GTC 2026 Day Two — Vera Rubin Partnerships, Robotaxi Deals, and the NemoClaw Ecosystem Expands
NVIDIA GTC 2026 Day Two brings Azure as the first hyperscaler on Vera Rubin NVL72, AWS committing to one million GPUs, BYD and Uber robotaxi partnerships, and 110 robots on the show floor signaling the physical AI era.
GTC 2026 Day Two — Vera Rubin Partnerships, Robotaxi Deals, and the NemoClaw Ecosystem Expands
Executive Summary
Day One of GTC 2026 was Jensen Huang's keynote — the vision. Day Two was where the ecosystem responded. Microsoft Azure became the first hyperscale cloud to announce Vera Rubin NVL72 systems going live. AWS committed to purchasing more than one million NVIDIA GPUs. BYD, Hyundai, and Nissan signed robotaxi partnerships running on NVIDIA DRIVE. And the NemoClaw open-source agentic AI framework that launched just 24 hours earlier already has partner integrations from ASUS, PEGATRON, Wiwynn, and Compal building Vera Rubin-based infrastructure to run it on.
This is not a hardware company anymore. This is a full-stack computing platform — and the ecosystem is aligning around it faster than anything since the early days of AWS.
AWS GPU Commitment
1,000,000+
NVIDIA GPUs committed by Amazon Web Services at GTC 2026
The Hyperscaler Race for Vera Rubin
Azure Goes First
Microsoft Azure was the first hyperscale cloud to confirm it will deploy Vera Rubin NVL72 systems. Each NVL72 rack contains 72 Rubin GPUs and 36 Vera CPUs, delivering what NVIDIA claims is 700 million tokens per second — a 350x improvement in inference throughput compared to the Hopper architecture from just two years ago.
Azure's early commitment makes strategic sense. Microsoft has been deepening its NVIDIA relationship since the Azure ND series launched with H100 GPUs in 2023. Vera Rubin represents the next generation of that partnership, and being first to market with the new silicon gives Azure a tangible differentiation against AWS and Google Cloud.
| provider | status |
|---|---|
| Azure | 95 |
| AWS | 80 |
| Google Cloud | 60 |
| Oracle Cloud | 45 |
AWS Responds with Scale
Not to be outdone, AWS announced a commitment to purchase more than one million NVIDIA GPUs. While AWS did not specify whether these are Vera Rubin units or a mix of Blackwell and Rubin, the sheer volume signals that NVIDIA hardware remains the default for cloud AI infrastructure.
This is particularly notable because AWS has invested heavily in its own custom silicon — Trainium and Inferentia chips — as alternatives to NVIDIA GPUs. The million-GPU commitment suggests that customer demand for NVIDIA hardware still far exceeds what custom silicon can address.
What This Means for Developers
If you are building AI applications on cloud infrastructure, the Vera Rubin rollout means:
- Inference costs will drop — 350x throughput improvement translates directly to lower per-token pricing
- Agent workloads become practical at scale — Multi-step agent workflows that currently cost dollars per run could cost cents
- Local vs. cloud calculus shifts — When cloud inference is this cheap, the argument for local GPU deployments weakens for all but the most sensitive workloads
As I covered in my GTC 2026 Day One analysis, the Vera Rubin platform is designed from the ground up for the token economy — where every computation is measured in tokens generated per second per dollar.
The Robotaxi Offensive
BYD, Hyundai, Nissan, and Uber
The physical AI announcements on Day Two were arguably more significant than the cloud partnerships. NVIDIA revealed robotaxi partnerships with BYD, Hyundai, and Nissan — three of the world's largest automakers — all running on the NVIDIA DRIVE platform.
Then came Uber. NVIDIA and Uber announced a deal to integrate robotaxis running NVIDIA DRIVE directly into the Uber network. This is the missing link that has held back autonomous vehicles for years. The technology exists. The regulatory framework is emerging. But without distribution, robotaxis remain a curiosity. Uber provides distribution to 150 million monthly active riders across 70 countries.
| Name | Value |
|---|---|
| BYD | 30 |
| Hyundai | 25 |
| Nissan | 20 |
| Uber Integration | 25 |
110 Robots on the Show Floor
GTC 2026 featured 110 robots at the show floor — the largest robotics presence in the event's history. This includes the BMW AEON humanoid robot deployment announced the day before the conference, which puts the first wheeled humanoid robot in an automotive factory at BMW's Leipzig plant.
