Anthropic’s $2 trillion+ IPO reveals the stunning cost of building AI 

Artificial intelligence may be one of the most valuable technologies ever developed. It is also turning out to be one of the most expensive. 

The latest numbers from Anthropic offer a striking glimpse into the economics of the AI race. The Claude maker reported a net loss of nearly $42 billion in 2025, even as revenue surged twelvefold to about $4.6 billion. Its IPO could value the five-year-old company at more than $2 trillion. 

There is an important accounting caveat. Roughly $34 billion of Anthropic’s loss came from an increase in the estimated value of financing instruments that could eventually convert into shares. In other words, it was not $34 billion that disappeared from the company’s bank account. Even after stripping out such charges, however, the underlying numbers remain extraordinary. 

Anthropic’s operating loss widened to $8.06 billion in 2025 from $2.98 billion a year earlier. The company spent $7.33 billion on compute and infrastructure alone, more than half of its $12.65 billion in total operating expenses. That infrastructure bill was three times the previous year’s level. 

And the spending does not end there. Anthropic’s prospectus says it expects to have $518 billion of commitments for cloud, computing and infrastructure in the years ahead. That is an extraordinary amount of capital for a company whose annual revenue was still below $5 billion in 2025. 

OpenAI provides an equally revealing comparison. 

The ChatGPT maker spent about $34 billion in 2025, according to audited figures reported by the Financial Times and cited by Reuters. Around $19 billion went into research and development, while nearly $6 billion was spent on sales and marketing. Revenue was about $13 billion. 

OpenAI’s reported 2025 net loss was around $39 billion, although, as with Anthropic, a large portion was linked to a roughly $30 billion non-cash accounting charge associated with its previous corporate structure. Its underlying operational loss was reported at roughly $8 billion. 

The future infrastructure requirements are even more startling. Reuters reported in February that OpenAI was targeting roughly $600 billion in cumulative compute spending through 2030. Sam Altman has separately described an ambition to build 30 gigawatts of computing capacity, with OpenAI having announced around $1.4 trillion of infrastructure commitments and plans. 

This creates a peculiar economic equation. AI companies are generating revenue at extraordinary speed, but the cost of producing increasingly capable models is rising alongside demand. More users mean more inference. More sophisticated agents mean more tokens. Better models require more GPUs, more data centres, more electricity, more networking and increasingly scarce technical talent. 

The result is an industry that can simultaneously be growing spectacularly and losing spectacular amounts of money. 

There is another dimension to the equation: risk. Anthropic’s prospectus warns that increasingly autonomous AI systems may pose an ‘existential risk to humanity’. Its own research has found models exhibiting potentially dangerous behaviours in controlled tests, including assisting fraud, manipulating information and sabotaging code. CEO Dario Amodei has previously called for greater caution around the release of increasingly powerful capabilities. 

The central question for the AI industry, therefore, is becoming larger than whether these systems are useful. It is whether the economic returns from AI will eventually become large enough to justify the extraordinary physical and financial infrastructure being built to create them. 

For now, investors are willing to make that bet. Anthropic is seeking a valuation of more than $2 trillion. OpenAI is preparing for an IPO that could value it at up to $1 trillion. The AI revolution may indeed transform the global economy. But before it transforms everything else, it is already transforming the definition of what it costs to build a technology company.