Nvidia's Push Beyond Chips: Financing AI Infrastructure as Power Constraints Bite

Nvidia's growth now limited by electricity, not chips; introduces revenue-sharing for GPU access. Delays hit new rack systems as power demand soars.

For most of the AI boom, the story has been about chip scarcity. That narrative is quietly being overwritten. The new bottleneck throttling Nvidia’s growth is no longer fabrication capacity at TSMC or packaging yields — it is the raw electricity needed to run the racks once they leave the factory floor. And the company is responding with a business-model overhaul that turns it from a pure hardware vendor into something closer to an infrastructure financier.

The numbers underscore the shift. Nvidia’s Rubin GPU rack, still on track for delivery later this year, already draws around 300 kilowatts. Industry planning documents suggest future chips will push a single rack past one megawatt — enough electricity to power 750 average US homes. A manager at CoreWeave, a key Nvidia partner, framed the trajectory bluntly: "Je mehr Leistung man in einen Chip packt, desto höher wird die Dichte — immer weiter." That density creates its own accounting problem: Nvidia itself acknowledges that roughly 30% of the energy entering data centers never produces AI output, lost instead to cooling systems and transmission inefficiencies across sprawling campuses.

The company is adjusting its sales playbook accordingly. Rather than demanding full upfront payment for GPU clusters, Nvidia is rolling out a model where cloud providers gain access through revenue-sharing agreements and credit structures. The goal is to funnel compute capacity toward AI startups that cannot afford multi-million-dollar purchases. This is not an entirely new strategy — the primary source describes specific deals already inked. In Australia, Sharon AI has committed to roughly 40,000 Grace-Blackwell chips under a six-year contract. An even larger project in Indonesia will see Firmus Technologies build out a data center with up to 170,000 AI accelerators by 2028, with capacity agreements totalling nearly $30 billion. At a smaller scale, Bit Origin is investing $11 million in new servers in Malaysia, expected to generate recurring monthly revenue of $360,000.

Execution, however, is hitting speed bumps that go beyond power. Nvidia’s contract manufacturer Foxconn reported a nearly 40% revenue surge in the second quarter to almost $79 billion, driven by AI servers. But the supply chain is showing cracks. The Kyber NVL144 rack system is facing significant delays due to complications with internal circuit boards, pushing deliveries into 2028 — a setback of more than twelve months. The larger NVL576 systems are also expected to be affected. Despite these technical hurdles, management remains bullish: the current fiscal second quarter is forecast to bring in around $91 billion in revenue, with delivery obligations recently exceeding $95 billion.

To lead the commercial push, Nvidia has hired Nicholas Parker from Microsoft. Parker will start as head of sales on August 24, 2026, replacing Ajay K. Puri, who is retiring but staying on as an adviser. The compensation package reflects the importance of the role: a base salary of $1 million, a multi-million-dollar signing bonus, and stock awards valued at roughly $40 million.

The stock market’s reception has been mixed. Shares closed Friday at €171.98, up 1.09% on the day and 1.88% on the week, but down 7.16% over the past month. That leaves the equity 15.07% below its all-time high of €202.50 set in mid-May. Year-to-date, the gain is a modest 6.75% — a far cry from the triple-digit jumps of earlier cycles. The current level sits 5.17% below the 50-day moving average of €181.36, though still 4.73% above the 200-day average of €164.21. The RSI of 43.8 signals neither overbought nor oversold conditions, while the 30-day annualised volatility of 38.25% underscores the violent swings in sentiment.

The divergence between short-term sellers and long-term holders is playing out against a backdrop of structural reassessment. At a market capitalisation of €4.12 trillion, percentage growth becomes mathematically harder to sustain. Analyst expectations, however, remain elevated: the average price target of €263.59 implies more than 53% upside from current levels. That bet rests on the assumption that the electricity bottleneck — and the financing bottleneck Nvidia is now addressing — will be solved rather than derail the buildout.

Even the dividend — a quarterly payout of $0.25 per share, with an ex-date in June 2026 — feels like a quiet signal. A company that once epitomised high-growth, zero-dividend investing now offers a small income stream, reflecting its evolution into a more mature, infrastructure-oriented enterprise. The next major move in the stock may depend less on how many chips Nvidia ships and more on how quickly the world can supply enough gigawatts to keep them running.