron 3 Super, part of the Nemotron 3 family of open models, is optimized for complex multi-agent applications. NVIDIA-Nemotron-3-Super-120B-A12B is a 120B-parameter model built on a hybrid Mamba-Transformer Mixture-of-Experts (MoE) architecture, activating only 12B parameters during inference. This design delivers strong performance on complex reasoning tasks while remaining efficient enough for production-scale deployment.

What makes this model particularly notable is its focus on agentic workloads. Rather than being optimized only for general chat, Nemotron 3 Super is built for scenarios where models need to reason across long contexts, call tools accurately, and support multi-step task execution.

Key Specifications:

• Model Size: 120B total parameters with 12B active parameters during inference.
• Architecture: Hybrid Mamba-2 + SSM + Transformer with Mixture-of-Experts (MoE) routing.
• Context Length: Supports up to 1M tokens for long-context reasoning.
• Accuracy: Leading accuracy on the Artificial Analysis Intelligence Index within its model size category.
• Minimum GPU Requirement: 2× H100-80GB
• Multilingual Support: English, French, German, Italian, Japanese, Spanish, Chinese

Key architecture innovations

Agentic AI systems require models capable of advanced reasoning, coding, and long-context analysis, while remaining efficient enough to operate continuously at production scale.

However, multi-agent workflows introduce additional complexity. Compared with traditional chat interactions, these systems generate significantly more tokens as agents repeatedly exchange conversation history, tool outputs, and intermediate reasoning steps across multiple turns. Over long tasks, this expanded context can increase inference costs and introduce challenges such as communication overhead and context drift.

To address these requirements, Nemotron 3 Super is built around several key architectural design principles.

• First, the hybrid Mamba-Transformer architecture improves the model’s ability to process long sequences efficiently while retaining the reasoning strengths of transformer layers.
• Second, the Mixture-of-Experts routing mechanism dynamically selects specialized expert networks for each task, enabling higher performance without proportionally increasing inference cost.
• Third, the model is designed as a fully open model stack, including open weights, training datasets, and training recipes. This allows developers to customize and fine-tune the model on their own infrastructure while maintaining control over data privacy and security.

A layer pattern diagram showing repeating blocks of Mamba-2/MoE pairs interleaved with attention layers

Source: NVIDIA

Together, these capabilities make Nemotron 3 Super particularly well suited for building complex multi-agent systems and enterprise AI workflows. And Nemotron 3 Super is optimized for tool calling, multi-step reasoning, and agent orchestration. These capabilities are essential for emerging AI agent systems, where models must interact with external tools, APIs, and data sources to complete tasks.

Benchmark performance

Nemotron 3 Super delivers leading accuracy across several agentic and reasoning benchmarks while maintaining high throughput efficiency. The model demonstrates strong capabilities in instruction following, coding, tool use, and long-context reasoning.

The benchmark results below highlight its performance across key agentic evaluation tasks.

A chart comparing Nemotron 3 Super accuracy on key benchmarks against similarly sized open models.

Source: NVIDIA 

Enterprise use cases

Nemotron 3 Super is designed for complex enterprise AI applications, particularly those that involve multi-agent reasoning and large knowledge contexts. These capabilities make the model well suited for a variety of industry use cases where advanced reasoning and scalable AI workflows are required. Below are several representative examples： 

Software Development: 

With strong coding and tool-calling capabilities, as well as the ability to work across large codebases, Nemotron 3 Super can support end-to-end software development workflows, including code generation, automated debugging, and testing.

Deep Research and Search: 

Nemotron 3 Super has demonstrated strong performance in enterprise research workflows, with capabilities well suited for comprehensive report generation and precise factual recall. This makes it a strong fit for literature review, competitive intelligence, and research automation.

Cybersecurity: High-accuracy tool calling enables autonomous agents to select the appropriate tools more reliably, reducing execution errors in high-stakes environments such as security operations and automated cybersecurity workflows.

Financial Services: 

Nemotron 3 Super can process large volumes of financial reports and long-context documents in a single workflow, helping analyst agents maintain context over extended tasks and improving both efficiency and analytical accuracy.

Run Nemotron 3 Super via API on Bitdeer AI Model Studio

Bitdeer AI Model Studio is a serverless inference platform where you can immediately access and run foundation models like Nemotron 3 Super through a simple API. 

Step 1: Select the Model

Log in to Bitdeer AI Model Studio , locate NVIDIA-Nemotron-3-Super-120B-A12B in the model list, and start using it for your AI workloads, priced at  $0.30/M in ·$0.80/M out. 

Step 2: Generate an API Key

Before using the model, you need an API key. Go to API Keys in the left navigation panel and click Generate API Key. 

Step 3: Call the Model via API

Once the API key is created, you can call the model using the Bitdeer AI inference API by including your API key in the request header.

With Model Studio and API access, developers can quickly start experimenting with Nemotron 3 Super and integrate the model into their AI applications.

Conclusion

With strong reasoning capabilities, long-context support, and an efficient hybrid architecture, NVIDIA-Nemotron-3-Super-120B-A12B provides a powerful foundation for building advanced AI applications. From complex reasoning and coding tasks to large-scale knowledge processing and agentic workflows, the model enables developers to tackle demanding workloads with greater efficiency and flexibility. 

Ready to experience NVIDIA-Nemotron-3-Super-120B-A12B in action? Explore the model on Bitdeer AI Model Studio and see how it can accelerate your next AI applications.

OpenClaw is a powerful AI agent framework that enables you to build intelligent assistants with multiple capabilities, including web search, code execution, and integration with various AI models. With Bitdeer AI Cloud, OpenClaw can be deployed online with flexible resource allocation and usage-based pricing, making it easier to run AI agents without long-term hardware constraints.

In this guide, we'll walk through the complete process of deploying OpenClaw on Bitdeer AI Cloud and configuring it to work with Telegram for an interactive AI assistant experience.

Prerequisites

Before getting started, ensure you have:

• A Bitdeer AI Cloud account
• Basic knowledge of Linux command line
• A Telegram account for bot integration

Step 1: Setting Up Bitdeer AI Cloud Instance

Ordering a Virtual Machine

First, navigate to the Bitdeer AI Cloud platform and place an order for a virtual machine instance.

Select an appropriate instance configuration based on your needs. For OpenClaw, we recommend:

• CPU: At least 2 cores
• RAM: Minimum 4GB
• Storage: At least 20GB SSD
• OS: Ubuntu 20.04 or later

Once your order is confirmed, wait for the instance to be provisioned.

Accessing Your Instance

After the instance is ready, you'll see it in your dashboard with a "Running" status.