Jensen Huang has been talking about "physical AI" for two years. Day Two of GTC 2026 is where the talk turned into hardware you can touch. Developers building robotic applications now have a complete NVIDIA stack — from simulation in Omniverse to inference on Jetson to deployment on DRIVE.
NemoClaw Ecosystem Day One
Partner Infrastructure
The NemoClaw launch on Day One immediately attracted hardware partners. ASUS, PEGATRON, Wiwynn, and Compal all showed Vera Rubin-based server configurations designed to run NemoClaw workloads at the partner showcase on Day Two.
This matters because NemoClaw is not just software — it is a reference architecture. When NVIDIA defines how autonomous agents should be sandboxed, secured, and deployed, and then hardware partners build servers optimized for those workloads, it creates a vertically integrated stack that is very difficult for competitors to replicate.
Nemotron 3 Super — The Agent Model
The Nemotron 3 Super model (120B parameters) that powers NemoClaw achieved a score of 60.47 percent on SWE-Bench Verified, a benchmark that tests AI coding ability against real software engineering tasks. This makes Nemotron 3 Super competitive with the leading coding models from Anthropic and OpenAI.
For enterprise teams evaluating NemoClaw, having a purpose-built coding model means their agents can perform code review, bug fixing, and code generation as native capabilities — without switching to a third-party model.
| model | score |
|---|---|
| Nemotron 3 Super 120B | 60.5 |
| Claude Opus 4 | 72.5 |
| GPT-4.5 | 65 |
| Gemini Ultra 2 | 58.3 |
Broader Industry Signals
UK AI Copyright Reports Due Tomorrow
While GTC dominates the news cycle, the UK government publishes two AI copyright reports on March 18 under the Data (Use and Access) Act 2025. These reports could set significant precedent for AI training data rights — an issue that affects every model maker at the conference.
Video Generation Leaps Forward
Two video generation breakthroughs landed this week. Lightricks shipped LTX 2.3 with native 4K video and synchronized audio generation. ByteDance, Peking University, and Canva released Helios, which generates full 60-second videos at real-time speed on a single GPU. Both are open-source.
These announcements reinforce the broader theme of GTC 2026: the AI stack is maturing so fast that capabilities once limited to research labs are shipping as open-source production tools within months of initial publication.
BMW Deploys First Automotive Humanoid Robot
BMW launched a pilot at its Leipzig plant deploying AEON, a wheeled humanoid robot by Hexagon Robotics. It is the first automotive factory deployment of its kind, and it arrives the same week that 110 robots populate the GTC show floor. The timing is not coincidental.
GTC Day 1
Jensen keynote — Vera Rubin, NemoClaw, Feynman reveal
BMW AEON
First automotive humanoid robot deployment at Leipzig plant
GTC Day 2
Azure, AWS, robotaxi partnerships, partner infrastructure
UK Reports
AI copyright reports published under Data Act 2025
What Comes Next
GTC 2026 has two more days. The partner and developer sessions running Tuesday through Thursday will reveal the practical details behind the keynote announcements — specific APIs, pricing, availability dates, and developer programs.
For developers building on the NVIDIA stack, the immediate action items are:
- Get on the NemoClaw waitlist — The alpha is open but capacity-limited
- Start with the NeMo Agent Toolkit — This is the most mature piece of the stack and is available now on GitHub. If you want a step-by-step walkthrough, I published a complete tutorial on building agents with NemoClaw and the NeMo Agent Toolkit today
- Watch the Vera Rubin pricing — Cloud pricing on Vera Rubin NVL72 systems will determine whether multi-agent workflows are economically viable at scale
- Track the NemoClaw prediction — I have a prediction that NemoClaw will reach 60 percent adoption among Fortune 500 agent deployments by Q1 2027. The Day Two ecosystem response is already providing early evidence
The token economy that Jensen Huang described in his keynote is not a vision for 2028 or 2030. The partnerships announced on Day Two show it arriving in 2026 — this year.