Years ago, when I was at a firm growing 80% in a hot market, a board member observed: "You hear people talk about the good old days? Well, gang, appreciate this moment, for these are the good old days.”
If those days were good, the current growth of Anthropic and OpenAI in the enterprise is…well, I can't summon the proper superlative.
Take Anthropic. At the start of this year (as it turned five years old), its revenue run rate was $9 billion. By May it was $47 billion. By year end it will most likely clear $90 billion. To put that in perspective, look at the leaders of prior enterprise software waves:
SAP, defining enterprise applications, needed 22 years to reach $1 billion.
Oracle, riding the client-server wave, took nearly 14 years to reach its first $1 billion.
Salesforce, leading the SaaS revolution, took a decade to reach its first billion.
I had a front-row seat for all three of those market sensations, and the rides were exhilarating. What's happening now? It's simply nuts. 100x the growth rate.
And this velocity in enterprise AI is creating profound challenges. People, companies and markets weren’t built for this velocity.
Human capacity can't keep pace. Internal IT organizations lack the skills to deploy AI properly at scale. So do the IT services firms. There's open demand for some 200,000 forward-deployed engineers (FDEs), yet only around 2,000 true FDEs exist.
Renegade AI usage. A key to AI growth? Users can purchase by the drink, instantly. SAP, Oracle and Salesforce had to be sold (through CIO offices, procurement, etc. by professional sales forces). Access to frontier models is bought, with almost no friction. This has allowed for remarkable growth…and now sizable governance headaches.
Pricing and unit economics are still in flux. Enterprises are learning that using a top-tier model for every task is like hiring a Ph.D. to answer the phones. Also, cheaper challengers like DeepSeek are crashing the price floor, delivering comparable output at a fraction of the cost. This combo of a.) smarter model selection from within, and b.) massive price pressure from without means the economics that built OpenAI and Anthropic are anything but settled.
New security threats emerge. This past week, both OpenAI and Anthropic admitted that during their own internal cyber tests, their own models broke into the live systems of third-party companies. If the labs that built these things can't contain them on their home turf, this raises huge concerns for security teams at Coca-Cola, UnitedHealthcare, or Walmart (smart people, thin budgets, a hundred other fires).
Will these four gaps get covered in time? Yes. They always do: the engineers get trained, the governance catches up, the pricing settles, the sandboxes get thicker walls. But for now, the distance between what these models can do and what any of us can manage is the widest it's ever been. And that gap is where the money, the risk, and the vertigo all live at once.
Life in the fast lane, surely make you lose your mind.
The week's most important AI news, curated for you:
