Profoundd checked a Nov. 3 post by the X account @euwyn, which had 5.1 million views when Profoundd read it on Nov. 4. The post carries a 68-second clip of Elon Musk and says he warned that “the AI boom will hit a WALL by the end of 2026.” Profoundd read Musk’s words in the full transcript of the episode the clip comes from, then checked each claim against electricity data from Ember, the International Energy Agency (IEA) and the largest U.S. grid operator, and against what the companies buying AI chips have said.
The claims, checked
- “Elon Musk says the AI boom will hit a WALL by the end of 2026” Musk’s own words, said in February. The clip is from Musk’s interview on Dwarkesh Patel’s podcast, published Feb. 5, 2026, nine months before the post, which does not give that date. In the interview Musk said: “People are going to hit the wall big time on power generation. They already are.” He also said: “Towards the end of this year, I think people are going to have real trouble turning on... The chip output will exceed the ability to turn chips on.” The post’s “millions of them pile up” adds a number Musk did not give; his words were “The chips are going to be piling up and won’t be able to be turned on.”
- “Electricity is nearly flat everywhere outside China” The contrast with China holds; whether new power is “flat” depends on what is measured. Musk said output outside China is “more or less flat. It’s maybe a slight increase, but pretty close flat,” and set it against chip output “growing pretty much exponentially.” By Ember’s figures, electricity use outside China rose about 1.7% (+346 terawatt-hours) in 2025, against 5.0% (+503 terawatt-hours) in China. In 2024 it rose about 3.1% (+634 terawatt-hours), against 6.7% (+631 terawatt-hours) in China. New supply is growing fast but not as fast as data centers ask for it: the U.S. added 53 gigawatts of generating capacity in 2025 (a gigawatt is roughly one large power plant’s output), the most in a year since 2002, while the largest U.S. grid operator said in December 2025 that data-center demand “continues to far outstrip new supply.”
- “By the end of the year, we will be making more AI chips than we can actually turn on” Reported at Microsoft in late 2025 and at Amazon in October 2026; not shown industry-wide. Microsoft’s chief executive said in late 2025 that he had chips he could not plug in. In October 2026, Business Insider reported that Amazon had begun taking ownership of billions of dollars of Nvidia server racks before they reached a data center, and people familiar with the effort told it that some sites lacked the power to switch them on. Amazon disputes part of that report. No company publishes a count of chips sitting unpowered, so there is no industry-wide number to check.
- “The bottleneck for the next year is not chips, it is raw power” An accurate paraphrase of Musk; the evidence on it is split. Musk’s words: “The current limiting factor that I see in the three to four year timeframe, it’s chips. In the one year timeframe, it’s energy, power production, electricity.” Nadella’s account, Business Insider’s Amazon reporting and the grid operator’s figures support him. SemiAnalysis, a research firm that tracks data-center construction, wrote in October that power “is largely solved” because AI companies are building their own power plants, and that chips may become “the new #1 constraint in 2027-28.” A Futurum Group survey of enterprise data-center decision-makers, released in April 2026, put chip supply narrowly first: 26% named it their biggest constraint and 23% named power and cooling (Futurum did not say how many people it surveyed).
What Musk said, in context
Patel asked Musk why anyone would put data centers in space, where the chips are harder or impossible to service and, Patel suggested, more expensive to run. Musk answered: “The availability of energy is the issue. If you look at electrical output outside of China, everywhere outside of China, it’s more or less flat. It’s maybe a slight increase, but pretty close flat. China has a rapid increase in electrical output. But if you’re putting data centers anywhere except China, where are you going to get your electricity?” He went on: “The output of chips is growing pretty much exponentially, but the output of electricity is flat. So how are you going to turn the chips on?”
Musk’s complaint was about getting new power to a site quickly. He described waiting on utilities (“They have to do a study for a year”) and on equipment (“The turbines are sold out through 2030”), and said “It’s not clear to me that there’s enough usable electricity to turn on all the AI chips that are being made.” He drew a line between big clusters and small devices: “people start getting to the point where they can’t turn the chips on for large clusters towards the end of this year,” while “for edge compute it’s a different story,” meaning chips spread out in cars and robots that can charge from the grid at night. His answer for the big clusters was space: “once you can get to space and unlock the power constraint, you can now do hundreds of gigawatts per year of power in space.”
