Microsoft (MSFT) CEO Satya Nadella didn’t mince words in his appearance on the Bg2 Pod with investor Brad Gerstner and OpenAI’s Sam Altman.
The tech leader made a startling admission that the biggest problem facing AI expansion right now isn’t chips — it’s power.
In what was a rare moment of candor, Nadella confessed that Microsoft has some of the most cutting-edge GPUs sitting idle because there’s basically nowhere to plug them in.
“I don’t have warm shells to plug into,” he said, referring to unfinished data center facilities that don’t have power or cooling capacity. It’s a stunning admission from arguably the world’s most resource-rich companies, and a potential reality check for the entire AI boom.
With industries across the globe racing to build smarter machines, it’s hitting a very human limit; there might not be enough electricity to keep the dream running.
Microsoft’s AI ambition meets a power wall
Microsoft’s all-in bet on AI may have run into an obstacle no algorithm can fix.
The tech giant’s relentless push to expand its AI footprint is being weighed down not by innovation, but by infrastructure. It’s not often you hear a CEO admit his biggest problem is finding enough plugs!
Too many GPUs, not enough juice
When quizzed on Brad Gerstner’s Bg2 Pod on whether he concurred with Nvidia CEO Jensen Huang that there’s “no chance of a compute glut,” Nadella gave a reality check.
“The biggest issue we are now having is not a compute glut, but it’s power,” he said, adding that the tech behemoth has “a bunch of chips sitting in inventory that I can’t plug in.”
That wasn’t metaphorical, but literal, in that without “warm shells,” or powered data centers ready for operation, Microsoft’s GPUs stay boxed.
AI’s power crunch, in four painful facts:
The strain Nadella described isn’t purely a Microsoft problem, but a full-blown industry slowdown. Here’s what it looks like across the cloud and compute landscape:
- Clouds tapping the brakes: Amazon Web Services (AWS) paused multiple new colocation data-center leases across several regions this spring. Microsoft followed suit earlier this year, when it said it was pausing select U.S. AI builds, including a $1 billion Ohio site, while repurposing two others.
- Google’s demand-response pivot: In keeping grids stable with AI workloads surging, Google inked deals with U.S. utilities to delay or reschedule non-urgent computing during peak demand.
- Regional grid bottlenecks are real: Ireland is estimated to be getting nearly 21% of its electricity from data centers. Similarly, in the U.S., PJM, America’s largest grid, reports data-center demand has surged so quickly that it’s basically rewriting planning processes.
- Utilities and firm power rush in: Constellation, Vistra, NextEra, and Brookfield Renewable are in a race to lock in 24/7 PPAs, while nuclear upstarts such as Oklo and NuScale continue drawing hyperscaler attention to plug the AI power gap.
- Chip supply vs. deploy gap: Nvidia’s GPUs are still in record demand, but deployment has been lagging. Several customers have postponed Blackwell racks due to heat and manufacturing issues, even before considering the power shortage.
Also, if you’re wondering how big this will get, the IEA expects global data-center electricity usage to increase 50% to nearly 945 TWh by 2030,roughly 3% of world demand. Goldman Sachs forecasts U.S. data-center load skyrocketing 50% by 2027.
What else did the Nadella-Altman conversation reveal?
Apart from the AI energy debate, the Bg2 Pod chat between Microsoft’s Satya Nadella and OpenAI’s Sam Altman offered a few under-the-radar insights investors shouldn’t miss.
First, Microsoft’s link-up with OpenAI isn’t just about bottom-line growth, but also effectively shaping the rules of the AI era.
Related: Top analyst drops jaw-dropping price target on Nvidia stock
The company owns roughly 27% of OpenAI, down from about a third, through a nonprofit-over-public-benefit-corp model, which holds close to $130 billion in stock. The first $25 billion is deployed for health, AI safety, and resilience projects.
At the same time, Microsoft Azure keeps exclusive rights to OpenAI’s stateless APIs through 2030, backed by a power-packed revenue-share deal to 2032.
More Tech Stocks:
- As Palantir rolls on, rivals are worth a second look
- Nvidia’s next big thing could be flying cars
- Cathie Wood sells $21.4 million of surging AI stocks
Moreover, OpenAI’s $1.4 trillion compute commitments may not sound like much compared to its reported 2025 sales, but Altman says growth is “steep.”
With the cost per unit of intelligence dropping, demand tends to explode, a cycle that may swing from shortages to gluts. Both Altman and Nadella have warned that state-by-state AI laws, such as Colorado’s, could potentially stall innovation.
Forget IPO hype, OpenAI plans to fund expansion from sales, along with new breakthroughs such as Codex and AI-driven discoveries by 2026.
Related: Warren Buffett’s Berkshire takes $6 billion out of the stock market
