The dominant narrative in recent years has cast artificial intelligence (AI) as unambiguously bullish. Software stocks are now testing that assumption.

It’s been a very bloody week for the latter, with nearly $1 trillion from software and services stocks wiped out following the so-called Claude crash, sparked by Anthropic’s release of new legal and enterprise tools linked to its Claude chatbot.

Thomson Reuters suffered its biggest one-day fall on record and has now lost more than half its value over the past six months. It was not an outlier.

Software shares had been weakening for months; the S&P 500’s software and services index has fallen by a quarter since October, with the latest rout accelerating an already uncomfortable trend. Shares in Salesforce, which employs more than 2,500 people in Ireland, had already lost some 40 per cent of their value even before the Claude crash. That’s just one example.

Some suggest the sell-off is overdone, saying security concerns, data ownership and regulatory constraints all limit how quickly AI agents can displace established software.

Nvidia boss Jensen Huang called it “illogical”. Arm CEO Rene Haas called it a “micro-hysteria”. Not only is the sector now “guilty until proven innocent”, said JPMorgan, it is “now being sentenced before trial”.

Perhaps, but even a slower pace of disruption need not be benign. Besides, the US software sector still trades at roughly eight times sales, says Bespoke Investment, suggesting valuation support “is a long way away”.

So far, what we have seen is a large market rotation, not a sell-off, as evidenced by investors piling into defensive names and the S&P 500 remaining near all-time highs. In that sense, the software rout seems to reflect an effort to price AI more accurately, a messy process of price discovery rather than outright panic.

AI bubble anxiety is everywhere, but are valuation concerns overstated?Opens in new window ]

After all, if the technology is as transformative as its advocates claim, it should bolster some business models, but threaten others. The market is beginning to draw that distinction. Whether it has drawn it too sharply remains an open question.