Dear reader,
Accenture’s shares were up 22% during trading on Thursday 1 October. Before that session, they had lost about a third of their value this year amid investors’ fears that AI would take work away from IT services firms (Reuters). The whole sector had been falling: by 21 August India’s Nifty IT index was down about 20% this year, wiping $73bn off its ten members (Reuters). On 1 October the others rose with Accenture: Cognizant by 8%, IBM by 3%, and the US-listed shares of Wipro and Infosys by 6–7% (Reuters). Accenture forecast revenue growth of 3% to 6% in local currency for the new financial year. The midpoint, 4.5%, exceeded the 3.9% analysts expected; acquisitions are expected to contribute two to two and a half percentage points.
Revenue for the year to 31 August was $74.2bn, up 6% in dollars, and fourth-quarter bookings reached $22.2bn (Accenture).
Accenture gives clients AI productivity and sells more work
On the call with analysts, Julie Sweet, the chief executive, said prices were broadly stable across fiscal 2026 but lower in many business lines in the fourth quarter. She put the decline down to intense competition, without pointing to AI (transcript).
Keith Bachman of BMO asked what AI-driven price reductions Accenture was building into contract renewals. Sweet said the firm was giving clients more productivity from AI and making up for it with new kinds of work and larger project scope. Management expects new work to outweigh the efficiencies in existing contracts, although the results release lists reduced demand among the risks from AI.
More than 65% of Accenture’s bookings are fixed-price
Angie Park, the finance director, said fixed-price contracts now account for more than 65% of bookings and their share is growing. They include outcome-based contracts; Accenture does not disclose that share separately.
In a fixed-price contract the AI saving stays with the supplier unless the contract requires it to share. In a time-and-materials contract the client pays for days worked, so when AI shortens the work, the supplier’s revenue falls.
Wipro’s finance director Aparna Iyer said in April that AI was “helping us to deliver our fixed price programs better” (Wipro). Across more than 25 providers, HFS measured revenue per employee up about 1.7% a year and operating margin per employee up about 6.5%; HFS notes that acquisitions and restructuring also explain part of the rise (HFS).
Clients are asking for discounts
In April, HCLTech’s chief executive C. Vijayakumar estimated that prices for services hit by AI would fall by 3% to 5% a year over the next few years, and by 2% to 3% at HCLTech. He expected software development to feel it more as models improved (HCLTech).
In August, Reuters reported that Persistent Systems’ clients were asking for the same work for 25% to 30% less. Jimit Arora, chief executive of Everest Group, described a market in which buyers had the stronger negotiating position (Reuters).
Some clients have already negotiated a share of the savings. According to Reuters, Cognizant will split AI savings with its client Daimler Truck. In its cloud management deal with E.ON, HCLTech is paid nothing for the first year and from the second year earns on efficiency gains and business outcomes (Reuters).
In an HFS Research survey of 202 senior leaders at companies with revenue above $1bn, 77% planned to renegotiate prices with their providers and 48% to move some work to another provider. Only 17% wanted to renew with little change (HFS).
Firms keep hiring, but the work is changing
Accenture’s headcount rose by about 5%, according to an analyst on the call, to roughly 814,000. Sweet said revenue per person increased partly because of AI. She expects slower overall hiring in the coming year, while still planning more entry-level recruitment.
Infosys hired more than 20,000 graduates in its last financial year and expects to hire at least 20,000 this year (Infosys). McKinsey said in July that it aims to increase client-facing consultants by 20% this year, mostly at entry level (McKinsey).
Wipro’s chief technology officer told Reuters in September that AI had freed capacity equivalent to 20,000 employees, which the company redeployed. Engineers could supervise agents, move to other projects or retrain (Reuters).
Some firms are cutting. Cognizant expects $200m to $270m of severance and other personnel costs in Project Leap, a programme it links to AI-led efficiency and to sizing its workforce with AI (Cognizant). In April Capgemini proposed up to 748 redundancies in Spain, 6.8% of its Spanish staff. El País reported that the company linked them to AI and that most of the people affected had no assigned project (El País).
