Last week Capgemini formally notified Spanish labor representatives of a collective dismissal process targeting its local workforce. 11,000 employees. Official reason: artificial intelligence and an oversized bench. The first formal meeting with unions is set for April 23.

Only months earlier, the French parent announced 2,400 layoffs in France, 7% of headcount, the largest cut in the company's history. And all of it right after paying €1.5 billion out to shareholders.

The easy read is the one newspapers are already running: "AI is destroying jobs." The hard read, the one that matters if you run a business that buys IT services, is different. Capgemini's layoff round is not a workforce adjustment. It is the first visible crack in a 25-year-old business model built on a single hypothesis.

That hypothesis just broke.

The story behind the story

Capgemini is not laying off people because AI is taking work away from its clients. It is laying off people because AI is taking work away from Capgemini.

The number everyone is repeating is the "bench": staff on payroll who aren't billed to a project. Translated out of corporate euphemism: people Capgemini bills when they're sitting at a client and keeps on balance sheet when they're not. The bench was the lung that let the offshore model absorb demand spikes and rotate junior consultants across accounts.

That lung has been swelling for months. The reason is simple: projects that used to take 30 people for 9 months now take 8 people for 3. Same clients. Same needs. Fewer billable bodies.

And then there's the other uncomfortable data point: €1.5 billion returned to shareholders weeks before the collective dismissal. That is not a footnote. That is the headline. Capgemini isn't in a cash crisis. It is restructuring a machine that no longer turns hours into margin the way it used to.

25 years of arbitrage: how the offshore empire was built

It helps to remember what this business was actually built on. Understanding the origin of the model makes it easier to see why it is breaking now.

The big IT integrators — Capgemini, Accenture, TCS, Infosys, Wipro, Cognizant, HCL — built their 2000s playbook on a single idea: labor arbitrage. Bill the client in New York, London or Paris at first-world rates. Execute in Bangalore, Pune or Chennai at emerging-market rates. Keep the spread.

A more polished variant showed up later: nearshore. Bill in Paris at local rates. Execute from Poland, Portugal or a tier-two Spanish city at 40-50% lower cost than the capital. Same arbitrage, better optics for the European buyer.

On top of that arbitrage sat the pyramid: one partner who signs, a handful of managers who coordinate, and a very wide base of junior consultants, developers and analysts who execute. The margin doesn't come from the senior people. It comes from the base, selling junior-hours at manager-hour prices.

For two decades, that model was unbeatable. You had an IT problem? You bought a pyramid. 50 people, 18 months, closed engagement. Nothing else competed.

And then everything changed.

Hit from the base: why AI collapses the pyramid

AI doesn't attack strategy consulting. It attacks execution. And execution is exactly where the base of the pyramid lives.

Think about what a junior consultant actually does on a typical integration project. Writes boilerplate code. Migrates data between systems. Documents APIs. Produces tests. Runs manual QA. Drafts user manuals. Builds status decks. All of it is already being done by Claude, Copilot or Cursor at a fraction of the time and near-zero marginal cost.

This is not a forecast. This is happening right now, in 2026, on billed engagements. A senior team with AI tooling is delivering the work a 30-person pyramid used to deliver. And clients are starting to see it on the invoice.

The old geographic arbitrage was monetized on the gap between a consultant-hour in Paris and a consultant-hour in Bangalore. AI has opened a new, far more aggressive arbitrage: consultant-hour versus token. In that comparison, no geography competes.

"Offshore didn't compete with onshore on quality. It competed on cost. AI doesn't compete on quality either. It competes on cost. Only this time offshore is on the expensive side of the equation."

What Capgemini is discovering — and what every other integrator will discover — is that AI doesn't replace their consultants: it replaces the business model that made those consultants profitable. Those are two very different things, and the second one is considerably worse.

The "bench" euphemism: what that number actually means

Back to the bench. Capgemini says the problem is that too many people are "unassigned." Let's translate.

The bench in an IT services firm is not an accident. It is a strategic decision. People are kept on balance because the cost of having them waiting is lower than the cost of missing an opportunity when an urgent project lands. It is a disposal insurance, paid in salaries.

