Blog posts do not usually erase tens of billions of dollars.
On Monday, February 23, 2026, one did.
Anthropic published a technical post about running Claude Code on COBOL, and IBM stock fell roughly 13% before the closing bell.
The drop is the easy part of this story.
What matters more is that the same pattern has repeated on other tickers ever since.
Key Takeaways
IBM shares closed about 13% lower on Monday, February 23, 2026, the day Anthropic published a post claiming Claude Code can modernize COBOL in quarters rather than years.
The trigger was a vendor blog post, not an earnings miss, a downgrade, or a lost contract.
IBM had been an Anthropic partner since October 2025, so the selloff repriced a relationship investors had welcomed four months earlier.
IBM's same-day piece argued the mainframe's value sits in the platform, and noted roughly 40 percent of COBOL does not run on mainframes at all.
The COBOL panic was fully unwound within about three months, and IBM's later weakness came from an unrelated revenue warning in July.
The same catalyst has since moved other tickers in both directions, lifting Cognizant on a partnership and pressuring Alibaba on an allegation.
The IBM stock drop tied to Anthropic happened on Monday, February 23, 2026.
No earnings came out that morning.
IBM reported no results that morning, and the move traced back to a document published by another company entirely.
The catalyst was a post on Anthropic's own blog, published that day, describing how its coding agent handles COBOL, the language still running critical systems across banking, airlines, and government. Investors read it as a threat to the expensive human work of modernizing those systems.
That reading is what moved the stock, and whether it was correct is a separate question that the rest of this piece takes apart.
Anthropic claimed that AI collapses the discovery phase of COBOL modernization, not that it replaces the mainframe.
Its February 23 post says Claude Code can map dependencies across large codebases, write documentation for workflows nobody remembers, and flag risks that would otherwise take analysts months to find.
The headline claim is about time: "teams can modernize their COBOL codebase in quarters instead of years."
Read closely, that is a claim about understanding old code cheaply.
Understanding has always been the expensive part, because the business logic often exists nowhere except inside the code itself.
One thing to keep in mind: this is a vendor describing its own product, not an audited benchmark or a customer-verified case study.
Yes, and that is the strangest part of the episode.
IBM presented the deal as a productivity win for enterprise developers, and the market took it well at the time.
Four months later the same vendor published a blog post, and the market repriced IBM as the thing being disrupted.
Neither company announced any change to that relationship between those two dates.
Only the story had.
That gap between a partnership and a perceived threat is worth sitting with, because it is the clearest sign that February 23 was a sentiment event rather than a business event.
His line was blunt: "Translating code is one thing. Modernizing a platform is something else entirely."
The substance behind it is that enterprise COBOL on IBM Z sits inside an integrated stack of operating system, transaction managers, databases, and security layers, and the guarantees customers pay for come from that stack rather than from the language.
He also put a number on the mismatch: by IBM's own count, roughly 40 percent of COBOL does not run on mainframes at all, but on Windows, Linux, and other distributed platforms.
If that is right, a large share of the COBOL story investors priced into IBM that day was never IBM's business to lose.
Four of the clearest examples came from 2026, and they did not all move in the same direction.
That is the detail most coverage of this theme misses.
The list is longer than four. An early-February session knocked several information-services names lower at once, which is how you know the market was pricing a category, not a company.
Substitution claims push the named target down.
Partnerships and endorsements put the named company on the tooling side of the story, and Cognizant is the clearest case where the shares rose on one.
Accusations push a third party down, on reputational and regulatory risk rather than on product economics.
Same engine, three different signs.
Anyone who treated every Anthropic headline as automatically bearish would have missed Cognizant entirely, and would have misread what the IBM move was actually worth.
Start by asking which revenue line the announcement actually touches, then how big that line really is.
Four questions do most of the work:
Is this substitution or integration? A partnership and a capability claim look similar in a headline and point in opposite directions.
Which segment is exposed, and what share of revenue is it? Modernization services and platform hardware are not the same business.
Is the company selling a platform or selling hours? Hours compress when discovery gets cheap, while platform guarantees do not.
Has the company answered? A same-day, specific rebuttal is information, and silence is also information.
Anthropic now ships on a visible cadence, with model releases and enterprise partnerships landing through the year rather than at random.
Knowing that turns each announcement from a shock into a scheduled test of a thesis you already hold.
The IBM case is the worked example: the first-day move priced a substitution story, and the months that followed priced the more boring reality underneath it.
It depends on the product, because the three ways to hold US stock exposure on MEXC do not keep the same hours.
That matters for this theme specifically, since a blog post can land at an hour when US exchanges are shut.
Tokenized Stocks give economic exposure and trade around the clock.
Stock Futures are USDT-settled contracts with no ownership attached, and leverage is set per contract.
The structural point is worth stating plainly: a tokenized stock is not the same legal claim as a share, and a futures contract is not a claim on the company at all.
Anthropic itself is private, so there is no ordinary share to buy, and exposure runs through an Anthropic pre-IPO perpetual that conveys no shares, votes, or dividends. That contract prices against a composite pre-listing index rather than a spot market, and MEXC converts it into a standard stock futures contract if an IPO completes.
What date did IBM stock drop on Anthropic news?
Monday, February 23, 2026, the day Anthropic published its COBOL modernization post.
How much did IBM stock fall?
About 13% in a single session, worth roughly $30 billion in market value, per market data at the time.
What did Anthropic say about COBOL?
That Claude Code can automate the discovery and documentation work that made COBOL modernization expensive, cutting timelines from years to quarters.
Are IBM and Anthropic partners?
Yes, they announced a partnership in October 2025 to bring Claude into IBM's software development tools.
Did IBM respond?
IBM published a piece on the same debate the same day, arguing that translating code is not the same as modernizing a platform, though it did not name Anthropic.
Has IBM stock recovered?
It regained much of the loss over the following months without fully closing the gap, as of September 2026.
Which stocks does Claude threaten most?
Businesses that bill for the hours AI compresses, such as legacy code discovery and application modernization services, rather than platform vendors.
Did Cognizant stock rise on its Anthropic deal?
Yes, it climbed roughly 8% across the two sessions after the July 2026 partnership, per market data at the time.
Can I trade US stock exposure when the US market is closed?
Tokenized Stocks and Stock Futures trade around the clock, while RealStocks follow the US securities market schedule.
The first day after an AI announcement is a sentiment print.
The quarters after it are the verdict, and in IBM's case the two said different things.
Keep the framework rather than the headline: find the exposed revenue line, size it, and check whether the company answered.
If you want to act on the theme, start with what each product actually gives you.