AI losses are about to spiral out of control: OpenAI loses $1.22 for every dollar earned
OpenAI reported $5.7 billion in revenue in Q1 2026 and still posted non-GAAP losses of $6.9 billion: it loses $1.22 for every dollar billed. The author warns the company could go bankrupt this year or in early 2027, and that Anthropic does not escape the problem either.
By Will Lockett (planetearthandbeyond.co / Substack) · June 26, 2026.
The article —partly behind a paywall— opens with a provocative thesis backed by concrete figures: the AI industry is, according to the author, one of the largest financial bubbles in history, and the mechanisms that would normally curb its expansion —antitrust regulation, financial oversight, market pressure— appear to have no effect on it.
**OpenAI's Q1 2026 figures**
The central example Lockett uses is OpenAI. The company reportedly posted revenue of $5.7 billion in the first quarter of 2026, an impressive figure in absolute terms. However, against that revenue it recorded a non-GAAP loss of $6.9 billion. The arithmetic is devastating: for every dollar of revenue, OpenAI loses $1.22.
The author further stresses that non-GAAP figures are inherently more favorable than GAAP (Generally Accepted Accounting Principles) ones, since they can exclude items such as restructuring costs and certain types of depreciation. OpenAI, he notes, has plenty of both, suggesting that the real loss under standard accounting criteria would be notably larger.
**The context: $38.5 billion in losses the previous year**
This is not an isolated phenomenon of the quarter. According to data cited in the article itself, OpenAI accumulated $38.5 billion in losses over the course of 2025. The pattern is not that of a company in a controlled investment phase seeking to reach critical mass, but of a structural hemorrhage that worsens as revenue grows.
Lockett argues that, at the current rate of cash destruction, OpenAI could declare bankruptcy during 2026 or in early 2027, even counting on the private capital injections it periodically receives.
**Anthropic is not spared either**
The article also calls into question Anthropic's narrative, which has tried to project an image of a company close to profitability. Citing analyst Ed Zitron, Lockett maintains that this apparent profitability rests on an unsustainable market position and on creative accounting —specifically, EBITDA profitability— that does not reflect the company's financial reality.
In general, EBITDA (earnings before interest, taxes, depreciation and amortization) is a useful metric for assessing operating cash generation, but it can be misleading in sectors with extremely high investment needs in computing infrastructure, as is the case with large language models. Excluding data center depreciation or debt interest can make a company with negative free cash flow appear 'profitable' on paper.
**The bubble thesis**
Lockett does not limit himself to pointing out the losses: he argues that the ecosystem as a whole is in a phase of active denial. Unlike other historical bubbles, where market forces eventually impose correction, the AI bubble would be artificially sustained by a combination of massive private investment, the absence of effective regulation, the lack of antitrust enforcement and what the author calls 'round-trip investments' —circular financial structures in which money appears to circulate between investors and companies without creating real value.
The article cuts off at its most substantive part due to the paywall, so it is not possible to reproduce the author's final proposals or conclusions about what could trigger the collapse or what strategies companies might attempt to avoid it.
**Limitations of this summary**
It should be noted that the available material corresponds only to the article's introduction, as the rest is reserved for paying subscribers. The cited figures (Q1 2026 revenue, non-GAAP loss, 2025 losses) come from the text itself and could not be independently verified within the scope of this summary. The article's tone is clearly one of opinion and criticism, not neutral financial analysis.
As sector context, the profitability difficulties of the large AI labs are widely documented: the cost of training and inference for frontier models, the need for massive computing infrastructure and the high salaries of specialized talent create cost structures that are difficult to cover with current API and consumer subscription prices.