Blackwell hasn't even cooled off, and Rubin is already here.

In July 2026, NVIDIA's next-generation GPU architecture Vera Rubin began shipping to major North American cloud providers, marking AI compute's official transition from the Blackwell era to the Rubin era. The first publicly announced large-scale Rubin deployment comes from an unexpected pairing: Microsoft signed a multi-billion-euro cooperation agreement with French LLM unicorn Mistral AI, planning to deploy thousands of Vera Rubin GPUs in France itself to build Europe's first gigawatt-scale AI compute cluster.

Why Europe? Why Mistral?

This deal's political significance rivals its technical significance.

Over the past two years, Europe's anxiety about AI infrastructure has intensified steadily. Frontier LLM training and inference compute are almost entirely concentrated in the U.S. — Microsoft, Google, Amazon, and Meta's data centers carry most of the world's AI workloads. Europe has homegrown stars like Mistral, but Mistral's model training relies heavily on American cloud vendors' GPU resources, a political soft target in the discourse around data sovereignty and industrial autonomy.

Microsoft and Mistral's partnership directly addresses this pain point: on European soil, using the latest generation of GPUs, building a gigawatt-scale compute cluster. This isn't a few-hundred-card experiment; this is genuine infrastructure capable of supporting frontier model training and large-scale inference. The French government has been actively pushing an “AI sovereignty” strategy, and this deal can be packaged as a landmark win for European AI autonomy.

For Microsoft, this is a shrewd strategic move. Partnering with Mistral gives Microsoft an anchor customer in the European AI infrastructure market while signaling to European regulators that “I'm not here to monopolize; I'm here to empower the local ecosystem.” For Mistral, Microsoft's financial and infrastructure backing means it can continue competing with OpenAI and Anthropic in the frontier model race without being entirely dependent on the whims of the big three U.S. cloud providers.

What Does the Rubin Era Mean?

Vera Rubin is NVIDIA's next GPU architecture after Blackwell. While full technical specs haven't been publicly disclosed, supply chain and early customer information indicate Rubin delivers significant improvements over Blackwell in compute density, energy efficiency, and interconnect bandwidth, with a particular focus on optimizing for inference workloads — aligning perfectly with the AI industry's center of gravity shifting from “training models” to “running models.”

Notably, “thousands” of Vera Rubin GPUs isn't massive in absolute terms — compared to Microsoft's hundreds of thousands of GPUs deployed in the U.S., a few thousand is just a starting point. What really moves supply chain expectations isn't this order's absolute size but its signaling value: Rubin has moved from “next-gen product” to “orderable product now.” Procurement decisions at cloud providers, LLM companies, and sovereign AI projects will begin to orbit Rubin rather than Blackwell.

This has indirect implications for China's AI industry. U.S. GPU export controls to China remain in place, and high-end Rubin versions will almost certainly be barred from the Chinese market. But China's domestic substitution is accelerating — Alibaba's T-Head open-sourced its SAIL software stack at WAIC, with Zhenwu chips shipping a cumulative 560,000 units. The global AI compute market is bifurcating into two systems: the North American system (NVIDIA + big three clouds) and the Chinese system (domestic chips + domestic software stack), while Europe attempts a third path between the two.

See you tomorrow.