The SaaS versus AI battle is on and the market is confused, writes Jason Walsh
Blogs
Image: ThisIsEngineering/Pexels
Back when artificial intelligence (AI) was merely a threat to mere artists, writers and translators, the market response was muted at best. At worst it outperformed averages, with boosters drunk on visions of creative workers rattling tin cups outside the cinema where the latest AI-made blockbusters are screened.
This year, that changed. Today, with AI apparently threatening people who really matter, things are a little different. The first casualty was the software industry – effortlessly profitable, and global capitalism’s hidden financial obesity problem.
Names from Atlassian to Salesforce have seen valuations slashed by a quarter or more, with the iShares Expanded Tech-Software Sector ETF plunging from its all-time high of $117 to $85. Storied names like Adobe, which, lest we forget, was once a major contributor to the history of computing, suffered: its share price collapsing more than 60% from its pandemic-era peak.
Even giants stumbled: despite being a major player in AI, Microsoft is down 18% year to date, hit first by a disappointing cloud growth forecast and then by the broader software selloff that intensified after Anthropic, developer of the Claude large language model (LLM), released Cowork, a desktop agent that lets non-technical users perform work tasks without paying endless software subscriptions.
Private equity, which is heavily invested in software companies, often in niche verticals, is now facing the music, according to recent reports in the Financial Times and the Economist. Stuffed with debt, private equity firms were heretofore able to gorge themselves on recurring revenue from software companies. Private credit (or, if you prefer, non-bank lending or shadow banking) is also worried. UBS estimates that between 25% and 35% of private credit is exposed to AI disruption risk, with potential default rates climbing to 13%. Apollo Global Management’s John Zito reportedly stunned an audience of investors in Toronto by asking: “is software dead?” Some firms have reportedly begun hiring consultants to check their portfolios for vulnerable businesses.
Then, this week, things took a turn for the worse.
Wall Street is now on a rampage, searching for, and dumping, anything that AI could get its claws into. Commercial credit rating and data analytics firms followed software off the cliff: S&P Global dropped roughly 12% after issuing a profit forecast below expectations. Its confrère, Moody’s, fell in sympathy. Late in the week, wealth management firms nosedived following the announcement of Hazel, a new AI-enabled tax planning tool developed by Altruist Corp, a fintech start-up led by former Morgan Stanley and Pimco staffers. Charles Schwab and LPL Financial both fell around 8% on the day.
To be clear, this is not a stock market crash. Overall, share indices have stayed largely static, but individual companies are falling, with a ‘sell the disrupted’ rotation under way. What we will now find out is if the market has moved from speculative fear toward fundamental repricing.
The craft and the code
But, just as the price tag is not the painting, the numbers don’t tell the whole story.
AI doesn’t eliminate software development. It separates conception from execution. The developer becomes, at best, a ‘prompt’ giver and result reviewer, transformed from a creator into a supervisor of output and a manager of machines.
The late critic Herbert Read would say this is the same thing industrialisation did to craft production, and that the result isn’t efficiency but degradation, because the knowledge that lives in the act of making gets lost when making is automated. The developer who writes code understands the problem differently from the developer who reviews AI-generated code, just as the carpenter who shapes wood understands the material differently from the factory supervisor who checks output. Chemist and philosopher of science Michael Polanyi called this ‘tacit knowledge’: we can know more than we can tell, and when that telling is all that remains, something essential has been lost.
This is not mere sentiment: knowledge acquired in practice, by the act of doing, is precisely the kind of knowledge that cannot be formalised and handed to a machine. Indeed, the doer may not even know they have the knowledge, or may know they have it but not quite know what the ‘it’ is. When that kind of knowledge atrophies the system becomes brittle in ways that are invisible until something breaks.
This, then, is the central irony. The investors dumping software stocks are engaged in exactly the same separation: they don’t really understand what AI does or does not do, any more than they truly understood what software did or did not do beyond collecting monthly subscription fees. They are trading a signal, not a substance.
In short, the market, which as we are endlessly told is a mechanism for price discovery, is pricing in vibes.
Of course, there has long been an alternative vision of computer use where every end-user is a ‘developer’. We never got there, substantially backing away from that idea in the mid-1980s, but if AI does mean we, each of us, take charge of our computing that doesn’t mean we understand how the computation actually works.
None of this is to say that the ongoing disruption is imaginary. Software-as-a-service (SaaS) was always, to some degree, a toll booth: overthrowing the previous model of buy-once-use-forever, it charged recurring fees for access to functionality that, once built, cost relatively little to maintain.
AI threatens that model directly, and some of these companies genuinely are overvalued, or at least they were. But the indiscriminate nature of the selloff, in which Thomson Reuters, Adobe and (surprisingly ‘sticky’) vertical niches like golf course management platforms are treated as a single category, tells us something about the quality of the analysis being applied.
The market, in short, knows something is changing but not quite what, certainly not by how much, and, most of all, not for whom. It is reviewing the output without writing the code. It may be right, and AI may drive the value of computation down even further, but it is perhaps reassuring to know that the problem is not new. Read, were he alive, would recognise the pattern.