For that insurance to make sense, projects need to keep landing at the pace they used to. And team sizes need to keep looking like they used to.

Neither is true anymore.

The RFPs landing today at the big integrators have two new characteristics. First, proposed team sizes are smaller because the client already knows — or suspects — that with AI the same job can be done with fewer people. Second, timelines are shorter because the client is benchmarking against vendors who already ship with AI. Combined effect: fewer billable hours per project, and probably fewer projects.

Capgemini's bench isn't bloated because they over-hired. It is bloated because structural demand for billable hours has contracted. That contraction is not reversing. It is accelerating.

What's coming: the reshaping of the IT consulting map

Capgemini won't be the only one. It is simply the first in Europe to move a cut big enough to make the front page. The others will follow.

What we are watching is the start of a reconfiguration along five axes:

1. The collapse of the pyramid's base. Junior consultants doing repetitive work — 60-70% of an integrator's headcount — stop being necessary. The pyramid flattens. Senior-to-junior ratios, which have sat at 1:6 or 1:8 for 20 years, move toward 1:2 or 1:3.

2. The disappearance of pure geographic arbitrage. If a model writes the code, it no longer matters whether the developer sits in Lisbon or Hyderabad. The relevant cost center stops being the developer's salary and becomes the technical lead's ability to orchestrate AI. That calls into question entire nearshore centers that were built over the last decade.

3. The pricing shift. The "time & materials" model — billing per consultant-hour — is dying. Clients don't want to pay for hours they suspect are actually tokens. The future is outcome-based pricing: pay for results, not effort. That is devastating for a P&L built on billable hours.

4. The return of senior consulting. What survives — and thrives — is the work AI cannot do: strategic diagnosis, architecture calls, data governance, business design. Four-to-five senior people work, not forty. Our advisory model was built on exactly that thesis.

5. A new kind of competitor. Boutiques that carry no pyramid now compete with the giants on equal footing. For 20 years, boutiques lost every big deal because they had no "capacity." In 2026 capacity is no longer given by headcount. It is given by the AI stack plus senior talent. That changes the market.

Capgemini knows all of this. That's why they are cutting. But cutting a pyramid without redesigning the business model doesn't save you: it just buys you one more year of decent margin while the structural problem keeps advancing.

What a CEO buying IT services should do now

If you are on the buyer side — not the seller side — this is the moment to rethink how you contract technology. What worked three years ago is expensive today and will be absurd soon.

1. Audit your outsourcing contracts as if they were debt. Any fixed-price, long-duration contract signed before 2024 is probably misaligned with what a team using AI can deliver today. Renegotiate or break clauses. The market moved under your feet.

2. Stop buying bodies. Start buying outcomes. If a proposal lands with "X FTEs for Y months," send it back. The right unit of purchase in 2026 is the outcome: what deliverable, by when, at what quality. How the vendor gets there — 40 people or 4 plus AI — is not your problem.

3. Ask where AI sits inside your vendor. If a large integrator cannot explain in 10 minutes how AI is embedded in your projects, assume it isn't. And assume they are billing you hours that should be tokens.

4. Don't confuse low price with value. The cheap nearshore team at $40/hour still doesn't compete with a senior team plus AI at $160/hour that ships three times the work. The question is not "what does the hour cost." It is "what does the outcome cost."

5. Shrink your vendor list and move it upmarket. Ten mid-tier vendors executing the same transactional layer is last decade's architecture. Fewer vendors, more senior, more AI, more outcome. That is the cost architecture of the next five years.

6. Build internal AI capability, even a small one. You need someone inside who can tell when a technical proposal actually leverages AI and when it is pastiche billed at pyramid prices. Without that, you will keep paying arbitrage to vendors who no longer have any.

The Capgemini story reads as a layoff story. It is something larger than that. It is the first public acknowledgement that a 25-year-old business model — arbitraging bodies across geographies — just ran out of arbitrage. AI didn't eat the consultants one by one. It ate the whole pyramid from the base up.

Anyone still buying services as if nothing had happened is funding a structure that no longer exists. And paying for it, on top of it, out of their own margin.