Nvidia is no longer just selling to US hyperscalers. Sovereign AI revenue hit $30 billion last fiscal year — more than triple the prior year — as European governments treat computing capacity as a national asset. The company's VivaTech showcase in Paris this week underscored a structural shift that is reshaping its revenue base.
Nvidia shares rose 1.1 percent to $206.99 on Thursday as the chipmaker used the VivaTech conference in Paris to showcase a European AI infrastructure push that has already generated $30 billion in sovereign AI revenue — more than triple the prior year's total.
"AI infrastructure is coming online. AI agents are running in production, start-ups are deploying applications and the French AI ecosystem is developing models, data sets and platforms designed around local languages, cultural context and European requirements," Nvidia wrote in a blog post on Wednesday.
The company's sovereign AI business, which topped $30 billion last fiscal year, is now a second revenue stream less tied to the Silicon Valley venture cycle. Planned deployments across Europe will provide more than 3,000 exaflops of Blackwell compute, with technology centers being built or expanded in Germany, Sweden, Italy, Spain, the United Kingdom and Finland. Nvidia's GB300 system ran 1.6 times faster than its predecessor in the latest MLPerf benchmarks, underscoring a performance gap that extends to the same hardware sold to sovereign clients without downgraded alternatives.
The big hope is French AI start-up Mistral, which recently raised $830 million in debt financing to build 200 megawatts of AI computing capacity across Europe by 2027. Nvidia has invested in Mistral. The two companies are also part of a consortium planning a 1.4-gigawatt data center campus near Paris that would rival the largest American facilities. Mistral's spending is modest by US hyperscaler standards — Alphabet's Google recently said it plans to issue $80 billion in equity for AI investment — but it signals a broader shift as European leaders push back against American AI dominance.
Geopolitics Opens a Door
The Trump administration's decision to ban foreign use of two Anthropic models last week has intensified the debate over European access to leading AI. French President Emmanuel Macron said at the G-7 conference this week: "We won't buy any models made by these [American AI] companies if overnight, you can just flip the switch."
That sentiment is creating a structural demand catalyst for Nvidia in Europe. Foxconn, the world's largest electronics manufacturer, used VivaTech to debut humanoid robots on the continent for the first time and formalized a manufacturing alliance with French computing firm Bull and Nvidia. Under the deal, components for the Vera Rubin NVL72 rack-scale AI supercomputer — Nvidia's third-generation architecture delivering 3.6 exaFLOPS of NVFP4 inference throughput at one-tenth the cost per million tokens versus the prior Blackwell generation — will be produced at Foxconn's Czech Republic facilities and assembled at Bull's factory in Angers, France. The systems will commercialize under the Bull brand as Europe's first domestically manufactured instance of Nvidia's most advanced AI server architecture.
The Vera Rubin NVL72, in full production since the first quarter of 2026, unifies 72 Rubin GPUs and 36 Vera CPUs connected by sixth-generation NVLink fabric at 260 terabytes per second — double the bandwidth of the prior generation. That fabric matters specifically for mixture-of-experts model architectures, where constant high-bandwidth communication between expert sub-networks is required. For European AI factory operators, the engineering result is lower operational cost per token.
The China Puzzle Persists
While Europe opens up, Nvidia's China business remains frozen. The Trump administration approved the sale of H200 GPUs to roughly 10 Chinese companies including Alibaba, Tencent, ByteDance and JD.com, but Nvidia has generated not a single dollar from these licenses and plans no data-center revenue from China in the near term. It is demanding full prepayment from Chinese buyers as a hedge against the risk that authorizations could be revoked. Before export controls, Nvidia held about 95 percent of China's advanced AI chip market.
A recent $25 billion bond issuance was oversubscribed fourfold, reflecting institutional confidence in Nvidia's ability to finance its AI infrastructure buildout. The stock is up 9.8 percent year to date and 41 percent over the past 12 months. Consensus price targets hover around $257.80, implying significant upside from current levels.
For investors, the European sovereign AI thesis offers a diversifying revenue stream that is less exposed to US hyperscaler concentration risk. Nvidia trades at roughly 35 times forward earnings. The question is whether European AI factory buildouts — still in their early stages — can sustain the growth trajectory that US cloud giants have provided. Mistral's 200-megawatt target and the 1.4-gigawatt Paris campus suggest the pipeline is real, but execution timelines will determine whether the valuation premium holds.
This article is for informational purposes only and does not constitute investment advice.
