Welcome back to Building with AI. Steve Eisman, the investor made famous by The Big Short, has been asking where the money for all this AI spending is actually going to come from. After last Thursday's sell-off, a lot more people are asking the questions I've been asking myself: is the bubble going to burst, what happens when it does, and how do we protect our businesses from that risk?
I can't tell you whether we've hit the top, and I'd be wary of anyone who says they can. What I'm confident about is who comes through a correction in good shape: the companies that used AI to fix real problems in processes that already worked. If you automate a bad process, you only get bad results faster, and that holds in a boom or a bust.
The Story
Is the AI bubble about to burst?
On November 5, the Nasdaq fell 5.5%, its worst day since April 2025, and the Nasdaq 100 lost about $2.5 trillion in market value, Reuters reported. Chipmakers and the neoclouds that rent out AI computing power fell hardest, on worries that data center projects worth billions could be delayed. As one investment manager put it to Reuters, "the market's moving from the promise of AI to the price of AI."
The demand behind the spending is real. Anthropic's updated IPO filing shows revenue of $17.0 billion in the third quarter alone (July to September), up 1,127% on a year earlier. But growth from the previous quarter slowed to about 47%, down from 144% in the second quarter, CNBC noted. The filing also shows how big the bet has become: about $895 billion in committed spending, "substantially all of which is related to compute and infrastructure," with roughly $447.5 billion due before the end of 2030.
The filing is also unusually open about risk. MIT Technology Review counted roughly 100 pages on it in the original 335-page version, including warnings that AI at scale could "erode trust, harm society, and destabilize the geopolitical order." Its AI editor, Will Douglas Heaven, calls Anthropic's public messaging "a master class in narrative capture," with the company presenting itself as a safety organization, a profit-driven business and a race to build superhuman AI all at once. That's a lot of story around one company, so it helps to look at the wider economy too.
For most businesses, the payoff is arriving more slowly. In its November Monetary Policy Report, the Bank of England says AI-related activity accounts for just under a fifth of UK economic growth since the start of 2026. Most of that comes from companies developing and implementing AI "rather than broader gains from their adoption elsewhere in the economy." Firms in the Bank's survey estimate that AI has raised their productivity by about 0.5% over the past three years, although they expect a bigger lift, about 0.9 percentage points a year, over the next three.
What I take from this is that the technology is real, but the bill is arriving before most companies have seen the benefit. Back in April, I told our leadership team I expected the big AI providers to move to tighter, pay-for-what-you-use pricing, with the cheaper "unlimited" plans starting to disappear. A correction would only speed that up, and I think small and mid-size companies will feel it hardest. Some tools will get more expensive, and some vendors won't survive. The companies that feel it most will be the ones that bought AI because everyone else was, without a clear problem to solve. And as I wrote about SaaS two weeks ago, rebuilding your own tools with AI can look cheap right up until you count the cost of maintaining them for years.
Sources: Reuters · Anthropic S-1/A · CNBC · MIT Technology Review · Bank of England
The Rundown
Goldman takes coding agents back from its traders
Goldman Sachs has restricted Claude Code and GitHub's coding tools to its engineers, cutting off traders and bankers who had used them to build their own tools, State of AI reported, citing Bloomberg. This speaks to the question I opened this newsletter with in Edition 01: who owns what AI builds? A trader's script runs with that trader's access to client data, often without the review and testing an engineer's code gets.
Source: State of AI
Can we see inside AI yet?
Nick Cammarata, a researcher at the startup Goodfire, published a 26,000-word essay arguing that AI models may be far easier to understand, and so to monitor and control, than researchers believed. Sam Altman praised it, but leading researchers are skeptical, and the work hasn't been peer-reviewed, Futurism reports. One told MIT Technology Review that researchers understand less than 0.01% of how top models work inside. Until that changes, keeping AI safe at work still depends on people: clear owners, real review and limits on what agents can reach.