The model developers are helping consultancies build deployment teams. On 2 October, Anthropic announced a $100m training programme to prepare 10,000 Frontier Deployed Engineers by the end of 2027. Its first cohorts include major consultancies and enterprise customers (Anthropic). In September, Accenture and Google Cloud announced a 1,000-person forward-deployed engineering team (Accenture).
Firms that sell developers by the day will feel it first
In my view, the first to feel this will be firms that sell developers by the day, through staff augmentation or time-and-materials projects. The figures do not show it yet. Large enterprises are only starting to change how they build software around AI agents. When they do, demand for developers sold by the day could fall by about a third. Or they may simply order much more software for the same money.
Clients also want to pay differently. In an HFS study, 65% of companies pay providers by effort or FTE, and 70% expect that to decrease within two years. The same share, 65%, expect less fixed-fee and milestone-based pricing. Outcome-linked and consumption-based pricing are expected to grow (HFS). So far it is a small part of the market. BCG’s chief executive Christoph Schweizer told the Wall Street Journal in May that three quarters of BCG’s largest AI cases used variable fees, but that across all its work the share was “significantly less than a third” (WSJ). Infosys’s chief executive Salil Parekh said in July: “It’s not that we are shifting massively to it” (Infosys). Moving from time and materials to fixed price will not be enough.
Two capabilities will keep their value: delivering and pricing results the client can verify, and deep specialist knowledge, of the client’s processes or of a field such as data science. Changing how a firm invoices gives it neither. A firm that sells developers by the day should build both without giving up its current business, and set the pace by what clients buy.
Demand for IT services can still grow. ISG recorded a 21% year-on-year rise in the annual contract value of EMEA managed-services awards in the second quarter, to $4.7bn; its index covers contracts worth at least $5m a year. Europe moved differently from the US: in IT outsourcing, contract value rose 21% to $3.2bn in EMEA and fell 23% to $3.6bn in the Americas. For 2026 ISG expects worldwide managed-services revenue to grow 2.1%, against 30% for services sold as subscriptions (ISG EMEA, ISG Americas).
A client that wants its share of the savings at renewal needs a baseline: what the same work cost before AI. Without one, it cannot check how much the supplier saved.
Gartner predicts that by 2028, 70% of enterprises will abandon agentic AI built by vendors’ forward-deployed engineers because running it costs too much and ties them too closely to the vendor. It recommends agreeing knowledge transfer and an exit plan before delivery starts (Gartner).
Briefing
Reuters reported that Anthropic’s confidential IPO prospectus shows nearly $4.6bn in 2025 revenue, twelve times the year before, an operating loss above $8bn and at least $518bn of infrastructure commitments over about a decade. Around 80% of those commitments are non-cancellable or payable regardless of use, while many large customers lack long-term contracts (Reuters, Reuters). Reuters said the listing could value Anthropic at more than $2tn, about 435 times its 2025 revenue.
Those spending obligations give Anthropic an incentive to secure longer customer commitments. If a buyer is offered a multi-year discount, the price concession needs to compensate for the flexibility it gives up.
OpenAI has shelved the planned October release of GPT-6.1 Astra after safety testing. Head of safety systems Saachi Jain said it “didn’t quite meet the bar”; WIRED reported concerns about staying within authorisation and accurately communicating completed work (WIRED). Any deployment awaiting that version now needs a revised schedule or a model that has already passed the buyer’s own evaluation.
On 30 September, Reuters reported that the Federal Trade Commission is investigating Anthropic, OpenAI and other AI labs over consumer risks. The agency plans compulsory information requests and testimony from executives and the research group METR (Reuters). The investigation uses existing powers over unfair and deceptive practices, so the absence of an AI-specific law does not prevent scrutiny of agent behaviour.
Summary
Accenture is giving clients AI productivity and making up for it with larger scope. Clients across the industry are asking for discounts, and 77% of large companies in the HFS survey want to renegotiate prices. In my view, firms that sell developers by the day will feel it first, once large enterprises start building software with AI agents.
Stay balanced, Krzysztof
Krzysztof Goworek is founder of Quintant — AI advisory that gets enterprises from experiment to production value.