In an opinion shared on X today, Tether CEO Paolo Ardoino warned that Big Tech’s efforts to build AI data centers depend on subsidized computing and hardware that lose value within three to five years. He believes four structural mismatches are putting the sector at risk.
This warning comes as hyperscalers invest record amounts in infrastructure despite not seeing any clear returns on investment.
The four issues that must be fixed
Ardoino said that AI companies are subsidizing computing to attract more users and are investing heavily in infrastructure that only lasts three to five years.
He listed four main problems:
- Token prices do not match costs.
- Profitability timelines do not align with investments.
- The maturity of capital does not match asset life.
- Finally, open-source AI could lower revenues.
The spending numbers are huge and still growing. In its midyear outlook released on June 24, JPMorgan raised its estimate for global AI-related capital spending through 2030 to $5.5 trillion, up from $5.1 trillion, and expects AI-related debt financing to reach $4.1 trillion.
The bank predicts hyperscaler capital spending will reach $650 billion this year and go over $1.1 trillion in 2027. Microsoft alone plans to spend about $190 billion in 2026, which is a 61% increase from the previous year.
Goldman Sachs estimates that Meta, Microsoft, Amazon, and Alphabet will spend a combined $5.3 trillion on capital expenses between 2025 and 2030. This year, these four companies plan to spend $725 billion, which is 77% more than last year’s $410 billion.
Alphabet also raised $84.75 billion for AI infrastructure, which was described as the largest US equity capital raise ever, according to reports.
No returns on these massive investments yet
Ardoino’s worries about profitability reflect a wider uncertainty about whether this spending will actually pay off. The average company will spend $11.5 million on AI this year, but most cannot show any clear return on investment. Data from the Bureau of Economic Analysis also shows that growth in the Information sector slowed to 1.5% in the first quarter of 2026, down from 3.2% in the third quarter of 2025.
His warning about open-source AI taking a larger share of revenues matches a trend that has been brewing for months. Companies that previously encouraged staff to maximize AI usage (also known as ‘tokenmaxxing’) are now scaling back as their CFOs are now questioning rising API costs.
Amazon dropped its internal leaderboard tracking employee AI use, Uber used up its 2026 AI coding budget in just four months and set a $1,500 monthly cap per employee, and Meta cautioned about 6,000 staff over rapidly increasing costs.
IDC projects that by 2028, 70% of leading AI adopters will use multiple models instead of relying on a single vendor, which could spark a price war.
Regulators are also concerned. The Bank for International Settlements warned in its annual report that a sharp drop in AI investment could hurt global stock markets more than past recessions.
The bank named AI as one of three main risks to the economy. Zhang Tao, the BIS chief representative for Asia and the Pacific, said “the speed of a correction could be much faster than previous banking crisis episodes.”
Not everyone is so pessimistic. Wedbush analyst Dan Ives said that the buildout is “an arms race” that no major company can afford to leave. He believes the sector will start making money in the next six to twelve months.
JPMorgan also expects profits to remain strong, predicting operating cash flow will go above $900 billion by 2027.
Great Hill Capital chair Thomas Hayes gave a more balanced view, saying that one or more major companies may announce lower capital spending in the next earnings report. For now, the upcoming earnings season will be important. If any of the big spenders cuts back, as Hayes predicts, it will be the first real test of the issues Ardoino pointed out.