Connect to your instance using SSH:

ssh username@your-instance-ip

Step 2: Installing OpenClaw

Quick Installation

OpenClaw provides a simple one-line installation script that handles all dependencies automatically:

curl -fsSL https://openclaw.ai/install.sh | bash -s -- --no-onboard

This command will:

• Install all required system dependencies
• Set up Node.js environment
• Install OpenClaw

Once the installation is completed, verify it:

openclaw --version

💡 Tip: If you encounter openclaw: command not found, you may need to source your .bashrc file for the current session or update your PATH:

source ~/.bashrc
# Or manually add to PATH if necessary
export PATH="$HOME/.npm-global/bin:$PATH""

Step 3: Configuring OpenClaw

Initial Setup

Run the OpenClaw setup wizard to initialize your environment:

openclaw setup

This will create the basic configuration structure for OpenClaw.

Configuring AI Models

After the setup completes, edit the OpenClaw configuration file to add your AI model settings:

vim ~/.openclaw/openclaw.json

Update the configuration with your preferred models:

{
  "agents": {
    "defaults": {
      "model": {
        "primary": "bitdeerai/moonshotai/Kimi-K2.5"
      },
      "workspace": "/home/ubuntu/.openclaw/workspace"
    }
  },
  "models": {
    "mode": "merge",
    "providers": {
      "bitdeerai": {
        "baseUrl": "https://api-inference.bitdeer.ai/v1",
        "apiKey": "your-api-key-here",
        "api": "openai-completions",
        "models": [
          {
            "id": "moonshotai/Kimi-K2.5",
            "name": "Kimi-K2.5"
          }
        ]
      }
    }
  }
}

You can find more details in model card here:

Getting Bitdeer AI Cloud API Key

To obtain your API key from Bitdeer AI Cloud:

1. Log in to your Bitdeer AI Cloud account
2. Navigate to the API Keys section in Models
3. Click "Generate API Key"
4. Copy the generated API key and save it securely

⚠️ Important: Keep your API keys secure and never commit them to version control.

Verify Model Configuration

After configuring your models, verify that they are properly set up by listing available models:

openclaw models list

This command will display configured models and their availability.

Onboarding Process

Complete the onboarding process to finalize your OpenClaw service:

openclaw onboard --flow quickstart

Setting Up Telegram Channel

During the onboarding process, select Telegram as your communication channel and provide your bot token.

After completing the onboarding process, the OpenClaw service will automatically start and run in the background.

Accessing OpenClaw Control Panel (Optional)

If you want to access the OpenClaw control panel via web interface, you can set up an SSH tunnel and access it using the gateway token.

First, find your gateway token in the configuration file:

cat ~/.openclaw/openclaw.json

Look for the gateway configuration section:

{
  "gateway": {
    "mode": "local",
    "auth": {
      "mode": "token",
      "token": "your-gateway-token"
    },
    "port": 18789
  }
}

Create an SSH tunnel from your local machine to access the OpenClaw control panel:

# Run this on your local machine
ssh -L 18789:localhost:18789 username@your-instance-ip

Then open your browser and navigate to:

http://localhost:18789?token=your-gateway-token

Use the gateway token from your configuration file to authenticate.

Step 4: Pairing with Telegram

Connect Your Telegram Chat

After the OpenClaw service is running with Telegram integration enabled, you need to pair your Telegram chat:

1. Open Telegram and find your bot
2. Send /start command to initiate the conversation
3. The bot will respond with a pairing code
4. Enter the pairing code in your OpenClaw terminal:

openclaw pairing approve telegram <pairing-code>

Interactive Chat

Once paired, you can start chatting with your AI assistant:

The bot can:

• Answer questions using the configured AI model
• Perform web searches
• Execute code snippets (with proper sandboxing)
• Handle file uploads and processing
• Maintain conversation context

Using Telegram Commands

OpenClaw provides several built-in commands:

Common commands include:

• /start - Initialize the bot
• /help - Show available commands
• /status - Check bot status

Conclusion

You now have a fully functional OpenClaw AI assistant running on Bitdeer AI Cloud with Telegram integration. This setup provides you with a powerful, customizable AI agent that can assist with various tasks while maintaining security and performance.

The combination of Bitdeer AI Cloud's reliable infrastructure and OpenClaw's flexible framework creates an ideal environment for building and deploying intelligent assistants. As you become more familiar with the platform, you can explore advanced features like:

• Advanced agent skills
• Integration with other channels

Happy building!

*Security Note: Apply appropriate access controls for sensitive actions.

Additional Resources

• Bitdeer AI Cloud Documentation
• OpenClaw GitHub Repository

Hack to Hire 2 has concluded earlier, but the use cases developed during the event continue to stand out. Rather than focusing on experimental models or theoretical benchmarks, the hackathon produced concrete system designs that address persistent business challenges across supply chains, retail operations, and document-heavy industries.

Built within a limited timeframe, these projects demonstrated how AI when paired with the right infrastructure and system architecture can be embedded directly into operational workflows. Even after the event, the use cases remain relevant as practical references for how enterprises can translate AI capabilities into real business value. In the following article, we will take a closer look at these prototype cases and further explore their practical applications and value.

Use Case 1: Market & Supplier Intelligence for Sustainable Supply Chains

Background

In the textile manufacturing supply chain, sustainability and supplier quality are no longer secondary concerns. Decision-makers must continuously assess compliance, performance, and risk across a large and evolving supplier base. However, relevant data is often distributed across internal records, supplier submissions, external documents, and public news sources.

Technical & Business Challenges

The primary challenge lies in fragmentation. Supplier information exists in multiple formats and systems, making it difficult to establish a unified, up-to-date view. As a result, sustainability compliance checks and quality assessments are largely manual, delaying critical decisions and increasing operational risk.

Solution Architecture

The proposed solution consolidates supplier data, documents, and external information through a centralized data pipeline. Within the system architecture, Bitdeer AI Cloud provides the virtual machine environment and data processing layer used to run ETL workflows, host the analytical dashboard, and support AI-driven insight generation.

Structured data is stored centrally, while AI models analyze supplier attributes to produce quality scores and contextual recommendations. The dashboard presents these insights in a unified view, enabling more informed and timely supplier management decisions

Outcome

The prototype illustrates how AI-supported analytics can transform supplier evaluation from manual reviews into a more systematic and insight-driven process, particularly valuable for sustainability-focused supply chains.

Use Case 2: Retail Store Intelligence for Planogram Execution

Background

Retail ground staff spend significant time executing product placement according to predefined planograms. Store managers must oversee large teams, while brand and product managers often lack direct visibility into how products are ultimately displayed on the shop floor.

Challenges

Manual reporting and visual checks are time-consuming and inconsistent. Without a structured feedback mechanism, it is difficult to assess execution quality or scale oversight across multiple stores.

Solution Architecture

The system digitizes the planogram workflow from instruction to verification. Product placement requirements are recorded and distributed to ground staff via a centralized interface. After execution, staff upload photos through a mobile web browser.

