Analysis · Feb 2026
OpenAI and Anthropic are the two most valuable AI startups on Earth. Together they've raised over $60 billion. Neither has turned a profit. Here's why that's about to change.
The Billion-Dollar Bonfire
In the first half of 2025 alone, OpenAI posted a $13.5 billion net loss — burning roughly $575,000 every hour. Anthropic, while more restrained, still burned through $3 billion in 2025. The AI industry's two flagship companies are haemorrhaging cash at a pace that makes the dot-com era look frugal. But unlike Pets.com, these companies have revenue — and it's growing at an unprecedented clip.
OpenAI tripled revenue from $3.7B (2024) to $13.1B (2025), driven by ChatGPT's 900 million weekly users and aggressive enterprise expansion. Anthropic's trajectory is even steeper — from $1B to $9B in the same period, a 9x leap fuelled almost entirely by enterprise API adoption. By February 2026, Anthropic's annualized run-rate hit $14B, with Claude Code alone generating $2.5B — more than most public SaaS companies earn in total.
Where the Money Comes From
The two companies pursue fundamentally different revenue strategies. OpenAI is consumer-first — ChatGPT subscriptions drive the bulk of revenue, supplemented by API sales and a growing enterprise segment. Anthropic is enterprise-first — 80% of its revenue comes from business customers, with over 300,000 companies using Claude and 500+ spending more than $1M annually.
This distinction matters enormously. Enterprise contracts are stickier, more predictable, and higher-margin than consumer subscriptions. Anthropic generates $2.10 per compute dollar versus OpenAI's $1.60, according to Fortune. OpenAI's diversification into video generation (Sora), web browsing (Atlas), hardware (with Jony Ive), and even robotics demands huge capital. Anthropic has deliberately avoided these expensive adjacencies, focusing compute on its core Claude models.
Where the Money Goes
AI doesn't run on hope. It runs on GPUs — thousands, then millions of them — packed into data centres consuming as much electricity as mid-sized cities. Training a frontier model costs north of $100M. Inference costs scale with every query. And talent? Senior AI researchers command equity packages worth millions.
In H1 2025, OpenAI spent $6.7B on R&D, $2B on sales & marketing (nearly double the entire 2024 budget), and $2.5B on stock compensation. The compute margin — revenue after model-running costs — improved from 52% (Oct 2024) to 70% (Oct 2025), a strong signal that unit economics are heading in the right direction. But with OpenAI committing to ~$600B in total compute spend by 2030, the upfront infrastructure bet is enormous.
The Great Divergence
Here's where it gets interesting. Both companies burn cash today, but their paths to profitability are dramatically different.
Burn rate drops to ~33% in 2026, ~9% by 2027. Positive cash flow expected by 2027–2028. Projected $17B cash flow by 2028.
Burn stays at ~57% through 2027. $74B operating loss projected for 2028. Cash flow positive targeted for 2029–2030. Needs ~$207B more in capital.
OpenAI will burn through roughly 14x as much cash as Anthropic before turning a profit. That's the cost of building a consumer empire spanning chatbots, video, browsers, hardware, and robotics. Anthropic's laser focus on enterprise AI keeps costs tightly coupled to revenue growth.
When Do They Make Money?
The Bigger Picture
The global AI market crossed $390B in 2025 and is projected to exceed $3.4 trillion by 2033, growing at 30%+ CAGR. Enterprise AI adoption alone hit $115B in 2026. This isn't a niche — it's becoming the infrastructure layer for the entire global economy. The question isn't whether AI as an industry will be profitable. It's whether the current frontrunners will capture enough of that value to justify the investment.
The Wild Card: DeepSeek
In January 2025, Chinese lab DeepSeek released a model competitive with OpenAI's best — trained for just $5.9 million versus OpenAI's $100M+. Nvidia lost $600B in market cap in a single day. DeepSeek's models are now priced at 1/4th to 1/6th of comparable US systems. If training costs continue to plummet, the "you must spend tens of billions to compete" thesis collapses — and that's good news for profitability. Cheaper models = higher margins for everyone.
Conclusion: Yes, AI Will Be Profitable
The losses are real, but so is the trajectory. Here's why profitability is a question of when, not if:
OpenAI's compute margin went from 52% to 70% in one year. Better hardware (next-gen chips), better algorithms (distillation, quantisation), and competition from DeepSeek are all driving inference costs down. This is the single biggest lever for profitability.
78% of companies now use AI in at least one business function. 8 of the Fortune 10 use Claude. This isn't hype-driven trial usage — it's operational integration. Enterprise contracts are sticky and high-margin.
Anthropic grew revenue 9x in 2025 while reducing cash burn from $5.6B to $3B. OpenAI tripled revenue while burn grew less than 2x. The crossover point is mathematically inevitable at these growth rates.
The AI market is projected to hit $3.4 trillion by 2033. Even capturing 5–10% of that gives these companies revenue in the hundreds of billions — more than enough to cover infrastructure costs at scale.
Amazon lost money for 9 years. AWS alone now generates $100B+ annually. Netflix burned billions on content before becoming consistently profitable. The pattern of heavy upfront infrastructure investment followed by margin expansion is the defining playbook of transformative tech companies.
AI will be profitable. Anthropic will likely get there first (2027–28).
OpenAI's path is longer and riskier, but the scale of the payoff — $280B revenue by 2030 — justifies the bet.
The companies burning cash today are building the infrastructure layer of the next economy.
Sources