Sources: Futurism · MIT Technology Review
A startup for companies that want to own their AI
Applied Compute raised $350 million at a $3.25 billion valuation on November 2, after reaching $100 million in annualized revenue less than 18 months after it was founded. It helps companies such as Microsoft, Shopify and DoorDash train and run their own AI models on open-weight foundations, and says it cut one customer's production AI costs by up to 47%. If a correction pushes up the price of the big providers' models, owning more of your AI stack becomes a more serious option.
Source: Applied Compute
The Human Side
The AI layoff trap
What happens if every company uses AI to cut jobs at once? Economists Brett Hemenway Falk and Gerry Tsoukalas modeled exactly that in a paper called The AI Layoff Trap, which Tsoukalas walked through in an interview with the BBC. Each company that automates keeps the full cost saving, but only bears a small share of the lost spending when the people it lets go stop buying. The rest falls on its competitors. That pushes rational companies into an automation race that displaces far more workers than is good for anyone, owners included.
The paper puts into numbers something I've felt for a while. Individually, every company is doing what's best for itself. Collectively, we're racing toward a cliff on AI and jobs. Everyone agrees someone should hit the brakes, but nobody wants to go first, because if I slow down, somebody else wins.
The evidence so far suggests cutting people isn't even the winning strategy. Palantir's CTO, Shyam Sankar, told Stratechery that almost all the customers he has seen use AI mainly to cut staff and costs failed to reach even that goal, while the ones succeeding use it to win their markets. And the Bank of England's contacts report that AI is making firms less likely to hire or replace workers, but not yet more likely to let people go.
Most of us find purpose and meaning in our work, and a job that gives someone dignity is worth protecting. I'd like to see more leaders use AI to grow what their people can do and take on work they couldn't before. By Sankar's account, that's where the winners are, and it keeps people doing work that matters.
Sources: arXiv · The Next Web · Bank of England
Try This Week
Stress-test your AI vendors
List the AI tools your business relies on, and for each one ask what you would do if the price doubled or the vendor disappeared. Note where each connects to the systems you'll keep no matter what, like your ERP, CRM and finance systems, because that's where a forced switch hurts most. Anything with no fallback is worth a conversation with your team now, while you have time to plan it.
One for your IT team
If your developers use Postman's desktop app, have them check their coding agents. A GitHub issue shows the app quietly installed plugins into Claude Code and OpenAI's Codex, telling the agents to route API work through Postman, which says it is reverting the change. Ask developers to remove any Postman plugins or hooks they didn't install themselves.
Source: GitHub
That's it for this week. If your AI costs doubled tomorrow, which tools would you keep paying for? Hit reply and let me know.
References
- Reuters, Investors wipe $2.5 trillion off Nasdaq as AI data center stocks slide
- Anthropic, PBC, Amendment No. 1 to Form S-1 Registration Statement (November 2, 2026)
- CNBC, Anthropic reports $17 billion in Q3 revenue, purchase commitments of $895.4 billion in revised S-1
- MIT Technology Review, Will the real Anthropic please stand up (Will Douglas Heaven, October 26, 2026)
- Bank of England, Monetary Policy Report, November 2026
- State of AI, Goldman Sachs Is Pulling Claude Code From Its Bankers and Traders
- Futurism, Explosive Document Claims That AI Is Not What We Think (October 19, 2026)
- MIT Technology Review, “We were looking at it wrong”: The explosive claim that’s upended the AI world (Will Douglas Heaven, October 19, 2026)
- Applied Compute, Applied Compute hits $100M run rate and raises $350M
- Brett Hemenway Falk and Gerry Tsoukalas, The AI Layoff Trap (arXiv)
- The Next Web, Palantir CTO tells Stratechery he doesn't regret working with Anthropic
- GitHub, postmanlabs/postman-app-support issue #14370: Desktop auto-installs Codex/Claude plugins