Within the architecture, Bitdeer AI Cloud hosts the operational dashboard and supports model inference used to analyze uploaded images. The results are automatically recorded and aggregated, allowing managers to review execution status and performance without manual follow-up.

Outcome

By combining visual evidence with automated reporting, the solution improves transparency across retail operations and reduces the manual burden on both frontline staff and management.

Use Case 3: Automated Tender Document Intelligence for SMEs

Background

Singapore’s construction industry issues thousands of government tenders each year, many consisting of hundreds of pages of documentation. For SMEs, participating in these tenders is critical but resource-intensive.

Challenges

Tender documents are typically received via email and reviewed manually. Processing a single tender can take several days, placing heavy strain on small quality assurance teams and increasing the risk of missed deadlines or errors.

Solution Architecture

The proposed pipeline automates tender intake and evaluation. Incoming emails trigger document processing workflows that extract, analyze, and assess tender requirements. AI and Retrieval-Augmented Generation (RAG) techniques reference internal document repositories such as certificates and project histories to perform eligibility checks.

Bitdeer AI Cloud provides the execution environment for these automated workflows, hosting the virtual machines that run document processing, AI inference, and the dashboard used to visualize results and audit pipeline actions

Outcome

The prototype demonstrates how AI-enabled automation can significantly shorten tender review cycles, helping SMEs respond more efficiently and compete more effectively in high-volume tender environments.

Infrastructure as an Enabler, Not the Focus

Across all three use cases, Bitdeer AI Cloud functions as the underlying execution layer that supports data pipelines, AI inference, and dashboard deployment. By providing a consistent and flexible runtime environment, teams were able to focus on workflow design and problem-solving rather than infrastructure setup.

This separation between infrastructure and application logic proved critical in enabling rapid iteration within the hackathon timeframe.

Lasting Takeaways

The solutions developed through Hack to Hire demonstrate more than short term experimentation. They provide concrete examples of how AI can be operationalized when designed around real business workflows rather than isolated models or theoretical prototypes.

Across all use cases, a consistent pattern emerges. Real value is created when AI is embedded into end to end processes and supported by infrastructure that enables rapid iteration and reliable execution. In this context, Bitdeer AI is positioned as a vertically integrated AI cloud platform that unifies AI data center infrastructure with a comprehensive AI cloud layer, providing built in tools that support the full AI lifecycle from development and training to deployment and production operations. Together, these projects illustrate a pragmatic path for enterprises adopting AI by focusing on workflow integration, operational impact, and measurable outcomes.

The rapid advancements in artificial intelligence require cutting-edge hardware capable of handling the immense computational demands of machine learning , deep learning , and AI-driven workloads. One of the most anticipated breakthroughs in this domain is the NVIDIA GB200 NVL72, a GPU designed specifically to meet the needs of high-performance computing and AI workloads. Leveraging the power of NVIDIA's innovative architectures, the NVIDIA GB200 NVL72 is poised to revolutionize AI development.

What is NVIDIA GB200 NVL72? 

The NVIDIA GB200 NVL72 is a rack-scale AI system designed for large-scale AI training and inference workloads. Built on NVIDIA’s Blackwell architecture, it features 36 Grace Blackwell Superchips, each integrating one NVIDIA Grace CPU and two Blackwell GPUs, totaling 72 GPUs and 36 CPUs in a single liquid-cooled rack.

Through a full NVLink and NVLink Switch interconnect, all GPUs operate as a tightly coupled system, allowing the platform to function as a single, massive GPU rather than a collection of discrete accelerators. This architecture significantly reduces inter-GPU communication overhead and is optimized for workloads that involve large model sizes, frequent parameter exchange, and long-running training or inference jobs that require stable, predictable performance at scale.

Technical evolution from H100 to GB200 NVL72

While H100 marked a major step forward for transformer-based AI, GB200 NVL72 builds on that foundation with system-level changes. Compared with H100, GB200 NVL72 introduces next-generation Tensor Cores and an updated second-generation Transformer Engine optimized for new FP4 precision, which enables doubling the compute capacity and model size compared to the H100’s FP8, specifically benefiting massive LLM inference. This directly benefits LLM training and inference, where matrix operations dominate runtime.

Interconnect is another key differentiator. H100 clusters rely on NVLink within nodes and high-speed networking across nodes, which can introduce latency and synchronization overhead at scale. GB200 NVL72 extends the NVLink domain to all 72 GPUs within the rack via the NVLink Switch System, offering 9x the aggregate bandwidth of previous generations and treating the entire rack as a single GPU.

Memory access patterns also improve at the system level. By enabling faster parameter sharing and collective operations, GB200 NVL72 improves utilization efficiency, especially for models that exceed the memory capacity of a single GPU. Finally, GB200 NVL72 is a liquid-cooled-native system. This shift is essential to handle the 120kW+ rack power density, ensuring sustained peak performance that would be physically impossible to achieve with traditional air-cooling.

These architectural changes are not only reflected in design principles, but also translate directly into measurable gains in real-world AI workloads. The following benchmarks illustrate how GB200 NVL72 compares with H100-based systems across large language model training, inference throughput, energy efficiency, and data processing performance.

source: from NVIDIA website

Business Use Cases: Where GB200 NVL72 Fits Best

The NVIDIA GB200 NVL72 is designed for AI workloads that require high computational throughput, efficient scaling, and sustained performance. Its rack-scale architecture supports enterprise scenarios where training speed, inference efficiency, and operational cost have a direct impact on business outcomes across multiple industries, including healthcare, finance, and AI-native technology companies.

Large-Scale AI Model Training

NVIDIA GB200 NVL72 is well suited for training large and complex AI models used in areas such as autonomous driving, healthcare, robotics, and natural language processing. With powerful Tensor Cores and increased system-level throughput, the platform reduces the time required to train deep learning models, enabling faster experimentation cycles and quicker progression from research to deployment.

High-Performance AI Inference

Beyond training, NVIDIA GB200 NVL72 supports inference workloads that demand fast and reliable response times. Applications such as video analytics, recommendation systems, and real-time decision support benefit from the platform’s ability to execute inference workloads efficiently at scale, helping enterprises deliver more responsive AI-powered services. And in  healthcare and life sciences, workloads like medical imaging models, genomics analysis, and drug discovery rely on large datasets and long-running training jobs. It can help reduce training time and improve resource utilization, enabling faster experimentation and more efficient progression from research to clinical or production environments.

Energy-Efficient AI at Scale

NVIDIA GB200 NVL72 is designed to balance performance with efficiency, making it suitable for large-scale AI deployments where power consumption and operating costs are key considerations. This is particularly relevant for cloud providers, enterprise AI platforms, and research institutions running continuous workloads, where efficiency at scale directly affects total cost of ownership.

