Power Grid, Not Chip Delays, Emerges as Nvidia’s Newest Bottleneck

Nvidia’s stock has been bouncing between two very different narratives this week — one rooted in physical infrastructure limits, the other in market jitters that briefly sent shares reeling. After dipping to 169.52 EUR during Tuesday’s session, the shares recovered to close at 172.50 EUR, leaving them roughly 15 percent below the May record and about 6 percent lower over the past 30 days. The year-to-date gain of roughly seven percent remains intact, but the path has turned choppy.

The sharp intraday swing owed less to any fundamental shift in demand than to a liquidity crisis in South Korea. The KOSPI index tumbled nearly five percent, triggering a so-called margin cascade in which investors were forced to sell highly liquid positions — Nvidia among them — to meet margin calls. Even strong earnings from Samsung could not lift the broader chip sector in that environment. Adding to the noise, weekend analyst reports claimed Nvidia’s next-generation Kyber NVL144 rack system would be delayed until 2028. The company quickly denied the rumors, stating that base Rubin systems are already in production and will ship to eight major cloud customers this autumn. The stock stabilised after the clarification, but the episode underscored the market’s hyper-nervous reaction to any whiff of a slowdown in the AI buildout.

Yet the real constraint on Nvidia’s growth is no longer chip availability or fabrication capacity — it is the power grid. Industry forecasts suggest that by 2027, 40 percent of AI data centres could face electricity shortages, with wait times for grid connections stretching to 36 months in key markets. Nvidia is tackling this head-on with the upcoming Rubin architecture, which will leverage the new HBM4 memory standard to cut energy consumption significantly. Recent field trials with utilities have also demonstrated that GPU clusters can dynamically throttle power usage during grid strain, freeing capacity within existing infrastructure.

The demand side of the equation is undergoing a structural transformation. While hyperscaler growth has normalised to roughly 12 percent quarter-on-quarter, the segment of national AI factories and regional clouds is expanding at 31 percent. Countries such as France, India, Germany and Brazil are treating AI infrastructure as a strategic national asset, no longer willing to entrust their data licenses to foreign cloud providers. These sovereign projects now account for about 14 percent of Nvidia’s revenue. The company is actively deepening its role as architect of this ecosystem: partnerships with Sharon AI, Firmus Technologies and others use revenue-linked financing models that allow smaller cloud builders to deploy vast compute capacity without massive upfront investment. Firmus, for instance, is constructing a giant AI campus in Indonesia.

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

At the same time, Nvidia faces emerging competitive pressure from China, where the AI firm DeepSeek is developing its own chips as part of a broader push for Asian technological independence. But analysts point out that Nvidia’s integrated hardware-software stack creates strong lock-in, a factor that has helped the stock hold above its 200-day moving average of 164.41 EUR, even as it trades about five percent below the 50-day line.

Valuation has become more palatable for bargain hunters. The price-to-earnings ratio has slipped to around 20, which Goldman Sachs describes as highly attractive given Nvidia’s roughly 80 percent market share. The average analyst price target stands at 264 EUR, implying upside of more than 55 percent from current levels. Short-term momentum, as measured by the RSI, sits near 45 — squarely neutral.

Beyond terrestrial data centres, Nvidia is also pushing into orbital infrastructure. Over 220,000 of its chips are reportedly running in SpaceX installations, and the Vera Rubin Space-1 module points to a future where the company supplies compute capacity for space-based AI. The quarterly dividend of $0.25 per share remains modest, but the real story for investors continues to be the shift from a supplier of chips to the operator of an operating system for entire economies.