Musk warns 15GW of AI compute may idle in 2027. The bottleneck is infrastructure, not chips.
Investors eye GE Vernova and Constellation.
Nvidia's chip demand stays intact.
What happened
Elon Musk warned that roughly 15 gigawatts of AI compute scheduled for next year may never get powered up in 2027, equivalent to 10 nuclear plants' capacity. He named transformers, wiring, liquid-cooling systems, chillers, and networking gear as choke points.
Why it matters
Goldman Sachs and Morgan Stanley have flagged the same bottleneck, with U.S. data-center power demand expected to rise from 31 GW in 2025 to 66 GW in 2027. Historically, only 50% to 60% of scheduled capacity comes online on time, so the investment focus shifts to hardware that connects GPUs to power.
What to watch
GE Vernova's total backlog reached $176 billion, with Electrification segment data-center orders more than double all of 2025. Constellation and Vistra offer a third path with less execution risk.
Ask the AI about this article →
The AI trade has long centered on chipmakers like Nvidia, but Musk's warning highlights a different constraint: the physical infrastructure needed to power and cool data centers. This shift is supported by Goldman Sachs' projection of power demand nearly doubling by 2027, and Morgan Stanley's estimate of a 38 GW shortfall through 2028.
Investors are now looking at companies that supply the "plugs"—transformers, turbines, and grid gear—as well as those that can monetize the shortfall without heavy capital expenditure. GE Vernova's growing backlog and Constellation's existing nuclear and gas fleets represent distinct risk profiles. The key takeaway is that the 2027 winners may be decided not by chip performance but by the ability to plug them in.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
A new poll from Embold Research and Heatmap Pro shows 75% of Americans now oppose building AI data centers in…
Caterpillar is applying lessons from its mining automation business to broader AI deployment, including a voic…
Broadcom's AI revenue is near $8.4 billion per quarter with a $73 billion backlog, and management sees over $1…
Broadcom CEO Hock Tan reiterated on the June 3 earnings call that AI semiconductor revenue will exceed $100 bi…
A randomized experiment with 1,053 freshmen at Bocconi University found that GPT-4o helped students earn signi…
South Korea is a major AI boom winner, home to Samsung Electronics and SK Hynix, the two largest memory chip m…
