A claim surfaces: AI companies surpass McDonald’s revenue. The number is $120 billion. It circulates in crypto media, retweeted as proof of AI dominance. But the code doesn't match the narrative. I trace the invariant where the logic fractures.
Context
The source is Crypto Briefing, a cryptocurrency publication. Its article states Anthropic and OpenAI combined have $120 billion in revenue, exceeding Starbucks and McDonald’s. No audited filings back it. No breakdown of revenue sources. The data is presented as fact, yet every industry benchmark contradicts it.
OpenAI’s actual annualized revenue as of late 2024 was approximately $3.7 billion. Anthropic’s was around $1 billion. Combined: under $5 billion. The $120 billion figure is two orders of magnitude higher. The only match is their valuations — OpenAI at ~$157 billion, Anthropic at ~$60 billion. The article conflates valuation with revenue. This is not a typo. It’s a structural failure of information integrity.
Core — Code-Level Dissection
Let’s verify from first principles. If OpenAI and Anthropic generated $120 billion in revenue, what would that imply about their infrastructure?
Assume 50% of revenue goes to inference compute costs (conservative for LLM deployment). That’s $60 billion spent on GPUs annually. At current H100 rental prices (~$2.50/hour), that buys 24 billion GPU-hours per year. The entire global supply of H100 GPUs is roughly 5 million units. That’s 43.8 billion GPU-hours total available. So two companies would consume 55% of all H100 compute. Nonsensical.
Now check the user economics. ChatGPT Plus has 10 million subscribers at $20/month — $2.4 billion annual. Enterprise deals add maybe $1 billion. API revenue at $0.01 per 1K tokens — to reach $100 billion in API revenue, you’d need 10^16 tokens processed. That’s 10 million times the entire GPT-4 training data. The physics doesn’t hold.
Metadata is memory, but code is truth. The revenue claim fails on every technical load test. The real story is not about AI beating McDonald’s — it’s about media producing data that cannot survive a simple cost model.
Contrarian — The Hidden Signal
The contrarian angle: the $120B figure is not a lone error — it’s a symptom of a broader pattern. Crypto media increasingly adopts AI hype to pump token attention. The article is not reporting; it’s advertising. The real economic shift is not AI’s revenue catching up to fast food — it’s the upstream capture by infrastructure providers. NVIDIA’s data center revenue alone exceeded $100 billion in 2024. That’s the real $120B story: the picks and shovels, not the miners.
Friction reveals the hidden dependencies. The dependence here is on investor capital. AI companies burn cash faster than they earn it. OpenAI lost $5 billion in 2024. Anthropic similar. The “surpass” narrative masks fragility. When the capital taps dry, the revenue will revert. The abstraction leaks, and we measure the loss.
Takeaway
The next crisis will emerge when these inflated projections fail to materialize. Capital will recede. The $120B mirage will evaporate, leaving only the infrastructure players standing. Precision is the only reliable currency. Question every number that fits the headline too cleanly.
