‘Everything about the AI bubble breaks,’ tech expert claims 75% of chips Nvidia sold remain unused

While Nvidia (NASDAQ: NVDA) has already courted controversy by claiming it had shipped 6 million GPUs after Blackwell came out by counting each of the two cores separately, the technology expert and prominent artificial intelligence (AI) bear, Ed Zitron, believes even the lower figure – 3 million actual units – might be a misrepresentation.

Specifically, the journalist explained that the number of data centers he could confirm as operational indicates that it is probable that as much as 75% – roughly two million – of the units actually sold are simply awaiting deployment and ‘gathering dust,’ unable to generate revenue.

Between the overall capital expenditure (CapEx) dedicated to acquiring GPUs and the other hardware, and the alleged pace of construction, this information can burst the ‘AI bubble,’ Zitron argued in the latest edition of his newsletter.

If Data Centers Aren’t Getting Built, Everything About The AI Bubble Breaks

Microsoft claims it added 2GW in a year, but might be operating less than 1GW

Elsewhere, to back his argument, Ed Zitron claims to have conducted extensive research across public sources, but also direct inquiries directed at companies such as Microsoft (NASDAQ: MSFT), and found scarce confirmation that the numerous announced projects have actually been brought online in a revenue-generating capacity.

Notably, CEO Satya Nadella stated in late 2025 that his company ‘added over 2 gigawatts of new capacity’ in one year and added in late April that the ‘Fairwater datacenter in Wisconsin is going live, ahead of schedule.’

Zitron claims to have been able to track down at most 1GW of capacity in total operated by Microsoft, while adding that, after reaching out to the company for clarification on the most senior executive’s claims, he was promised a reply that never arrived.

Finbold’s own efforts to determine the actual state of the Stargate project in Narvik, Norway, after OpenAI gave up on the site and handed it over to Microsoft in April, corroborate the difficulties in finding precise information on the state of data centers.

Specifically, after examining a variety of public sources, including press releases and media reports, the most decisive remark available in English came in the form of a Google Maps review that stated it was ‘under construction’ – meaning the singular clear source was of indeterminate credibility.

Most major data center projects appear ‘stuck in development Hell’

Meanwhile, Zitron’s attempts to find precise information on the state of the many announced projects led him to report that the facilities actually constructed are far smaller than even 1GW, and that they took an average of about 18 months to move from groundbreaking to completion.

The situation is further muddled by the many data centers ‘stuck in development Hell,’ and the tendency of reports about bringing locations online to imply the entire campus is operational while explicitly referencing only some of the phases.

Thus, the technology reporter appears to have continued his previous argument about the majority of the reported AI demand coming exclusively from OpenAI and Anthropic – and even that, thanks to sprawling and often generous contracts with ‘Magnificent 7’ big tech giants – by concluding that even the alleged supply constraints are merely the results of very little capacity actually coming online.

Does the xAI-Antropic partnership signal exploding demand or imploding construction?

To back the argument, Zitron referenced Anthropics’ agreement with xAI regarding the use of the aging Colossus-1 facility, whose capacity, estimated at 300MW, appears trivial relative to the gigawatts reportedly coming online every year.

The agreement is made even stranger given Elon Musk’s public enmity toward the world’s other most prominent AI company. Famously, the South African-Canadian-American billionaire described Claude – Anthropic’s flagship platform – as ‘misanthropic and evil’ in a February 12 X post.

According to Ed Zitron, the deal appears both desperate and indicates that no significant capacity expansion is expected during the coming quarters, despite the numerous announcements, including those that claim construction is ‘ahead of schedule.’

Why is big tech announcing data centers in space as construction on Earth stalls?

While there is a possibility that the technology expert and reporter failed to find existing data centers rather than prove their non-existence, the lack of clear communication on behalf of companies vastly increasing their CapEx to fund construction is, arguably, quaint.

Indeed, given the relative novelty of such AI-focused facilities, it would not be beyond the pale to expect blue-chip big tech giants to be proud of the milestones they accomplished, especially as they increase capacity into multiple gigawatts.

Thus, ‘why would they reach such milestones while providing no definitive announcements’ – effectively keeping their breakthroughs a secret – also represents a compelling question.

The recent trend of unveiling plans for orbital facilities – as exemplified by an understanding recently reached between Google (NASDAQ: GOOGL) and SpaceX – makes the lack of confirmation even more pointed, especially since, while the coldness of space might be a benefit for cooling, the lack of atmosphere is not.

Should investors start running out of patience regarding the ‘AI boom?’

Lastly, though there is little reason to doubt Zitron’s due diligence, there is a risk that the technology expert has fallen victim to confirmation bias.

Indeed, he has been speculating that a large number of chips Nvidia claims to have sold are ‘gathering dust’ in warehouses for months while rapidly becoming one of the most prominent – and vehement – AI bears on the internet.

The possibility of bias is further increased by his other recent work, in which he dismissed the comparisons between the state of artificial intelligence expenditure and Amazon’s (NASDAQ: AMZN) long road to profitability, while highlighting the vast difference between verifiable reported AI revenues and CapEx.

Nonetheless, the scale of investment, the exposure regular people have to the ongoing AI boom, the lack of clear communication, frequent data obfuscation, and the seemingly perpetually shifting timetables for profitability, AGI, and even data center construction do raise ample and uncomfortable questions.

Featured image via Shutterstock