Enabling Next-Generation AI Applications

By supporting massive datasets and increasingly complex models, NVIDIA GB200 NVL72 enables organizations to develop and deploy next-generation AI applications across industries. This includes multimodal AI systems, large shared foundation models, and AI platforms that serve multiple business units or customers. The platform allows enterprises to scale AI capabilities incrementally while maintaining performance and architectural consistency as model complexity grows.

Bitdeer AI and NVIDIA GB200 NVL72 Clusters

As part of our ongoing expansion of the NVIDIA GPU offerings on Bitdeer AI Cloud platform, the addition of NVIDIA GB200 NVL72 extends the platform’s ability to support larger and more communication-intensive AI workloads. At the same time, NVIDIA H100 remains an integral part of our platform, continuing to support a wide spectrum of training and inference use cases. Together, these GPU models form a complementary compute foundation that addresses different model sizes and deployment needs.

By offering both platforms, Bitdeer AI allows customers to select GPU configurations that best match their model size, workload characteristics, and operational requirements.

Data centers have always reflected the dominant computing paradigm of their time. For many years, enterprise software, web services, and databases shaped how infrastructure was designed and operated. These workloads emphasized reliability, steady performance, and efficient resource sharing.

Artificial intelligence introduces a fundamentally different demand profile. Training and deploying modern AI systems requires large-scale parallel computation, rapid data movement, and significantly higher power density. As a result, a new class of infrastructure has emerged alongside traditional facilities: the AI data center. This article explains how traditional data centers and AI data centers differ in design, application, and business value.

What Is a Traditional Data Center?

A traditional data center is built to support a wide variety of general-purpose computing workloads. Its architecture prioritizes flexibility and operational stability across many applications.

Core characteristics include:

• CPU-focused compute, optimized for sequential and transactional processing
• Moderate power density, typically compatible with air-based cooling
• General-purpose networking, designed primarily for client-to-server traffic
• Multi-tenant workload support, often using virtualization technologies

This model remains well-suited for enterprise systems, web platforms, and business-critical applications that require predictable performance and high availability.

What Is an AI Data Center?

An AI data center is engineered specifically to support machine learning and artificial intelligence workloads. These environments are designed around accelerators rather than general-purpose processors.

Key attributes include:

• GPU or accelerator-centric clusters designed for parallel computation
• High power and thermal density, requiring advanced cooling approaches
• High-bandwidth, low-latency interconnects to support intensive data exchange
• Tightly integrated compute, storage, and networking to minimize bottlenecks

Instead of serving many unrelated workloads, AI data centers are optimized to run fewer but extremely compute-intensive jobs efficiently. Unlike the standard Ethernet used in traditional centers, AI hubs often employ specialized InfiniBand or RoCE (RDMA over Converged Ethernet) fabrics to ensure the network doesn't become a bottleneck for the GPUs.

Data Center Tier Classification: Tier I, Tier II, and Tier III

Before comparing traditional data centers and AI data centers, it is important to understand the commonly used data center tier classification system. Tier levels are widely used to describe a data center’s reliability, redundancy, and maintainability, rather than its computing performance.

Tier I Data Center

A Tier I data center represents the most basic level of data center infrastructure. It typically relies on a single path for power and cooling, with little or no redundancy. Planned maintenance or unexpected failures usually require a full system shutdown. Tier I facilities are best suited for non-critical workloads where availability requirements are minimal.It has an expected uptime of 99.671% (28.8 hours of downtime annually).

Tier II Data Center

Tier II data centers introduce limited redundancy in key components such as power supplies and cooling systems. While backup components exist, the infrastructure generally still operates on a single distribution path. Tier II improves reliability compared to Tier I, but maintenance activities may still cause service interruptions. This tier is commonly used for workloads with moderate availability requirements. It has an expected uptime of 99.741% (22 hours of downtime annually).

Tier III Data Center

Tier III data centers are designed with concurrently maintainable infrastructure. Critical systems typically follow an N+1 redundancy model and support multiple power and cooling paths, allowing maintenance to be performed without interrupting operations. Tier III is widely adopted for enterprise-grade and cloud data centers and is suitable for mission-critical applications, including many AI workloads.It has an expected uptime of 99.982% (1.6 hours of downtime annually).

It is important to note that tier classification reflects infrastructure reliability rather than workload type or computing capability. AI data centers are often built on Tier III or higher standards, with additional design considerations for high-density power delivery and advanced cooling.

Architectural and Operational Analysis

The architectural gap between traditional and AI data centers has widened rapidly. CPUs are designed for versatility, but AI workloads demand parallel execution across thousands of identical operations. GPUs and AI accelerators fill this gap, but they also introduce new constraints.

AI data centers must address:

• Power delivery challenges, as high-density racks draw dramatically more energy
• Thermal management, where air cooling alone is insufficient
• Network congestion, since model training depends on constant GPU-to-GPU communication
• Data locality, ensuring storage and compute remain tightly coupled

Operationally, this shifts the focus from managing virtual machines to managing clusters as a single system. Operational success is no longer measured only in utilization rates, but in training time, inference latency, and performance per watt.

Below is a key comparison between Traditional Data Center and AI Data center:

Business Use Cases and Industry Impact

AI data centers are rapidly becoming a core source of competitive advantage for modern businesses. They enable organizations to move beyond isolated AI experiments toward large-scale model training, high-throughput inference, and AI agents that automate complex workflows such as customer support, software development, and data analysis. The ability to scale compute, data movement, and power density directly influences speed of innovation and time-to-market.

Across industries, AI data centers are reshaping how value is created. Financial institutions use them for real-time risk modeling and fraud detection, healthcare organizations for advanced medical imaging and diagnostics, and manufacturers for simulation, digital twins, and predictive maintenance. These AI-driven workloads are increasingly mission-critical rather than experimental.

As AI adoption accelerates, businesses are aligning infrastructure strategy around AI data centers as long-term strategic assets. Rather than serving as supporting systems, they are becoming the primary engines for innovation, operational efficiency, and competitive positioning, defining how quickly and effectively organizations can scale intelligence across their operations.

Conclusion

As AI systems grow in scale and complexity, infrastructure decisions are no longer just technical choices but strategic ones. The ability to combine cloud flexibility with purpose-built AI data center design is becoming essential for organizations that want to deploy advanced AI reliably and efficiently.

At Bitdeer AI, we build our cloud platform around this reality. By leveraging our group’s global infrastructure resources, including power capacity and data center sites, and combining them with flexible cloud capabilities, we provide an AI cloud foundation designed specifically for advanced workloads on advanced NVIDIA GPUs.This integrated approach enables scalable AI deployment while maintaining performance, energy efficiency, and operational resilience. We allow our customers to focus on building and deploying intelligence, while we manage the underlying AI infrastructure securely, sustainably, and at scale.

