A conversation with Gemini made me realise the most obvious trade of the AI era was hiding in plain sight — and it's not Nvidia.

The AI Short I Thought of Too Late (And Why It Still Makes Sense)

I was in a deep rabbit hole with Gemini yesterday — running through business model analysis, asking it to stress-test ideas — and somewhere in that conversation it hit me: the most obvious short of the AI era wasn't a flashy tech bet. It was the companies whose entire business model is selling human hours.

TL;DR: AI is structurally threatening the labour-arbitrage model that HR leasing and IT consulting giants like Randstad, Capgemini, and Accenture are built on. The thesis is simple — and it's already showing up in their numbers. Capgemini in particular still looks technically and fundamentally weak. This is not financial advice. Do your own research.

Why This Matters — The Business Model Is the Problem

HR leasing and IT consulting firms have one core product: humans. Specifically, the ability to source, contract, and bill humans at scale. Randstad places temporary workers. Capgemini and Accenture sell consulting days — armies of developers, analysts, and project managers deployed against enterprise problems.

The margin in that model comes from the spread: what the client pays per day vs. what the worker costs. Volume and headcount are the engine. It's not a bad business — these are large, durable companies — but it is a business that depends on humans being the cheapest way to get cognitive work done.

AI is attacking that assumption directly. Not all at once. But structurally, and faster than their earnings calls are admitting.

Having spent years in operations-heavy environments — the kind where you're constantly pressure-testing headcount decisions against output — I can tell you that the conversation is already happening in enterprise boardrooms. The question is no longer "should we use AI tools" but "how many fewer external consultants do we need if we do."

How Is AI Affecting IT Consulting Companies?

The impact isn't coming from robots replacing consultants overnight. It's more insidious than that.

First, there's productivity compression. GitHub Copilot, Cursor, and similar tools are making individual developers meaningfully faster — some studies suggest 30–55% productivity gains on certain coding tasks. If your client used to need a 20-person development team from Capgemini, maybe now they need 13. That's a billing day problem.

Second, there's commoditisation of junior work. A huge chunk of consulting revenue comes from billing out analysts and junior devs who are essentially doing structured, repeatable cognitive tasks — documentation, code review, QA, report generation, data wrangling. These are exactly the tasks LLMs are eating. The high-margin senior architects are probably fine. The pyramid below them is not.

Third, there's enterprise AI adoption timing. Large companies are slow — but they move. And when they move on AI tooling internally, the first thing they cut is external headcount doing automatable work. That's a direct hit to Capgemini's and Accenture's top line.

Accenture themselves have tried to get ahead of this by pivoting their narrative to "AI transformation partner" — essentially telling clients: don't fire us, hire us to implement the AI that replaces the people we used to sell you. It's a smart pivot. Whether it's enough is a different question.

Is Capgemini a Good Short? Fundamentals and Technicals

This is where I'll be direct: I think Capgemini is an interesting short thesis. I am not telling you to short it. This is my personal read, not financial advice — please do your own research before touching any position.

Fundamentally:

Capgemini trades at a P/E that, for most of its recent history, priced in steady mid-single-digit revenue growth. That growth story is getting harder to tell. In 2024, their organic growth turned negative in several quarters — something that hadn't happened in years. Bookings growth has slowed. The client conversations about AI are translating into delayed hiring decisions on Capgemini's end, which shows up as revenue softness.

Their response — repositioning as an AI services company — is logical but slow. Margins on AI transformation projects are lower in the early years (you're selling strategy and tooling, not headcount at scale), and the retraining cycle for their workforce is long and expensive. The market hasn't fully repriced this transition risk yet.

Revenue mix is also worth noting: a significant portion of Capgemini's work is in financial services, manufacturing, and public sector — all industries where AI adoption is accelerating but procurement cycles are slow. That lag means the impact is still coming, not fully arrived.

Technically:

Capgemini's stock has underperformed the broader European tech index meaningfully since early 2023. The chart has shown a pattern of lower highs, and multiple sell-side downgrades have shifted consensus from buy to hold. Volume on down days has been heavier than on recovery bounces — not a clean setup, but a telling one.

Support levels from the 2020–2021 re-rating have largely been retested. If those break, the next floor is considerably lower. Again — not a prediction, just what the chart is showing.

The counter-argument: Capgemini is large, diversified, and has navigated tech cycles before. A broader market rally or an AI hype second wave could lift the whole sector regardless of fundamentals. Shorts in slow-moving structural stories can bleed you before they pay off.

What Most People Get Wrong About Shorting AI Disruption

People think AI disruption is binary. Company exists, then AI arrives, then company dies. That's not how it works — and it's why straightforward "AI will kill X" shorts often disappoint.

The real timeline is messier. These companies will adapt partially. They'll pivot their language. They'll hire AI PMs and call themselves transformation firms. Accenture is already doing this aggressively. Some will survive in a smaller, different form. The disruption shows up in margin compression and slower growth before it shows up in existential crisis — and slow-burn compression is hard to short profitably unless you're sizing and timing well.

The more interesting version of this thesis isn't "these companies go to zero" — it's "these companies are priced for a growth trajectory that no longer exists, and the repricing takes two to four years."

That's a different, more patient trade. Which, for most retail investors, is also a harder trade to hold.

I wrote more about patience as a core investing edge in my piece on value investing frameworks — the same logic applies here.

The best trades are usually obvious in retrospect — the trick is figuring out whether 'too late' means you missed it or you're just early to the second chapter.

What to Actually Do

Map the business model before the ticker. Ask: does this company sell human hours or outcomes? If hours, ask what happens when hours get cheaper. That's your starting filter.
Watch bookings, not just revenue. For consulting firms, bookings growth is a leading indicator. Revenue is lagging. If bookings are already soft, the revenue hit is coming 6–12 months later.
Read the earnings calls, not the headlines. Capgemini and Accenture's investor communications are carefully worded — but the language around AI is telling. "We are repositioning" usually means "our existing model is under pressure."
Size for a slow story. Structural shorts in large, established companies can take years to play out. If you're going to express this thesis, size it like a long-term view, not a swing trade — and account for the possibility that a bull market lifts all boats temporarily.
Look at Randstad separately. The HR leasing model is arguably more exposed than IT consulting in some segments — temp workers in logistics, admin, and basic tech roles are highly automatable. Worth its own analysis.
None of this is financial advice. I'm thinking out loud. These are the questions I'm asking myself. Do your own research, talk to someone qualified, and don't size anything off a blog post — including this one.

I thought of this trade a year too late. But the underlying logic hasn't expired — it's just moved from 'early thesis' to 'mid-cycle confirmation.' That's still worth paying attention to.