Hook
Over the past seven days, a single number has been circulating through the crypto-twitter ecosystem with the persistence of a virus: Anthropic and OpenAI's combined annual recurring revenue has allegedly surpassed $115 billion, "closing in on Microsoft." The source is Crypto Briefing, a publication better known for token coverage than enterprise software analysis. No methodology. No breakdown. No citation. Just a number that, if true, would rewrite the entire landscape of enterprise technology.
My eye is on the horizon, not the hourly candle. And from where I sit, this number does not survive contact with reality.
Context
Let me establish the baseline. According to the most generous public estimates from The Information and Bloomberg, OpenAI's annualized revenue for 2024 sits somewhere between $3.7 billion and $4 billion. Anthropic, the more conservative of the two, hovers around $1 billion. Combined, we're looking at roughly $4.7 billion in annualized revenue—not $115 billion.
The gap between these figures is not a rounding error. It is a chasm that suggests either a fundamental misunderstanding of what "ARR" means, a deliberate conflation of contract value with recognized revenue, or simply a number manufactured to serve a narrative.
To put this in perspective: Microsoft's commercial cloud revenue—which includes Azure, Office 365 commercial, and LinkedIn commercial—generated approximately $160 billion in fiscal 2024. Even if we accept the Crypto Briefing figure at face value, $115 billion would represent roughly 72% of Microsoft's entire cloud business. Two companies with a combined workforce of perhaps 5,000 employees would be generating revenue per employee that would make Goldman Sachs look like a struggling startup.
The bust was not an end, but a necessary pruning. And what we're witnessing now is the pruning of analytical standards in financial media.
Core
Let me walk through the mathematics of why this number fails basic scrutiny, because the exercise reveals something important about how narratives propagate in the digital asset ecosystem.
First, the revenue-per-employee test. OpenAI employs approximately 1,500 people. Anthropic employs roughly 800. Combined, that's 2,300 employees generating $115 billion in ARR. That works out to $50 million per employee. For context, Microsoft generates about $1.1 million per employee. Apple, the most efficient large company on Earth, generates about $2.5 million per employee. The Crypto Briefing figure would require Anthropic and OpenAI to be 20 to 45 times more efficient than the most productive companies in human history.
Second, the market size test. The entire global enterprise software market is estimated at roughly $500 billion annually. For OpenAI and Anthropic to generate $115 billion in ARR, they would need to capture nearly a quarter of the entire enterprise software market within three years of launching commercial products. Salesforce, which has spent 25 years building its enterprise footprint, generates about $35 billion in annual revenue. The claim that two startups have already achieved 3.3 times Salesforce's revenue is not just improbable—it's absurd.
Third, the infrastructure test. Based on my audit experience with digital asset protocols and my work modeling compute costs for AI companies, I can estimate that serving $115 billion in AI revenue would require inference compute costs of at least $30-40 billion annually. That's roughly 15-20 times the current total revenue of NVIDIA's data center business. The physical infrastructure simply does not exist to support this level of AI service delivery.
So what's actually happening here? I believe we're witnessing a specific type of data corruption that I've seen repeatedly in the crypto ecosystem: the conflation of "committed contract value" with "recognized revenue." When enterprise customers sign multi-year agreements with AI providers, they often commit to total contract values that include future capacity reservations, professional services, and support fees. These numbers can be 5-10 times the actual annualized revenue. A $10 billion multi-year commitment from a major enterprise customer might be reported as "$10 billion ARR" by a sloppy analyst, when the actual annualized figure is closer to $2-3 billion.
There's also the possibility of simple unit confusion. The original source may have intended "$11.5 billion" (11.5B) and lost a decimal point somewhere in transmission. Even that figure would be generous—it would still be 2.4 times the combined public estimates—but it would at least be in the realm of plausibility for a forward-looking projection that includes committed but undelivered capacity.
Contrarian
Here's where I diverge from the obvious takeaway. The instinctive response to this kind of data corruption is to dismiss the entire AI revenue narrative as hype. That would be a mistake.
The underlying trend—that AI companies are experiencing unprecedented revenue growth—is real. OpenAI's revenue grew roughly 200% year-over-year in 2024. Anthropic's growth was even faster, albeit from a smaller base. The question is not whether AI is becoming a significant enterprise software category. It clearly is. The question is whether the rate of growth justifies the valuation multiples currently being assigned to AI companies.
This is where the crypto parallel becomes instructive. In 2021, we saw DeFi protocols report "total value locked" figures that conflated deposited assets with genuine economic activity. The TVL narrative drove massive capital inflows, which created a self-fulfilling prophecy of growth—until the music stopped. The same dynamic is now playing out in AI, where "ARR" is becoming the new "TVL": a metric that can be gamed, inflated, and weaponized for fundraising purposes.
The contrarian insight is this: the Crypto Briefing article, despite its laughable data, is actually a useful signal. It tells us that the AI hype cycle has reached the stage where even crypto media is trying to capture the narrative. Historically, when narratives cross from specialist media into adjacent ecosystems, we're approaching peak enthusiasm. The fact that crypto investors are being pitched AI revenue stories suggests that the marginal buyer of AI equity is becoming less sophisticated.
Takeaway
The $115 billion figure will be forgotten within weeks, replaced by the next attention-grabbing headline. But the pattern it represents—the weaponization of financial metrics to serve narrative goals—deserves more scrutiny.
My eye is on the horizon, not the hourly candle. And on that horizon, I see a market that is increasingly bifurcated between companies with genuine revenue traction (OpenAI, Anthropic, Microsoft's AI division) and a speculative ecosystem that trades on narrative momentum rather than fundamental value. The former will survive the coming consolidation. The latter will provide the cautionary tales for the next cycle.
The question investors should be asking is not whether AI companies are growing—they clearly are. The question is whether the growth is sustainable, profitable, and defensible. And that question cannot be answered by a single headline number, no matter how impressive it looks in a crypto newsletter.
Disillusionment is data. Act accordingly.