In the era of data-driven applications and AI-powered innovations, businesses require reliable, scalable, and high-performance database solutions. To meet this growing demand, we are excited to showcase RDS (Relational Database Service) on our AI cloud platform. Designed to simplify database management while ensuring security and scalability, our RDS solution is optimized for PostgreSQL, providing a seamless experience for developers and enterprises alike.

Key Features of Our RDS Service

1. Fully Managed PostgreSQL: Our RDS supports PostgreSQL, a reliable and highly extensible open-source database. With broad community support and a rich ecosystem of extensions for analytics, geospatial data, full-text search, time-series workloads, and more, it adapts easily to a wide range of use cases. 
2. Database Monitoring: Monitor your databases with Prometheus, Grafana, and Loki through our RDS, enabling real-time metrics, clear visualizations, and efficient log analysis to maintain performance and database health. 

3. Security & Compliance:
   • VPC (Virtual Private Cloud) Isolation: Provides network-level security by isolating database instances from unauthorized access.
   • Encryption: Protects sensitive data and prevents unauthorized access, ensuring compliance with security best practices.
4. Maintenance-Free Server Management: Offloads the burden of hardware maintenance and OS management, allowing users to focus on application development without worrying about infrastructure upkeep.

Advantages of Using Our RDS Service

Building on these capabilities, our RDS provides several clear advantages:

• Simplified Database Management: Automates routine database tasks such as monitoring, reducing operational overhead.
• Enhanced Security: Data Encryption ensures that your data remains protected from external threats.
• Optimized for Performance: With PostgreSQL’s advanced indexing, querying, and transaction capabilities, RDS provides a powerful solution for data-heavy applications.
• Automatic Failover: With clustered deployment mode, auto failover feature is enabled to prevent application disruption over primary node failure.

Use Cases

These features enable PostgreSQL on Bitdeer AI Cloud to support a wide spectrum of real-world scenarios:

• Backend Databases for Enterprise Applications : Support mission-critical systems such as ERP, CRM, and large-scale SaaS platforms with reliable, high-performance PostgreSQL infrastructure that ensures consistency, availability, and smooth transactional operations.
• Vector Database for Native RAG Workflows : Store and query high-dimensional embeddings using PostgreSQL vector extensions to power Retrieval-Augmented Generation, semantic search, recommendation engines, and other embedding-driven AI applications.
• Graph Database for Graph RAG and Relationship Intelligence : Leverage PostgreSQL graph and JSONB capabilities to model complex relationships, enabling graph queries, entity linking, fraud detection, and Graph RAG scenarios that require deep relational reasoning.
• OLAP Compatibility for Seamless Data Lake Integration : Support analytical workloads through PostgreSQL’s OLAP-friendly features, columnar extensions, and connectors to data lake systems. Enable unified storage, fast aggregations, and ETL workflows without moving data across multiple systems.
• Full-Text Search for Log Storage and Querying : Use PostgreSQL's native FTS capabilities to index and query log data at scale, enabling fast retrieval, filtering, and pattern detection for observability and operational analytics.
• Time-Series Database for Dashboards and Visualization : Store high-volume time-series data such as metrics, events, and telemetry using PostgreSQL time-series extensions. Power real-time dashboards, monitoring systems, IoT analytics, and operational insights with efficient ingestion and querying.
• Web Applications & SaaS :Provide scalable and dependable database infrastructure for CMS platforms, e-commerce systems, mobile applications, and multi-tenant SaaS products requiring predictable performance and strong data integrity.
• AI & Machine Learning Workflows : Serve as a unified data layer for AI pipelines by storing structured datasets, powering real-time analytics, and integrating with AI frameworks for tasks such as inference, recommendation, and predictive modeling.

Getting Started

Deploying a PostgreSQL instance on our RDS platform is straightforward. You can select your desired configurations, set up monitoring with Prometheus, and benefit from security features within a few clicks. 

Step1: Enter the RDS dashboard

Step2: Select Instance Type

Step3: Select PostgreSQL Extensions

For better understanding, please see the full tutorial video below.

Conclusion

Our RDS service is designed to empower businesses by providing a fully managed PostgreSQL solution that combines ease of use, scalability, and enterprise-grade security. Whether you are driving AI-powered innovation, managing enterprise workloads, or running large-scale web services, Bitdeer AI Cloud ensures optimal performance and reliability. By adopting our cloud-based RDS, organizations can streamline operations, enhance data-driven decision-making, and advance their digital transformation.

AI is becoming essential to the financial industry’s broader digital transformation goals. As data volumes rise and customer expectations shift toward immediacy, precision, and always-on service, financial institutions are recognizing AI’s potential to improve decision-making, strengthen operations, and increase the speed and quality of customer engagement. With tools, models, and deployment frameworks becoming more mature, AI is moving beyond experimentation and into production across multiple business lines.

Across banking, insurance, and asset management, AI adoption is concentrated around three major domains where it consistently demonstrates measurable value: Customer Intelligence, Risk and Decisioning, and Market and Data Intelligence. Together, these areas represent the foundation of how modern financial institutions generate insights, manage complexity, and deliver differentiated services.

Customer Intelligence and Experience Transformation

Financial institutions are shifting from product-centric engagement to data-driven, personalised customer journeys. AI enables deeper understanding of individual behaviour, intent, and financial circumstances, leading to more relevant and timely interactions.

1. Hyper-Personalization and Targeted Offerings

AI synthesizes behavioral data, spending patterns, life-stage indicators, and risk profiles to tailor credit limits, loan pricing, protection products, and investment guidance. This precision improves conversion and strengthens long-term loyalty.

2. Intelligent Customer Service and Autonomous Support

AI-powered service agents resolve inquiries, retrieve account information, and complete tasks instantly while supporting multiple languages and channels. Institutions benefit from reduced operational costs and more consistent service quality.

3. AI-Augmented Advisory and Scenario Planning

Advisors and clients use AI-generated analysis portfolio diagnostics, risk-return scenarios, economic simulations to make more informed decisions. This elevates advisory capabilities and increases planning accuracy across retail and wealth segments.

Increasingly, financial institutions view customer intelligence as a strategic differentiator, using AI to power personalized journeys at scale.

Risk, Compliance, and Core Decisioning

Effective decisioning lies at the heart of financial operations. AI strengthens the precision, speed, and consistency of these processes, enabling institutions to manage uncertainty with greater clarity.

1. Risk Management and Fraud Prevention

Machine learning models detect anomalies across transaction networks, devices, accounts, and behavioural patterns. Compared with rule-based systems, AI adapts quickly to new threats and reduces false positives which is critical for fraud, AML, and risk surveillance.

2. Automated Underwriting and Credit Decisioning

AI-based scoring models incorporate wider data sets and produce more holistic assessments of creditworthiness while supporting clear rationales. As a result, approval cycles shorten, portfolio quality improves, and new credit products become feasible.

