Jensen Huang sees a bottleneck coming. The Nvidia chief executive spent the weekend in Seoul hammering out a pact with SK Group chairman Chey Tae-won, officially sealed on Monday, to secure supplies of High Bandwidth Memory for the company’s next-generation Vera Rubin supercomputing platform. The warning from Huang is blunt: without a steady flow of HBM chips, the entire artificial intelligence sector could stall for years.
The urgency of the trip reflects a broader tension inside the company. On one hand, Nvidia just reported a record quarter. Revenue in the first quarter of fiscal 2027 — the period ending in April 2026 — hit $81.6 billion, an 85% jump from a year earlier. Its data-center division alone contributed $75.2 billion, up 92%. Those are numbers the company has never posted before.
Yet the stock has struggled to hold the gains. Shares closed the week at €178.08, down 5.42% on the week and roughly 12% below the all-time high of €202.50 hit in mid-May. The technical picture shows the price clinging to its 50-day moving average near €174.40, with the relative strength index at 45 and annualised volatility around 44%. Since the start of the year, the stock still shows a solid gain of just over 10%.
What’s unsettling investors is the growing risk that Nvidia’s biggest customers are building their own chips. Roughly half of data-center revenue comes from hyperscalers — Google, Amazon, Microsoft and Meta. Each is pouring resources into custom silicon: Ironwood TPU, Trainium, Maia 200 and MTIA accelerators, all designed to reduce dependence on Nvidia hardware. Semiconductor analyst Jay Goldberg of Seaport Research puts it directly: “That has the potential to completely destabilise Nvidia — I consider it a significant risk.”
The numbers back up the concern. TrendForce projects that application-specific integrated circuits (ASICs) will grow 44.6% in 2026, compared with just 16.1% for standard GPUs. For the first time, custom AI chips are expanding faster than Nvidia’s core product.
Huang counters that the overall demand curve is so steep that it renders such comparisons academic. During the analyst call, he described the demand environment as “parabolic” and pointed to a broadening customer base of AI startups, enterprises and governments that have no intention of building their own chips. Bernstein analyst Stacy Rasgon agrees: “I’m not really interested right now in who wins or loses. That’s the wrong question.” His argument is that AI agents are driving compute needs so forcefully that the only real constraint is production capacity — and whoever can deliver, sells.
Should investors sell immediately? Or is it worth buying Nvidia?
For the current quarter, Nvidia guided revenue of around $91 billion, a figure that excludes China data-centre sales.
The Seoul pact is designed to ensure that delivery capacity stays intact. SK hynix will supply the DRAM memory for the Vera processors, while Samsung and Micron are also being tapped to provide HBM4 chips for the Rubin architecture. That diversification is intended to cushion the supply constraints that Huang sees looming in the high-end memory segment.
Inside the company, there are signals of confidence. On May 18, the board authorised an additional $80 billion share buyback programme with no expiration date. The quarterly dividend was lifted from $0.01 to $0.25 per share, payable on June 26 to holders of record on June 4. During the first quarter alone, Nvidia returned roughly $20 billion to shareholders. UBS strategist Maxwell Grinacoff called the dividend increase a potential precedent, saying it could “open the door for other large S&P 500 companies with solid free-cash-flow yields to introduce or raise dividends.”
Yet insider activity has raised eyebrows. Over the past 90 days, executives and directors sold shares worth about $387 million. The largest single trade was a sale of 500,000 shares by director Mark A. Stevens. Institutional investors, however, have largely held their positions.
Bank of America remains bullish, reiterating its “Buy” rating with a $350 price target. The firm expects the content value per gigawatt of AI infrastructure to double from $40 billion with the current Blackwell generation to $80 billion once Vera Rubin ships.
The fundamental question for Nvidia is whether the hyperscalers’ own chip investments will eventually eat into the company’s order book. Huang’s management team expects total AI spending by these giants to reach $1 trillion by 2027. The answer to that question — and the success of the Seoul supply chain pivot — will determine the stock’s trajectory for the foreseeable future.
Ad
Nvidia Stock: Buy or Sell?! New Nvidia Analysis from June 8 delivers the answer:
The latest Nvidia figures speak for themselves: Urgent action needed for Nvidia investors. Is it worth buying or should you sell? Find out what to do now in the current free analysis from June 8.
Nvidia: Buy or sell? Read more here...
