Chip giant Nvidia on Thursday, 9 July 2026 publicly congratulated xAI on the launch of Grok 4.5, confirming that the new model was trained on Nvidia GB300 NVL72 rack-scale systems and is purpose-built for coding, agentic tasks, and knowledge work.

Context

In its post on X, Nvidia stated: 'Congrats to xAI on Grok 4.5 — trained on NVIDIA GB300 NVL72 systems and purpose-built for coding, agentic tasks, and knowledge work. This is what happens when world-class AI infrastructure meets world-class AI research.' The announcement confirms that xAI, the AI company founded by Elon Musk in 2023, used Nvidia's latest Blackwell-generation hardware to train its newest frontier model. The GB300 NVL72 is a rack-scale system designed to handle the extreme compute demands of large-scale model training.

Grok is xAI's flagship large language model series, which began with the open-sourcing of Grok-1 in 2023. Each successive version has targeted more demanding use cases, and Grok 4.5 appears to deepen that trajectory with a specific focus on agentic and coding workflows — areas where enterprise demand has grown sharply.

Policy Backdrop

Nvidia's role as the dominant supplier of AI training hardware has become a structural feature of the global AI industry. The company's Blackwell architecture, announced in 2024 as the successor to Hopper, underpins the GB300 NVL72 systems that xAI used for this training run. Nvidia's dense GPU clusters have become the de facto standard for frontier model development as model sizes and compute requirements continue to scale rapidly.

The partnership also reflects a broader pattern in which US-based AI laboratories build large domestic training infrastructure using Nvidia hardware, even as export controls limit access to advanced chips in other markets. For enterprises and developers in India and across the Asia-Pacific region, the availability of models trained on such infrastructure has direct implications for the quality and capability of AI tools accessible to them.

Stakeholders and Impact

The most immediate beneficiaries are enterprise developers and AI researchers who use Grok models for coding assistance and agentic applications — tasks where a model must plan, reason, and execute multi-step workflows autonomously. Coding-focused AI tools have seen rapid adoption across software teams globally, including in India's large technology services sector.

For Nvidia, the public congratulations serve a dual purpose: they validate the GB300 NVL72 platform's readiness for frontier training workloads and reinforce the company's positioning as an essential partner to leading AI laboratories. The company's Jensen Huang-led pivot toward AI infrastructure has made such partnerships a central part of its market narrative.

What's Next

Attention will now turn to xAI's roadmap beyond Grok 4.5, including any further expansion of its training clusters and the scope of the model's public or enterprise availability. On the hardware side, Nvidia's next-generation Rubin architecture and its associated rack systems are expected to define the next wave of frontier training capacity. As model developers and chip makers continue to co-evolve their roadmaps, the compute-capability feedback loop is likely to accelerate iteration cycles across the industry.