3. Compliance and Reporting Automation

AI extracts, verifies, and reconciles information from documents, forms, and internal records to accelerate KYC reviews and reporting tasks. Intelligent validation ensures greater completeness and accuracy in financial reporting and disclosure.

AI in this domain enhances operational resilience and supports high-volume, high-accuracy decision workflows across the enterprise.

Market Intelligence and Data Automation

In a market defined by rapid information cycles and increasingly complex data streams, AI provides institutions with analytical leverage and real-time insight.

1. Predictive Analytics and Forecasting

Models anticipate shifts in liquidity demand, pricing risk, loan performance, customer churn, and macroeconomic indicators. These forecasts support better planning, capital allocation, and risk mitigation.

2. Algorithmic Trading and Market Analytics

AI identifies patterns across large, fast-moving data sources market data, news sentiment, macro signals and supports traders with improved strategy optimisation and execution.

3. Advanced Data Synthesis, Document Intelligence, and Narrative Reporting

LLM-driven systems interpret filings, contracts, statements, and regulatory documents, transforming unstructured information into structured, analysable formats. AI also generates narrative explanations of results, enabling faster reporting cycles and supporting emerging “continuous close” practices where financial insights are updated in near real time.

These capabilities give institutions a more complete and timely view of their financial and market environment.

As this overview shows, AI is reshaping nearly every function in modern banking, with impact spanning pricing, onboarding, operations, servicing, and risk. The challenge is no longer identifying use cases, it’s building the infrastructure to scale them consistently across the organization. This is where a unified AI platform becomes essential.

Figure 1. Banking Example (Source The World Economic Forum )

How Bitdeer AI Power Financial Solution

To operationalize AI at scale, financial institutions need a unified environment that integrates high-performance compute, secure data infrastructure, production-ready models, and automated workflows. 

Bitdeer AI provides a unified cloud environment that combines scalable GPU compute with secure data and model-development tools. Through AI Studio covering notebooks, RAG pipelines, and training jobs that teams can build, refine, and deploy open-source LLMs efficiently while relying on integrated databases and vector search for data-driven financial workflows.

On top of this foundation, the Agent Builder allows institutions to create custom agentic workflows using open-source models, tool integrations, and database queries. These agents automate tasks such as document understanding, KYC data extraction, validation, and summarization, helping financial teams streamline high-volume processes with greater consistency and speed.

Conclusion

AI is reshaping how financial institutions operate, powering more personalised customer experiences, more accurate decisioning, and faster, insight-driven reporting. As AI capabilities mature, the institutions that succeed will be those able to deploy them consistently and securely across their core operations. With an integrated platform designed for high-performance, production-grade AI, Bitdeer AI enables financial organizations to scale these capabilities with confidence and unlock the full value of intelligent transformation.

The Hack to Hire hackathon series, jointly organized by Bitdeer AI and GECO Asia concluded the first event earlier in September, bringing together emerging engineers, developers, and data scientists to tackle real-world business challenges through AI. Over several days, participants used Bitdeer AI Cloud’s GPU virtual machines and AI model library to build practical prototypes across multiple industries, including publishing, logistics, and consumer goods.

The event demonstrated how access to scalable computing and pre-trained AI models can dramatically shorten development cycles, transforming ideas into deployable solutions within days. It also reflected Bitdeer AI’s commitment to nurturing the next generation of AI talents while enabling enterprises to accelerate digital transformation through intelligent infrastructure.

Case1: Building Data-Driven Market Insights

Background: A Singapore-based publishing company with over three decades of history is one of the largest scientific publishers in the Asia-Pacific region. It produces more than 600 new titles and 130 journals annually in multiple languages, serving a diverse academic audience worldwide.

Challenges: The company’s teams spent extensive time manually researching and categorizing information from universities, journals, and books. Without real-time analytics or insight generation, the management struggled to identify high-performing content areas, forecast demand, or understand relationships between academic institutions and sales performance.

Solution: Participants built an AI-powered analytics platform on Bitdeer AI Cloud using natural language processing and data-mining models. The system analyzed large datasets to uncover hidden connections between topics, readership, and market trends. With configurable prompts, users could automatically generate customized insights such as emerging research areas or optimal sales channels without technical expertise.

Outcome: The prototype provided a foundation for intelligent decision-making, enabling the company to shift from manual data exploration to predictive market analysis. Executives could visualize real-time performance indicators, identify growth areas, and respond proactively to industry trends.

Case2: Automating Pharmaceutical Supply Chain Management

Background: A logistics company operating across Asia focuses on pharmaceutical distribution, custom packaging, and transshipment to Japan and Indonesia. Its operations relied heavily on manual processes for warehouse inspections and inventory tracking.

Challenges: Recording product information and medical kits required significant time and manual effort. Warehouse data was captured in spreadsheets, with limited analytical visibility. These inefficiencies not only delayed reporting but also introduced human error, hindering the company’s ability to monitor operations effectively.

Solution: Using Bitdeer AI Cloud’s GPU infrastructure, participants developed a mobile-friendly web application that combined photo capture with AI image recognition. Warehouse staff could take photos of cartons or medical kits, which were then uploaded for real-time analysis. AI models automatically identified SKUs, verified packaging, and updated inventory records.

Outcome: The system significantly reduced manual data-entry time and error rates while improving inventory accuracy. Real-time analytics dashboards gave supervisors complete visibility into logistics operations and stock movement. The prototype illustrated how cloud-based AI can drive operational efficiency and compliance in the highly regulated pharmaceutical logistics sector.

Case 3: Building an Integrated AI Ecosystem in FMCG

Background: With over 50 years of market presence, this leading FMCG company had built a strong brand heritage. However, its digital infrastructure had become increasingly fragmented over time. The deployment of disparate systems for customer service, marketing, and internal operations led to entrenched data silos and disrupted information flow.

Challenges: The organization grappled with two core operational gaps: a lack of centralized visibility for business reviews and the absence of a 24/7 customer service channel. This left decision-makers without a consolidated view of performance, while corporate chatbots remained offline after hours. Internally, significant productivity was lost as staff manually handled routine HR and IT inquiries.

Solution: The hackathon team created a comprehensive AI ecosystem using Bitdeer AI Cloud. It comprised three interconnected components:

1. A customer-facing chatbot powered by large language models to provide real-time, natural language interaction for users seeking information or support.
2. A business intelligence (BI) dashboard built on Microsoft BI, aggregating marketing, sales, and inventory data into interactive visualizations for leadership review.
3. A staff-facing assistant chatbot that automated HR and IT queries, improving efficiency and response time for internal teams.

Each module was deployed and tested using Bitdeer AI Cloud’s model library and virtual machines, enabling fast iteration and integration across the company’s systems.

Outcome: The prototype established a unified AI foundation for both customer engagement and internal operations. It improved responsiveness, enhanced data visibility, and reduced administrative workload demonstrating how enterprise-grade AI can drive organizational agility and connected intelligence.