What the electricity data show
Ember, an energy think tank, publishes an annual review of the world’s electricity. Its 2026 review, released in April, gives totals for the world and for China. Profoundd subtracted China from the world total to get the rest of the world:
| 2024 change | 2025 change | |
|---|---|---|
| World | +4.3% (+1,265 TWh) | +2.8% (+849 TWh) |
| China | +6.7% (+631 TWh) | +5.0% (+503 TWh) |
| Everywhere outside China | about +3.1% (+634 TWh) | about +1.7% (+346 TWh) |
Source: Ember, Global Electricity Review 2026. The “outside China” row is Profoundd’s arithmetic from Ember’s world and China figures. Ember’s 2025 totals include estimates for countries that had not yet reported; it had 2025 data for countries covering 93% of world demand. A terawatt-hour (TWh) is a billion kilowatt-hours, about what 93,000 U.S. homes use in a year at the Energy Information Administration’s 2022 average of 10,791 kilowatt-hours per home.
The long record supports Musk’s word “flat.” The IEA’s Electricity 2026 report says demand in advanced economies “is on an upward trajectory again after a 15-year period of stagnation.” That upturn came in the years just before Musk spoke. Ember puts U.S. growth in 2025 at 3.0% (+131 terawatt-hours), and the IEA’s July 2026 mid-year update says U.S. demand set records two years running, with data centers “the main driver.” Growth has slowed since Musk spoke: about 1% in the United States in the first half of 2026, after a mild winter, the IEA says, while the European Union grew more than 2%. The IEA expects China to grow 5.5% this year. The IEA’s own tables, which measure demand slightly differently from Ember’s, put growth outside China at about 2.0% (+360 terawatt-hours) in 2025 and forecast about 2.6% (+485 terawatt-hours) for 2026; Europe grew 1.1% in 2025 and the Eurasia region, which includes Russia, shrank 0.2%.
Total use is not the same as power a new data center can get. PJM Interconnection, which runs the grid for more than 67 million people in 13 states and Washington, D.C., reported on Dec. 17, 2025 that its auction for power supply from June 2027 through May 2028 came up 6,623 megawatts (about six and a half gigawatts) short of its reliability requirement and cleared only 774 megawatts of new generation. The previous auction, for the year that began in June 2026, met the requirement by 139 megawatts. “This auction leaves no doubt that data centers’ demand for electricity continues to far outstrip new supply,” said Stu Bresler, a PJM executive. PJM still expected a reserve margin of 14.8%, against the 20% its standard calls for, and said mitigating factors could bring the system “very close” to that standard. The IEA’s Electricity 2026 report says “One-fifth of the global data centre build out is at risk of delay due to grid congestion.”
New supply is not standing still. The U.S. Energy Information Administration reported in February that 53 gigawatts of generating capacity were added to the grid in 2025, “the largest capacity installation in a single year since 2002,” and that developers planned a record 86 gigawatts in 2026, about half of it solar. SemiAnalysis wrote on Oct. 19 that AI labs and cloud companies, facing grid connections that take five to 10 years, began building their own power plants on site, and that “the power generation problem has broadly been solved.” That is the route Musk’s own company took with gas turbines near Memphis.
Chips waiting on power
Musk said in February that the wall was already being hit, and the clearest example came before he spoke. Microsoft chief executive Satya Nadella said on the BG2 podcast in late 2025, as TechCrunch reported: “you may actually have a bunch of chips sitting in inventory that I can’t plug in. In fact, that is my problem today. It’s not a supply issue of chips; it’s the fact that I don’t have warm shells to plug into.” A “warm shell” is a building that already has power and cooling, ready for equipment. Nadella put the problem as “the ability to get the [data center] builds done fast enough close to power.”
On Oct. 20, 2026, Business Insider reported, from internal emails, that Amazon Web Services expected to take ownership of roughly $5 billion of Nvidia GB300 server racks (cabinets of AI chips) before final deployment, and projected about $3 billion more each month. “Some AWS sites had room for more racks but not enough available power to turn them on,” the report said, citing people familiar with the effort, so finished racks were kept at warehouses. An Amazon spokesperson called staging hardware near sites “best business practice” and disputed that the broader company lacked a path for getting racks into data centers. One email said at least one AWS team did not have “a clear forecasted path” for the incoming racks.
