Nvidia’s Quiet Evolution: Inside the AI Factory Financing Strategy That’s Reshaping the Stock

Nvidia shifts from chip vendor to co-investor in AI infrastructure, financing cloud providers and eyeing Vera Rubin's 2026 launch despite Kyber rack delay.

Nvidia has quietly rewritten its playbook. Since July 1, the company has moved beyond its traditional role as a chip vendor, now extending credit lines to smaller cloud providers and striking revenue-sharing agreements. The idea is to turn itself into a co-investor in the global AI infrastructure buildout, rather than just a supplier. As many as 40,000 Grace-Blackwell units are already running in what Nvidia calls “AI campuses” built through this model, and the company is using its $4.3 trillion market cap to bankroll the very customers it hopes will buy its next generation of hardware.

Shares have responded positively, closing at 184.60 euros on Friday after a 7.97 percent weekly gain. The stock’s 14-day relative strength index of 58.6 leaves room for further upside before hitting overbought territory, and it sits comfortably above both its 50-day moving average of 181.22 euros and the 200-day average of 164.78 euros. Still, at 8.84 percent below the 52-week high of 202.50 euros set on May 14, the recovery remains incomplete — a reflection of the competing narratives swirling around the stock.

On one side is the operational momentum. In the first quarter of fiscal 2027, revenue surged 85 percent year over year to $81.6 billion, with the data center segment alone up 92 percent and networking inside that unit jumping 199 percent. Nvidia is also widening its moat: a new robotics team focused on embodied intelligence and simulation has been formed, and the company launched the BioNeMo Agent Toolkit for protein structure prediction and molecular docking. A large share buyback program — described as being in the high double-digit billions — underscores management’s confidence.

On the other side, a notable setback looms. The Kyber rack system designed for the upcoming Rubin-Ultra chips has been delayed until 2028 due to manufacturing issues, handing AMD and Google a potential opening. That news has tempered some of the enthusiasm, even as the company begins to talk up its next architecture. Vera Rubin, the successor to Blackwell, is scheduled to begin production in the second half of 2026, optimized for efficiency per watt rather than raw compute. It will feature closed-loop water cooling that nearly eliminates local water consumption — a direct response to the mounting energy and environmental pressures on hyperscale data centers.

The strategic shift toward financing smaller cloud providers dovetails with what Nvidia calls the “Year of Inference.” Analysts estimate that inference — the real-time generation of outputs from trained AI models — now accounts for 80 to 90 percent of total compute load in data centers. That trend plays directly into Vera Rubin’s design, but it also exposes Nvidia to new risks: the company now bears credit exposure to the health of its AI-factory partners. The average analyst price target of 264.16 euros implies 43.1 percent upside from Friday’s close, based largely on expectations that “sovereign AI” projects in India, Brazil, and the United States will drive the next wave of growth. Whether that bet holds will depend on how many of those national initiatives come online — and how many of Nvidia’s smaller customers survive the journey.