Tech
The AI Correction Will Not Be Evenly Distributed | Computer Weekly
When the numbers coming out of the biggest AI companies get reported, the coverage is almost always the same: revenue up, growth accelerating, the boom is real. What almost nobody asks is what kind of revenue it is. In AI right now, that question is being skipped entirely. It’s the only one that matters.
Any investor who has sat across from a founder in a pitch meeting knows that headline revenue is just the starting point. The real questions come after: Is this B2B or B2C? Is it contracted or casual? Does the use case suggest land-and-expand potential, or is this customer already at their ceiling? Is the product embedded in something the customer cannot easily stop doing, or is it a nice-to-have competing with shrinking budgets and fading attention? These questions are table stakes at the startup level. They have almost completely vanished from the conversation about the companies now defining the AI landscape.
Take Anthropic and OpenAI. By most coverage, OpenAI is the dominant player – larger revenue, broader adoption, a product that has become genuinely cultural. That may all be true. But when you ask what colour that revenue is, the picture gets more complicated. OpenAI’s CFO confirmed that roughly 75% of its revenue comes from consumer subscriptions. ChatGPT has somewhere in the range of 800 million weekly active users – and only about 5% are paying subscribers. That is an enormous base resting on consumer willingness to pay for something most people still access for free, competing with curiosity, with free alternatives, and with whatever captures attention next. Consumer subscriptions cancel quietly and they cancel fast.
Anthropic’s revenue is built on integration
Anthropic’s revenue is smaller. But look at where it comes from. Approximately 80% comes from enterprise customers. Over 500 companies now spend more than $1 million annually on Claude. Eight of the Fortune 10 are customers. Claude Code, a tool embedded directly into developer workflows, went from zero to $2.5 billion in annualized revenue in roughly nine months. The result is a monetization gap that rarely gets discussed: Anthropic generates roughly $211 per monthly user while OpenAI generates roughly $25 per weekly user. That is not a small difference. It reflects what happens when revenue is built on integration rather than attention.
When a business has embedded AI into its compliance process, its coding infrastructure, or its data operations, switching is not a casual decision. It is an engineering project, a procurement process, and an organizational headache. That friction is not a bug; it is the entire point. It is what makes a dollar of Anthropic’s revenue structurally different from a dollar of consumer subscription revenue, regardless of the size of the number attached to it.
Lessons from SaaS
This is not a new lesson. The 2022 SaaS correction made it visible at a category level. When pressure hit, it did not hit evenly. Public SaaS multiples fell an average of 67% from their 2021 peak – but within that average, some companies saw multiples fall 90% while infrastructure and security tools largely held. The companies that took the worst hits were not necessarily bad businesses with bad products. They had the wrong colour revenue for a pressure environment. The market treated them as equivalent until the moment it didn’t.
AI will produce extreme divison
AI will produce a more extreme version of that divergence. Two reasons. First, the hype cycle is larger than anything SaaS produced – the speed of adoption, the scale of investment, and the cultural footprint of these products have created a wider gap between perceived value and embedded value than we have seen before. Second, the consumer-versus-enterprise variance is wider. SaaS was predominantly a business product. AI has gone consumer in a way SaaS never fully did, which means a much larger share of current AI revenue sits in the category most vulnerable to pressure. When that pressure arrives, the disaggregation will be severe and it will not look like a uniform correction. It will look like two completely different industries reporting results in the same earnings cycle.
The boom-or-bust framing that dominates AI coverage is the wrong question. Some of this is a boom. Some of it is not. The difference will not show up in total revenue figures until it is too late to be useful information. The question worth asking now is simpler and harder: which revenue survives pressure? That answer depends entirely on use case, contract structure, and how deeply the tool is actually embedded in how people and businesses work. We do not yet have a clean public way to measure it. That is exactly the problem.
Judah Taub is the founder and managing partner of Hetz Ventures, an Israeli early-stage venture capital firm specializing in cybersecurity, data, and AI infrastructure.