Massive AI data centers are the primary driver of the sector’s $5 trillion spending, according to Raso’s analysis.
AI BUBBLE DEBATE: MATTEO RASO SAYS GROWTH IS JUSTIFIED
*Matteo Raso argues the AI sector’s sky‑high spending is a one‑off build‑out, not a looming crash. He backs the claim with spending data, Anthropic’s S‑1 filing, and debt‑service math.*
A heated debate rages over whether the AI sector is perched on a speculative bubble. Matteo Raso, author of a detailed online essay, challenges the prevailing narrative that trillions of dollars in AI valuations are destined to implode. He lays out hard numbers: $3 trillion in committed spending, $1.9 trillion already expended, and a total industry outlay near $5 trillion. Raso argues that most of this cash fuels data‑center construction, a capital expense that will not recur once the infrastructure is in place. He backs his case with Anthropic’s S‑1 filing, which shows a $2.08 return per $1 of compute and a revenue surge from $41 million in 2023 to $16.3 billion in early 2026. By juxtaposing debt‑service calculations with Alphabet’s historic revenue, he suggests the sector can meet its obligations without a catastrophic crash.
How much has the AI industry spent so far?
The AI sector has pledged at least $3 trillion in spending commitments. Reported outlays already total $1.9 trillion, pushing overall industry expense to roughly $5 trillion. These figures encompass data‑center construction, compute purchases, and related capital projects. The spending is described as unprecedented, and critics cite the opacity of financing vehicles as a barrier to precise accounting. Raso stresses that the bulk of this outlay is capital‑intensive infrastructure, not recurring operating costs.
Can AI firms turn a profit after the data‑center build‑out?
Anthropic’s S‑1 filing shows a $2.08 return for every $1 of compute, indicating profitability once infrastructure costs are excluded. The filing also records revenue climbing from $41 million in 2023 to $386 million in 2024, $4.585 billion in 2025, and $16.3 billion in the first half of 2026. Raso notes that other major AI players report similar trajectories. He adds that firms can instantly cut free‑tier usage or pause costly research to improve margins, while convertible notes allow debt repayment via equity rather than cash.
What does the AI debt burden look like?
If the sector must service $5 trillion of debt over 20 years, annual payments would be about $250 billion. For context, Alphabet generated $162 billion in 2019 revenue, rising to $213 billion when adjusted for inflation. Raso acknowledges the roughness of the math but argues the required cash flow is not impossible. Some liabilities are convertible notes, repayable with shares. Anthropic recorded a $72 billion loss in share‑based debt in the first six months of 2026, yet its $1 trillion valuation provides ample equity to satisfy lenders.
Is the AI bubble a realistic threat?
The “AI bubble” narrative assumes over‑valuation will trigger a market collapse, potentially wiping out trillions of dollars. Raso rejects that assumption, citing the rapid revenue expansion and the one‑off nature of most capital spending. He points to untapped demand: AI can augment programmers, mathematicians, lawyers, and doctors, and self‑driving cars already outperform human drivers in tests. While skeptics fear AI’s societal impact, Raso argues that financial flows will follow adoption, making a sudden, catastrophic bust unlikely.
Frequently asked questions
How much has the AI industry spent on data centers?
The AI industry has committed at least $3 trillion and has already spent $1.9 trillion, bringing total expense to around $5 trillion.
Is AI currently profitable after infrastructure costs?
Anthropic’s S‑1 filing shows a $2.08 return per $1 of compute and revenue growth from $41 million in 2023 to $16.3 billion in early 2026, indicating profitability once infrastructure costs are excluded.
What annual revenue is needed to service AI industry debt?
Paying off $5 trillion over 20 years requires roughly $250 billion per year, compared with Alphabet’s $162 billion (2019) or $213 billion adjusted revenue.
Raso’s analysis reframes the AI hype as a phase of massive infrastructure investment followed by a profitability window, not a speculative house of cards. If the sector can sustain the $250 billion annual debt service and convert its capital spend into revenue, the feared bubble may never burst. The real question now shifts from “Will AI collapse?” to “How quickly will the world absorb the new AI capacity?”
Sources: Hacker News (https://matteoraso.github.io/im-not-convinced-the-ai-bubble-exists.html#im-not-convinced-the-ai-bubble-exists)