Bitdeer AI Cloud: The Foundation for Scalable Innovation

All three projects were developed and deployed on Bitdeer AI Cloud, which provided participants with GPU computing, pre-trained AI models, and flexible development environments. The platform allowed teams to build, train, and deploy models without complex setup or infrastructure management.

By streamlining AI experimentation and deployment, Bitdeer AI Cloud enabled participants to focus on solving business problems and delivering measurable outcomes. The hackathon served as a real-world demonstration of how robust AI infrastructure accelerates both learning and innovation.

Accelerating the Future of AI Development

Hack to Hire demonstrated how advanced cloud infrastructure and AI technologies can support digital transformation across a wide range of industries. In just a few days, participants developed functioning prototypes that addressed key business challenges in sectors from manufacturing to retail, improving efficiency, operational intelligence, and innovation capabilities.

Bitdeer AI delivers high-performance, cost-effective, and production-ready AI solutions, helping organizations move projects from concept to implementation and achieve measurable business impact. The event also underscored a growing reality: scalable AI infrastructure is increasingly essential for driving intelligent growth, optimizing operations, and enabling innovation across industries. This successful round paves the way for the next as Hack to Hire returns soon with more opportunities for innovation and collaboration.

As artificial intelligence evolves into a full-scale industrial ecosystem, it is also consuming an unprecedented amount of electricity. Whether it is generative models, large language models (LLMs), or multimodal agents, their operation within data centers relies on a common physical foundation: power capacity.

Behind the rapid progress of AI, energy consumption has become a growing concern. Some have noted that “the endgame of AI is compute, and the endgame of compute is power.” As models expand and real-time inference becomes the norm, the industry faces a new and pressing question: Do we have enough electricity to power it all?

The Exponential Rise in AI Power Demand

According to the International Energy Agency (IEA), global data center electricity consumption is expected to double by 2030, reaching around 945 terawatt-hours (TWh) annually. Among them, AI-optimized high-performance data centers may consume up to four times more electricity than they do today equivalent to adding the annual power consumption of a medium-sized developed country to the global grid.

Traditional data centers typically consume between 10 and 25 megawatts (MW), while hyperscale AI-focused facilities can exceed 100 MW, with annual usage comparable to that of 100,000 households. These AI-centric facilities continue to expand rapidly to accommodate ever-larger models and rising demand for AI services.

Training GPT-3 reportedly consumed roughly 1.3 gigawatt-hours (GWh) of electricity. Generating a medium-length response of around 1,000 tokens with GPT-5 consumes an average of 18.35 watt-hours (Wh) equivalent to running a microwave oven for about 76 seconds and can reach up to 40 Wh, nearly 8.6 times higher than GPT-4’s 2.12 Wh. Every new generation of AI models, especially those with hundreds of billions of parameters, requires longer training cycles, denser GPU clusters, and sustained high power draw.

In short, electricity has become the new currency of computation.

How Power Capacity Shapes AI Development

“Speed-to-Power” has emerged as a key metric of competitiveness. In the global race for AI infrastructure, the speed at which a data center can secure and activate a stable power supply often determines how fast AI capacity can be deployed. Even with sufficient land, cooling, and connectivity, projects can stall if grid access is delayed.

The expansion of electricity generation and transmission infrastructure remains slow. From planning and permitting to construction and operation, the process can take years, while AI demand grows on a quarterly cycle. Utilities must balance the needs of data centers with stable residential and industrial supply. The United States, home to the world’s largest concentration of AI data centers, is already nearing grid stability thresholds in several regions.

AI training workloads can trigger sudden and intense power surges, placing stress on local grids. When massive GPU clusters ramp up simultaneously, they can cause voltage fluctuations, particularly in systems integrated with variable renewable energy. This highlights that significant improvements in AI-related power infrastructure are still required.

Bridging the AI Power Gap

The accelerating power consumption of AI has become a central challenge for enterprises. As electricity availability becomes a defining factor in AI development, many technology companies are enhancing their internal optimization and diversifying their energy strategies to strengthen resilience and reduce reliance on external grid expansion or policy incentives.

The first priority is energy efficiency and computational optimization. Through model compression, sparse computation, quantization, and dynamic scheduling, AI developers are striving to extract more compute per watt. Data center architectures are also evolving, with liquid cooling and high-density rack designs driving continuous improvements in PUE (Power Usage Effectiveness).

The second approach is energy diversification and storage development. Many AI infrastructure operators are deploying distributed generation and on-site storage systems such as campus-scale solar, wind, and battery energy storage systems (BESS) to balance peak loads and increase energy flexibility.

In addition, intelligent power management systems are becoming standard. AI-driven energy monitoring platforms can track real-time power fluctuations across GPU clusters, predict peaks, and automatically adjust supply strategies to achieve simultaneous optimization of computational scheduling and energy efficiency.

Finally, enterprise collaboration is emerging as a key enabler. Leading AI companies and energy providers are co-investing in next-generation power infrastructure, signing long-term power purchase agreements (PPAs), and accelerating the integration of green electricity and regional energy interconnections.

Overall, the industry is shifting from simply scaling compute to achieving smarter energy utilization pursuing higher efficiency and stability in a world of finite electricity.

From Resource to Strategic Asset

Electricity capacity is no longer just a technical metric. It has become a strategic lever connecting digital innovation, industrial strategy, and the global energy transition.

In the coming decade, competition in AI will not only revolve around models or chips but also around who can secure reliable, efficient, and sustainable power the fastest.

Conclusion

As global AI demand accelerates, power capacity has shifted from being a constraint to becoming a strategic advantage. The next breakthroughs in AI will depend not only on better semiconductors and smarter algorithms but also on who can deliver scalable, clean, and resilient power infrastructure.

At Bitdeer AI, we are expanding our dedicated AI power capacity to 200 MW of IT load by 2026 as part of our long-term strategy to ensure sustainable, high-performance infrastructure across our global data center network. We believe that in the era of intelligence, every watt of energy fuels innovation.

GPUs have become the backbone of modern Artificial Intelligence (AI), High-Performance Computing (HPC), and Generative AI. NVIDIA has played a pivotal role in this transformation, evolving from a graphics accelerator vendor into the enabler of factory-scale AI platforms. Each architectural generation has been driven by one principle: resolve systemic bottlenecks that limit compute, memory, or scalability.

This article provides a systematic review of NVIDIA GPU architecture, from CUDA programmability to NVLink and NVSwitch breakthroughs, and extends to Blackwell and the upcoming Vera Rubin platform, exploring how GPU evolution is shaping the present and future of intelligent computing.

What is a GPU

To understand the direction of GPU evolution, it is essential to first examine its structure. A GPU (Graphics Processing Unit) was originally designed for image rendering but has since evolved into the core engine driving AI and HPC. Unlike CPUs (Central Processing Units), which emphasize low-latency execution of single threads, GPUs are designed for massive parallelism and high throughput.

