Nvidia’s Vera Rubin Launch Arrives as Sovereign AI Revenue Triples to $30 Billion

For a company that built its dominance on hyperscaler demand, Nvidia’s next chapter is looking decidedly more geopolitical. Sovereign artificial intelligence – the business of building national compute infrastructure for governments – generated $30 billion in sales during fiscal 2026, more than triple the prior year. That figure, representing roughly 14% of Nvidia’s $216 billion total revenue, has fundamentally reshaped the customer base just as the chipmaker prepares to roll out its next-generation Vera Rubin platform.

The shift comes at a delicate moment. Nvidia’s stock staged a 46.55% recovery from its June 23, 2025 trough to reach €182.60 ahead of this week’s annual shareholder meeting, but the recovery has stalled. Shares slipped 1.98% to €178.98 in recent trading, lagging the broader semiconductor sector’s year-to-date gain of roughly 11%. The 52-week high of €202.50 from May now sits about 12% above current levels, while the consensus analyst target of €260.83 implies more than 40% upside – a premium that hinges on whether the sovereign pivot and the upcoming product cycle can offset mounting headwinds.

Chief among those headwinds is the rapid erosion of Nvidia’s once-ironclad pricing power. Rental rates for B200 compute capacity have plunged from $6.11 to $4.22 per hour, signaling that the acute supply shortage that supercharged margins is fading. That decline has reinforced concerns among bears such as Michael Burry, who warns that AI valuations are stretched and that hyperscaler customers – Microsoft, Meta and Google – are increasingly designing their own chips for internal workloads. Nvidia still operates 81% of the world’s fastest supercomputers, but the structural moat is showing cracks.

The sovereign AI business is at least a partial answer to that vulnerability. Governments seeking digital independence have embraced Nvidia’s turnkey infrastructure: France deployed 18,000 Grace Blackwell systems, the UK secured significant compute capacity through 2030, and Deutsche Telekom runs the world’s first industrial AI cloud on Nvidia hardware. These state-level contracts now account for a meaningful slice of data-center revenue, which has become roughly evenly split between hyperscalers and other buyers including AI clouds, industrial firms, enterprises and sovereign clients. That diversification reduces, but does not eliminate, the concentration risk that has long worried investors.

The company is also counting on fresh hardware to reignite growth. At the ISC High Performance 2026 conference, Nvidia unveiled the Vera Rubin platform – a combination of Rubin GPUs and Vera CPUs delivering over seven exaflops of AI compute – alongside the Halos security architecture for robotics. The Leibniz-Rechenzentrum in Germany and Los Alamos National Laboratory have already committed to the platform, and systems from Dell and Supermicro are slated for availability in the fourth quarter of 2026. Halos targets a robotics security market that analysts estimate could reach $200 billion by 2035, giving Nvidia a foothold in physical AI beyond training.

Should investors sell immediately? Or is it worth buying Nvidia?

Yet the production ramp is not without friction. Blackwell chips, which drove Nvidia’s record data-center revenue, remain constrained by bottlenecks in HBM memory and advanced packaging. The Vera Rubin launch will test the company’s ability to scale manufacturing while managing a complex supply chain that is already absorbing a $4.5 billion write-down on H20 inventory and lost revenue of roughly $8 billion from China export restrictions. Since February 2026, US licenses have permitted small H200 shipments to select Chinese customers, but Chinese authorities have blocked those imports entirely – a reminder that the geopolitical calculus cuts both ways.

The immediate focus now shifts to Wednesday’s shareholder meeting, scheduled for 18:00 CET on June 24. The formal agenda, based on the proxy statement filed May 12, includes routine governance matters. But the market will be listening for signals on one metric in particular: the share of inference-based workloads in total revenue. Nvidia’s internal mantra – “inference equals revenue” – reflects the belief that real-world AI deployment, not just model training, will sustain demand. A rising inference share would validate the infrastructure build-out; stagnation would suggest that hyperscaler and sovereign investments remain speculative rather than self-funding.

Technically, the stock sits in neutral territory. The relative strength index reads 51.2, neither overbought nor oversold. The price is hovering just above the 50-day moving average of €180.86, a level it must reclaim to regain near-term momentum. The 100-day average at €168.35 and the 200-day average at €163.34 form the next lines of defense, though the latter still stands roughly 10% below current prices. A clean break above the 50-day line would reopen the path toward the May high of €202.50, while a failure risks a retest of the mid-June lows.

The next hard catalyst lands on August 26, 2026, when Nvidia reports quarterly earnings. The company has guided for around $91 billion in revenue, and the market will scrutinize whether Nvidia can translate 35 new European supercomputer projects into sustainable margin expansion. A successful Vera Rubin launch in the fourth quarter would provide a tangible growth driver, but it comes against the backdrop of falling chip rents and hyperscaler self-development. The sovereign AI surge offers a counterweight, but it has yet to be fully discounted by a market that still prizes hyperscaler growth above all else.

The shareholder meeting on June 24 will give management a platform to frame the narrative. The question is whether investors are ready to buy the story that governments – not just cloud giants – are now the engine of Nvidia’s next phase.