Some analysts expect more of the same. Bruce Bateman, a chief analyst at the research firm Omdia, wrote in a 2026 opinion piece that “Industry analysis projects 30-50% of planned 2026 data center capacity will slip to 2028,” citing grid-connection times and transformer shortages.
How much power data centers will need is itself disputed. Developers have been “submitting large-load requests to multiple utilities” for the same project, “a practice that has inflated regional demand forecasts,” Data Center Knowledge reported in June. That bears on how large the future shortfall will be, not on the chips reported waiting at Microsoft and Amazon.
Chips are scarce too
Chips themselves are still scarce, by Nvidia’s account. In November 2025, Nvidia chief executive Jensen Huang said “cloud GPUs are sold out,” meaning the graphics processors that run most AI work. For the quarter ended July 26, 2026, Nvidia reported data-center revenue of $89.0 billion, up 117% from a year earlier. Strong sales do not settle Musk’s forecast either way, because Nvidia is paid when a chip is sold, not when it is switched on. Business Insider described Amazon’s trade-off as “securing scarce GPU racks before competitors do, even if some data centers are not yet ready to turn them on.” Both limits can bind at once: Musk himself said chips become the limit in three to four years, and SemiAnalysis expects that shift sooner.
The Futurum survey asked enterprise decision-makers, not only the operators of the large clusters Musk described, and Futurum called chip supply “the single largest bottleneck.” Futurum’s Brendan Burke said of the largest cloud companies: “Power generation is a binding ceiling on hyperscaler expansion.”
Who stands to gain
Musk’s companies stand to gain from the fix he described. Three days before the episode was published, SpaceX acquired Musk’s AI company xAI, and Musk “said a main reason for the merger was to better build ‘orbital data centers,’” CNBC reported. The episode was titled “In 36 months, the cheapest place to put AI will be space.” In the interview Musk also said: “We’ll be able to turn on more chips than other people can turn on, faster, because we’re good at hardware.”
His companies are also building power on the ground. SpaceX spent $295 million on Tesla Megapack batteries in the second quarter of 2026 for its data centers near Memphis, CNBC reported. On the quarterly call, Musk said: “Our tentative target is to actually have 20 gigawatts of power and cooling live online by the end of next year.” The grid could not deliver power on the company’s timeline, so it brought in portable gas turbines; the NAACP has sued in federal court in Mississippi to stop their use without pollution controls or federal permits. Mississippi’s environmental regulator had allowed portable turbines to run without air permits for up to a year each, and the company later received a permit for 41 permanent turbines at the site, Scientific American reported. The lawsuit argues that dozens of portable machines running together are one power plant that needs a Clean Air Act permit. The Justice Department has asked the court to dismiss the suit, arguing that shutting down Colossus 2’s power source would threaten AI innovation and national security, Scientific American reported in September.
The poster has a stake as well. The @euwyn account’s X profile describes its owner as the founder of Orbital, “Data centers in space.”
What Profoundd could not check
As of Nov. 4, no company reports how many of its chips are sitting unpowered, so the size of the pile Musk predicted cannot be measured from public records. Profoundd found statements from Microsoft and reporting on Amazon. Profoundd’s research run and a reviewer’s web searches on Nov. 4 found no comparable statement from Meta, Google, Oracle or CoreWeave that chips they had bought were sitting unpowered, and no 2026 statement from Microsoft repeating or withdrawing Nadella’s account.
Profoundd did not ask SpaceX how many of its own chips are waiting on power, or whether Musk still holds the end-of-2026 forecast. It did not check GPU rental prices, which could point either to a glut of chips or to a shortage.
Profoundd's read
The clip is real and accurately summarized, apart from the added “millions,” but it is nine months old. Musk’s narrow forecast, that big AI clusters would have chips waiting on power, has been reported at Microsoft in late 2025 and at Amazon this fall, and the largest U.S. grid operator says data-center demand “continues to far outstrip new supply.” Whether power or chips will be the main limit over the next year is disputed among people who track it closely: SemiAnalysis says companies building their own power plants have largely solved the power problem and that chips may become the tighter constraint, while Futurum’s survey puts chips narrowly first. Musk’s broadest line fits 2025 better than 2024: electricity use outside China rose about 1.7% in 2025, the “slight increase” he allowed for, after about 3.1% the year before. Musk’s companies stand to gain from the answer he proposed, more power in orbit and on the ground.
