The story you’re being told is that Anthropic’s finances are spectacular. Leaked documents, podcast hype, and anonymous sources paint a picture of a company printing money on its way to a fall IPO. According to Marcus on AI, that picture leaves out most of the important parts.
Gary Marcus is watching this unfold in real time, and what he sees is a coordinated hype cycle running on undisclosed math. Anthropic itself is in an SEC-monitored “quiet period” ahead of its expected fall IPO, which limits what the company can say. But as Marcus on AI puts it: “Anthropic may be ‘quiet’, but its fans are not.”
📊 What’s Actually Being Claimed
The bullish case is being built from a handful of eye-popping numbers, most of them from unnamed people or leaked files:
- A Reuters report says Anthropic is “projecting 2028 revenue of roughly $190 billion to $200 billion,” sourced to “two people familiar with the company’s financials.”
- Bloomberg received leaked documents talking up an amazing Q2.
- On the All-In podcast, investor Gavin Baker claimed Anthropic was “making money on every token.”
- Podcaster Dwarkesh Patel projected the company “likely ends the year with ~$100-150B of revenue,” a claim he later partly walked back.
Notice the pattern. The most dramatic figures ride on anonymous sourcing and math nobody has shown. Marcus on AI’s core point is that the projections arrive without the context that would let anyone judge them.
🔍 Why the Skepticism Matters
Here’s what stands out. A revenue projection is not revenue. A 2028 forecast of $200 billion is a story about the future, not a line on a current balance sheet. And “making money on every token” is a claim about unit economics that quietly ignores the enormous fixed costs of training frontier models, the compute bills, and the talent war.
Dwarkesh Patel partly retracting his own $100-150B figure tells you how soft these numbers are. When the person making the projection backs away from it within weeks, that’s a signal about the reliability of the whole genre of claim.
This is significant because the timing isn’t an accident. A quiet period is supposed to prevent selective hyping of a company before retail investors can buy in. When the company goes silent but leaks and friendly podcasters fill the gap with unverifiable optimism, the effect is the same as a marketing campaign, without the accountability.
💡 The Broader Pattern
Anthropic isn’t operating in a vacuum. The entire generative AI sector is running on forward-looking valuations that assume today’s growth curves hold for years. OpenAI, the foundation model race, the datacenter buildout, all of it depends on investors accepting projections in place of proven profit. Marcus has been one of the most consistent voices arguing that the gap between AI hype and AI economics is wider than the market admits.
An IPO is where that gap gets tested with real money from real people.
✅ Practical Takeaways
For practitioners, investors, and operators watching this space:
- Separate projections from results. Ask whether a number describes what happened or what someone hopes will happen.
- Follow the sourcing. “According to two people familiar with the financials” is not the same as an audited filing. Weight it accordingly.
- Watch unit-economics claims closely. “Profitable per token” can be true while the company still burns billions overall. Both can hold at once.
- Note who’s talking and their stake. Investors and podcasters talking up a pre-IPO company often benefit from the narrative.
- Wait for the S-1. The actual IPO filing will force disclosures that leaks and podcasts never will.
Marcus has a long record of calling AI hype early and being proven right on the fundamentals, even when the market ran the other way for a while. Whether Anthropic’s real numbers justify the excitement is a question the filing will answer, not the leakers. Until then, treat the spectacular figures as marketing, not fact.
For the full breakdown of what’s been left out, the original analysis at Marcus on AI is worth reading in full.