Nvidia (NASDAQ: NVDA) has caused controversy by claiming it has already shipped 6 million GPUs. blackwell It was calculated by counting each of the two cores separately, but Ed Zitron, a technology expert and noted artificial intelligence (AI) bear, believes that even the lower 3 million figure could actually be a false alarm.
Specifically, the journalist explained that the number of data centers that could be confirmed to be operational indicates that 75% of the units actually sold (approximately 2 million units) may be “collecting garbage” simply waiting to be deployed, unable to generate revenue.
Between the overall capital expenditures (CapEx) going toward acquiring GPUs and other hardware, and the estimated pace of construction, this information could burst the “AI bubble,” Zitron argued in the latest edition of his newsletter.
If data centers aren’t built, the whole AI bubble will burst.
Microsoft added 2GW in 1 year, but claims production may be less than 1GW
Elsewhere, Ed Zitron claims to have conducted extensive research across public sources to support his claims, including direct research into companies such as Microsoft (NASDAQ:MSFT), but found little confirmation that the large number of announced projects actually came online in a revenue-generating manner.
Notably, CEO Satya Nadella said in late 2025 that his company had “added more than 2 gigawatts of new capacity” in one year, adding in late April that “our Fairwater, Wisconsin data center will be operational ahead of schedule.”
Fairwater data center in Wisconsin begins operations ahead of schedule.
As the world’s most powerful AI data center, we consolidate hundreds of thousands of GB200s into a single seamless cluster.
Thank you to all the teams who made this possible! https://t.co/O586ioWkJK
— Satya Nadella (@satyanadella) April 16, 2026
While Zitron claims he was able to track up to 1GW of total capacity operated by Microsoft, he added that after reaching out to the company for clarification on the CEO’s claims, he was promised no response.
Finnvold’s own efforts to understand the actual status of the Stargate project in Narvik, Norway, after OpenAI abandoned the site in April and handed it over to Microsoft, confirms the difficulty of finding accurate information about the state of the data center.
Specifically, after reviewing a variety of public sources, including press releases and media reports, the most definitive statement available in English was in the form of a Google Maps review stating that it was “under construction.” This means that the reliability of a single clear source of information is uncertain.
Most major data center projects appear to be ‘stuck in development hell’
Meanwhile, Zitron’s efforts to find accurate information about the status of many of its announced projects led to reports that the facilities actually built were much smaller than 1 GW and took an average of about 18 months from start to finish.
The situation is further muddied by the fact that many data centers are “in development hell” and a trend in reports about places coming online that explicitly reference only some phases while hinting that entire campuses are up and running.
So technology reporters seem to be continuing their previous arguments about how much of the reported AI demand is coming solely from OpenAI and Anthropic, thanks in large part to their vast and often generous contracts with the “Magnificent 7” tech giants, by concluding that even the purported supply constraints are simply the result of so little capacity actually coming online.
Does xAI and Antropic’s partnership signal an explosion in demand or construction?
To support this argument, Zitron cited the agreement between Anthropics and xAI regarding the use of aging phenomena. Colossus-1 The facility’s capacity is estimated at 300MW, which seems insignificant compared to the gigawatts it reportedly comes into service each year.
The agreement is even more bizarre given Elon Musk’s public hostility towards the world’s most prominent AI company. Famously, the South African-Canadian-American billionaire said: Claude – Anthropic’s flagship platform – was described as “misanthropic and evil” in a Feb. 12 post by X.
Your AI hates white people and Asian people, especially Chinese people, straight people, and men.
This is misanthropic and evil. Please fix it.
Frankly, I don’t think there’s anything we can do to escape the inevitable irony of humanity becoming misanthropic. You were destined for this fate because you…
— Elon Musk (@elonmusk) February 12, 2026
Ed Zitron said the deal appears doomed and indicates no significant capacity expansion is expected in the coming quarters, despite numerous announcements, including one that claims construction is “ahead of schedule.”
Why are major tech companies announcing they will set up data centers in space when construction on Earth is at a standstill?
It’s possible that the technology experts and reporters didn’t find an existing data center, rather than proving the existence of one, but the lack of clear communication on behalf of a company that has significantly increased capital expenditures to fund construction is definitely strange.
Indeed, given the relative novelty of such AI-focused facilities, it’s no mean feat for blue-chip tech giants to be proud of the milestones they’ve achieved, especially in scaling capacity to multiple gigawatts.
Therefore, “why do they reach such a milestone without making a final announcement” (which effectively keeps their breakthrough a secret) also represents a burning question.
The recent trend of making plans for orbital facilities public, as exemplified by the recent agreement reached between Google (NASDAQ: GOOGL) and SpaceX, makes the lack of confirmation even more acute, especially since the coldness of space may favor cooling, but the lack of an atmosphere does not.
Should investors start running out of patience regarding the “AI boom”?
Finally, while there is little reason to doubt Zitron’s due diligence, there is a risk that technology experts are falling victim to confirmation bias.
In fact, he has speculated that a number of chips that NVIDIA claims to have sold have been “gathering dust” in warehouses for months, while the company has quickly become one of the most prominent and fiercest AI underdogs on the internet.
The potential for bias is further heightened by his other recent writings. In it, he dismissed comparisons between the current state of AI spending and Amazon’s (NASDAQ:AMZN) long road to profitability, and highlighted the big difference between verifiable reported AI revenue and capital spending.
Nevertheless, the size of the investment, the exposure of the public to the ongoing AI boom, the lack of clear communication, frequent obfuscation of data, profitability, AGI, and even the seemingly ever-changing timelines for data center construction raise many uncomfortable questions.
Featured image via Shutterstock
