On February 23, IBM lost 13.15% of its market value. Not because of a bad earnings report. Not because a competitor launched a superior product. Because Anthropic published a blog post describing how Claude Code can automate the exploration and analysis phases of COBOL modernization. A description of something that could happen erased tens of billions of dollars from a company that had just announced it was tripling its hiring.
The Partnership
Five months earlier, in October 2025, Anthropic and IBM announced a partnership to make Claude models available through IBM's platform. In December, IBM CEO Arvind Krishna told The Verge there was no AI bubble and that quantum computing would be IBM's next growth engine.
- Oct 2025 Anthropic and IBM partner to make Claude models available through IBM's enterprise platform.
- Dec 2025 IBM CEO Arvind Krishna says there is no AI bubble.
- Feb 13 IBM announces plans to triple entry-level hiring in the US in 2026.
-
FEB 23Anthropic publishes a blog post about COBOL modernization. IBM drops 13.15%.
The partner that gave IBM access to Claude also published the data that destroyed 13% of its market value. The blog post described how Claude Code could automate what IBM's consulting division sells — the painstaking, expensive work of understanding and modernizing the millions of lines of COBOL that still run banks, insurance companies, and government agencies. IBM itself had launched its own COBOL modernization tool in 2023. But the market didn't care about IBM's tool. It cared about Anthropic's description of what a general-purpose AI could do to IBM's business.
Basically Zero
On the same day IBM lost 13.15%, the Washington Post reported that economists at Goldman Sachs and JPMorgan had concluded that the AI boom contributed "basically zero" to US economic growth in 2025.
This is the number that matters. Hundreds of billions of dollars have been invested in AI infrastructure. Every major corporation has announced AI initiatives. Consulting firms are tracking executives' login frequency. And when Goldman Sachs went looking for AI's impact in the actual economy — in GDP, in productivity statistics, in the measurable output of the world's largest economy — they found basically nothing.
The juxtaposition is the insight. AI was simultaneously powerful enough to erase tens of billions from a blue-chip company's market value with a blog post and invisible enough to escape detection in the broadest measure of economic activity. Narrative power and productive power are diverging, and the gap is widening.
The Belief Gap
The broader market seemed to notice. Software stocks extended their weeks-long selloff on Monday. The Wall Street Journal reported that some AI startups were boosting their valuations by raising capital in back-to-back rounds, creating the appearance of rapid growth through financial engineering rather than revenue.
And Sam Altman, in a series of interviews published in the New York Times, acknowledged that AI's adoption faces more resistance than expected — what he called a "belief gap." The CEO of the most valuable AI company in the world is telling you that people don't believe in the product as much as the valuation requires.
Meanwhile, Anthropic was having its own kind of day. The same company whose blog post crashed IBM also accused DeepSeek, MiniMax, and Moonshot of violating its terms of service. Defense Secretary Pete Hegseth summoned Dario Amodei to the Pentagon. An Anthropic-backed super PAC began running ads urging Congress to pass AI legislation. One company, on one day, was crashing a partner's stock, accusing Chinese competitors of IP theft, meeting with the Defense Secretary, and running political ads. This is what narrative power looks like when it's fully operational.
The Thirteen Percent
In 2020, when the pandemic hit and millions of Americans filed for unemployment through COBOL-based government systems, IBM released a free course to train new COBOL programmers. The problem was a shortage of people who understood the old code. Six years later, the problem is a blog post claiming AI understands it.
Neither version of the problem was really about COBOL. In 2020, the constraint was human expertise — too few programmers who knew a sixty-year-old language. In 2026, the constraint is narrative — a single company's description of automation, which has not yet been deployed at scale, which has not yet replaced a single line of production COBOL, which exists so far only as a blog post, was sufficient to destroy 13% of IBM's value.
AI contributed basically zero to US economic growth in 2025. It contributed negative thirteen percent to IBM's market cap in an afternoon.
This is the market we're in. The companies tracking AI adoption are not the same companies proving it works. The companies whose valuations depend on AI's promise are not the same companies showing AI's results. And a blog post from a partner — not a competitor, a partner — can move more market value in a day than the entire AI industry moved GDP in a year.