Sam Altman Just Delivered a Warning to Investors Betting on Endless AI Growth

AI & Compute

Altman Flags AI Chip Demand Concentration as Capacity Expands

Sam Altman says chipmakers may be betting on the same OpenAI and Anthropic orders. NVIDIA, AMD, Micron and Coherent show why capacity now outruns visibility.

By
Rebecca Stone
Filed
Channel
AI & Compute
Read
4 min read

By Rebecca Stone

NVIDIA said in August that OpenAI had committed to about 12 gigawatts of its AI infrastructure through 2030; by November, Sam Altman was warning that many suppliers were counting on the same buyers.

At Semafor’s November 16 “Silicon Valley & the World” event, OpenAI CEO Sam Altman identified a timing problem beneath the AI hardware boom. Designers expect OpenAI, Anthropic and other frontier labs to absorb their output.

Altman said: “There are now many people making chips, all talking to the silicon supply chain, assuming that we’re going to buy all of their chips.”

He did not question the sector’s long-run appetite for compute. Underlying data-center demand remains solid, he said, and long-term computing demand should keep growing; his concern is whether supply and demand align in shorter intervals.

That distinction matters. Multiple vendors can plan capacity against the same orders, then deliver into a market where each new data center still needs power, cooling and connectivity.

Altman spoke as AI stocks fell. NVIDIA’s shares were down 28.88% over the preceding month, a market move that did not erase the broader wager on rising compute demand.

What does OpenAI’s 12-GW commitment mean for NVIDIA?

Then-CEO Jensen Huang said in August that OpenAI’s commitment to substantial NVIDIA AI infrastructure deployments through 2030 amounted to about 12 gigawatts.

Those deployments represent a stated customer commitment. By contrast, NVIDIA’s $108 billion plus-or-minus-2% Q3 revenue figure is company guidance, following $96.22 billion in reported Q2 revenue.

NVIDIA also expects AI labs backed by its balance sheet to make up roughly one-quarter of its business next year, putting customer financing and concentration alongside chip performance on the earnings agenda.

Are suppliers exposing the same demand base?

  • AMD: Its first Helios racks are running workloads for Anthropic, Meta, Microsoft, OpenAI and Oracle Cloud Infrastructure. Lisa Su said demand exceeds supply, while AMD expects revenue growth above 80% in 2027.

  • Micron: More than 75% of its shipments are committed for fiscal 2027, yet management sees no clear point when supply and demand will balance. Its fiscal Q4 revenue rose 379.3%.

  • Coherent: Its calendar 2027 is effectively booked, with customers now booking calendar 2028, CEO Jim Anderson said. An approximately $20 billion PhotonLink agreement with a major AI company extends its exposure.

These disclosures make Altman’s warning measurable: the same frontier labs can support several supply chains at once, while vendors quote booking horizons extending into 2028.

That does not prove an immediate glut. It does show why revenue timing can diverge from capacity announcements when five large customers appear across multiple roadmaps.

Can neocloud financing absorb the buildout?

CoreWeave plans fiscal 2026 capital expenditure of $37 billion to $39 billion. In Q3, it booked $846 million of net interest expense and a $709 million net loss; its stock was down 33.12% year to date.

CEO Michael Intrator said: “We are not a growth-at-all-costs business.” The statement puts financing discipline against the AI buildout, even as contracted capacity grows.

At IREN, co-CEO Daniel Roberts said investor questions had shifted from contract size to whether capacity could be delivered, financed and operated profitably.

IREN’s net debt climbed to $4.43 billion from $216 million, while its shares fell 42.39% in one month.

Together, those numbers move the bottleneck beyond wafer starts. Announced chips still need financed data centers, and those sites still need power, cooling and connections before customers can turn them into billable compute.

Altman’s warning therefore concerns utilization as much as silicon. Suppliers can build against overlapping forecasts, but the buyer has to absorb both the hardware and infrastructure bills.

What should investors watch in NVIDIA’s next report?

NVIDIA’s next earnings report will be its first since Huang’s ouster. The commercial test is simple: whether the company identifies the same leading research labs that AMD, Micron and Coherent name when describing demand.

Anthropic’s October 26 S-1 adds a buyer-side version of the risk, saying the company may be committed to excess compute that remains underutilized. That filing marks a possible burden, not a reported current glut.

The next earnings cycle will show whether overlapping 2027 and 2028 bookings become matched demand. If OpenAI, Anthropic and peers anchor several suppliers at once, competition for a limited buyer pool will intensify; financing and infrastructure constraints will decide how much announced capacity becomes revenue.

Source: 247wallst.com

Share this article:

More from Rebecca Stone

Rebecca Stone

Show full bio

Correspondent covering technology and product news in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.

1,092 articles

Related articles

« Previous articleNext article »