Markets spent this year sorting technology companies into AI winners and AI losers. In the sell-off some call the SaaSpocalypse, investors fled whole categories on that logic, and IT services was filed under losers to the tune of almost $280 billion. That filing was wrong, for two reasons the bear case never examined.

Enterprises must refactor decades of legacy estate before any AI works, and cheaper delivery expands demand rather than shrinking it. The last four weeks suggest the market is quietly refiling the sector as AI neutral. Progress, but short of the mark.

Over that period Accenture has gained 30%, Cognizant 32% and Capgemini 16%. Two of the three raised guidance on the way up.

The case against the loser thesis went up here on 28 July. I hate to say I told you so, which is a lie. Analysts love saying it. It is most of the job satisfaction.

Repricing, not rethinking

The sharpest move came on 29 July, the day after publication, which I promise was a coincidence. Cognizant closed up 11.2%, Sopra Steria 13.6%, Alten 19.5%, Accenture and Capgemini both above 5%. The whole European and American complex re-rated in a session.

The reasons were mundane. Jefferies upgraded Indian IT on 27 July, Cognizant reported on the 29th and lifted full-year earnings guidance, Capgemini reported on the 30th and raised revenue growth guidance to 8.5-9%.

Read that again. In the fortnight the market supposedly grasped that AI was destroying IT services, two of the largest providers on earth raised guidance. Cognizant also confirmed it is deploying Anthropic’s Claude at Travelport to modernise how that company builds and maintains software, booked as revenue in the release that raised its outlook.

One caveat for the record. This was not a called bottom. Nifty IT turned on 1 July, almost four weeks earlier, so the Indian recovery was already running.

The first force: somebody has to refactor the estate

The bear case rested on a jump nobody seemed keen to examine. AI automates a growing share of technology work, so enterprises need less human effort, so services revenues collapse. The middle step is fine. The last does not follow.

Before a bank becomes AI-native, somebody must modernise decades of legacy systems, fix the data, implement governance and take responsibility when it fails at two in the morning inside a regulated business. That refactoring is no footnote to the AI revolution. For most enterprises it is the revolution, and an enormous services market.

The second force: the price elasticity of AI

Suppose AI lets a provider deliver a programme with 30% less effort. The assumption is the client spends 30% less. But what if it makes five more programmes viable?

Think of renovating a house you still live in. If the cost halves, you do not hand back half the budget and admire your smaller kitchen. You do the roof and the wiring too. Enterprise estates are not short of things needing repair.

Cloud collapsed the unit cost of computing and the world consumed vastly more of it. No law requires cheaper services to mean lower total spending. Elasticity, not automation, decides how large this industry becomes, and almost nobody is modelling it.

The same logic breaks the habit of reading headcount as revenue. Cognizant is training 5,000 engineers and 10,000 business operators on AI tools by Q4, out of 356,700 staff. Fewer people is not a smaller business if each produces more.

Losers, neutrals and winners

Neutral is not the final answer either, and the market has already begun separating the field.

Capgemini gained 16% against roughly 30% at the other two, and fell on results day despite raising revenue guidance, because normalised EPS dropped almost 12% on financing, tax and restructuring costs. Growth arrived with a bill attached. Infosys cut guidance.

The real AI losers sit inside this sector rather than defining it. Providers using AI merely to strip cost from existing delivery enter a productivity race they cannot win: clients will demand the savings, competitors will match them, and the margin is gone within two contract cycles. The winners will change what they sell and how they advise, rather than simply dropping their price.

Before anyone gets too pleased

Four weeks is not a thesis. A 30% month is a retracement, not a re-rating. Accenture still sits 38% below its 52-week high, Cognizant 33% and Capgemini 30%. The market has stopped pricing extinction without starting to price growth, which is what neutral looks like.

What would overturn the argument is bookings deteriorating across providers over consecutive quarters while AI-attributed revenue climbs. That would mean productivity is eating demand and the bears were early, not wrong. Cognizant’s 1.3x book-to-bill and Capgemini’s 1.04 say not yet.

The market priced the end state before pricing the transition. The transition may finally be getting a price.

I wouldn’t dream of mentioning it twice.