Nvidia

Cobus Greyling argued in October 2025 that Nvidia had moved past hardware into a full software ecosystem: open source Nemotron models, including Nemotron-Nano-12B-v2-VL-FP8, the compact ARM64 DGX Spark workstation, and a data flywheel feeding continuous fine-tuning. His comment on the Spark states the mechanism plainly: "No easy AMD/Intel swaps, but that's the point right?" Small models lower the cost of experimenting, and the hardware underneath keeps that experimentation inside one environment.

Tariq Krim, writing for Cybernetica a month later, read the market as a rational realignment of capital rather than a bubble, and placed at its center an infrastructure duel between Nvidia GPUs as de facto standard and Google's TPU with its integrated ecosystem. An October 2025 Bloomberg report supplied the counterweight: OpenAI and Nvidia investing billions into each other inside a circular web of deals, valuations climbing ahead of durable monetization.

The European notes keep arriving at the same dependency. Arthur Mensch told the Assemblée nationale inquiry in May 2026 that Mistral AI is a very minority shareholder in Campus IA at Saint-Arnoult, a 35 Md€, 100 hectare, 1.4 to 1.6 GW site backed by MGX/Abu Dhabi and Nvidia. Three months later, Timothée Lacroix's plan for 1 GW of European compute by 2030 left two dependencies unresolved: Nvidia GPUs, and Microsoft as anchor tenant.

A May 2026 BFM Business debate added a smaller detail with the same shape: Nvidia reportedly pays bonuses in tokens. Sovereignty arguments here end where the GPU order begins.