This architecture gives GPUs an unrivaled advantage in handling workloads that require thousands of tasks in parallel, making them indispensable for deep learning training, inference, and scientific simulations. For more details on key GPU parameters refer to our earlier article “Demystifying GPUs for AI Beginners: What You Need to Know”.

The Evolution of NVIDIA GPU Architectures

While GPUs excel at throughput through parallel design, they have also faced long-standing bottlenecks: programmability challenges, limited memory bandwidth, power efficiency constraints, and multi-GPU communication overhead. NVIDIA’s history of architectural innovation has been a process of systematically overcoming these constraints.

From Graphics to General-Purpose Computing (1999–2012)

Key Milestones:

• Tesla (2006): CUDA programmability opened GPUs to scientific and industrial computing
• Fermi & Kepler (2010–2012): Expanded memory hierarchy, improved efficiency, enabled supercomputers

Industry and Business Impact:

This was the stage when GPUs escaped the “graphics only” box. CUDA made it possible to run parallel workloads in physics, weather forecasting, and financial simulations. It was the foundation for enterprises to start viewing GPUs as compute engines rather than gaming chips. Enterprises gained access to affordable high-performance computing, which reduced barriers for R&D in pharmaceuticals, finance, and engineering

The AI Breakthrough Era (2014–2022)

Key Milestones:

• Pascal (2016): FP16 precision, NVLink 1.0, enabling large-scale deep learning
• Volta (2017): Tensor Cores, breakthrough for neural network training
• Ampere (2020): TF32 and INT8, scaling both training and inference

Industry and Business Impact:

This was when GPUs became synonymous with AI. Tensor Cores in particular changed the economics of training neural networks, cutting costs and time dramatically. NVLink enabled distributed AI across multiple GPUs, a prerequisite for modern LLMs. Deep learning moved from research labs into production. Voice assistants, computer vision in retail and manufacturing, and predictive analytics all became commercially viable. Companies that adopted early built durable competitive advantages in automation and personalization.

The Generative AI Era (2022–2024)

Key Milestones:

• Hopper (2022): FP8 precision, NVLink 4.0, and the introduction of confidential computing features designed for secure large-scale training.
• Blackwell (2024): NVLink 5.0, Grace CPU integration, data-center scale AI factories

Industry and Business Impact:

These platforms weren’t just accelerators but entire AI factories. Hopper made training trillion-parameter models possible, while Blackwell introduced factory-scale scalability, integrating CPUs and GPUs seamlessly. Enterprises could deploy generative AI copilots, real-time recommendation systems, and domain-specific AI platforms. Generative AI shifted from experimental pilots to core business strategy, transforming productivity, customer engagement, and competitive differentiation.

Beyond Hardware: Ecosystem and Scalability

The evolution of NVIDIA GPUs is not only about raw compute, memory, and interconnect, but also the supporting ecosystem that translates advances into usable performance.

• CUDA (Software Foundation): Translates GPU parallelism into programmability, offering libraries such as cuBLAS, cuDNN, and TensorRT.
• Tensor Cores: Redefined training efficiency for neural networks.
• Memory Hierarchy: Transitioned from GDDR → HBM → HBM3e, reaching TB/s-class bandwidth. Hopper expanded cache coherence, while Blackwell increased both capacity and speed.
• NVLink (Interconnect): Overcame CPU-GPU and GPU-GPU communication bottlenecks. Hopper’s NVLink 4 delivered ~900 GB/s aggregate bandwidth per GPU; Blackwell’s NVLink 5 doubled to ~1.8 TB/s and added unified memory addressing.
• NVSwitch (Scalability): Extended NVLink into full-switch fabrics, enabling multi-GPU systems to function as a single logical accelerator—critical for distributed AI training and cluster-scale AI factories.

Each step in this roadmap removed a bottleneck and expanded what enterprises could achieve with AI.

Business Implications

For AI, this represents a shift from single-device limits to industrial-scale systems. Training efficiency has accelerated with specialized compute, memory capacity has scaled with HBM innovations, and multi-GPU fabrics now enable trillion-parameter model deployment.

• Large-scale model training is no longer confined to hyperscalers. With access to modern GPU clusters, enterprises can now build and fine-tune large language models, creating proprietary AI assets that differentiate them in the market.
• Inference services have matured into enterprise-grade platforms, enabling copilots, assistants, and automated decision systems to scale reliably across entire workforces.
• Industries such as finance, manufacturing, and healthcare can now operationalize digital twins, scenario simulations, and predictive analytics that were once economically prohibitive.
• At the same time, leaders must balance the benefits with rising energy costs, infrastructure complexity, and sustainability challenges and decisions that increasingly shape competitive positioning and industry consolidation.

GPUs have evolved from accelerators into the backbone of enterprise-scale AI. The question is no longer how fast they run but what new possibilities they create. The focus has shifted from speed to scale and from experimentation to core infrastructure. Competitiveness now depends on aligning GPU strategy with business outcomes, turning compute capability into productivity gains, innovation capacity, and long-term market advantage.

Future Development: Vera Rubin (2026)

Following Blackwell, NVIDIA announced in September 2025 that its next-generation Vera Rubin architecture will launch in 2026. The Rubin CPX GPU and Vera Rubin NVL144 platform deliver 8 exaflops of compute, 100TB of fast memory, and 1.7 PB/s memory bandwidth, designed for million-token context processing and generative video. Vera Rubin points directly to ultra-long context AI applications, where models will no longer be limited to thousand-token dialogues but can instead process entire codebases, hours-long video, and multimodal histories, driving advances in AI agents, automated code generation, and creative media production.

While Rubin’s capabilities are groundbreaking, they also amplify existing challenges: ensuring stability and consistency across million-token contexts, balancing extreme compute and memory with sustainable deployment, and strengthening data security and privacy in multi-tenant environments. These issues will define the Rubin era and beyond.

Conclusion

GPU evolution is not just about increasing raw compute, but about the coordinated advancement of compute, memory, and interconnect. Each architectural upgrade has solved a new bottleneck, enabling larger and more complex AI models. From programmability to distributed scalability, GPUs have become the engine that moves AI from research to industrial-scale deployment.

The upcoming Vera Rubin platform highlights the next direction: not only faster, but also more specialized. Designed for long-context, multimodal, and system-level AI, Rubin represents the transformation of GPUs from accelerators into the core infrastructure of AI factories.

At Bitdeer AI, we build cloud infrastructure designed for this evolution, combining high-density GPU clusters, optimized cooling, and resilient network architecture. Our platform supports diverse AI workloads at scale, from model training to inference, enabling enterprises to deploy large, complex AI applications efficiently. With seamless scalability and integrated management tools, our cloud makes AI simple.