Computacenter’s record half: the AI money is in America, not the UK

Computacenter reported a record first half on 8 September. Revenue rose 71.6% to £6,845.2m, adjusted profit before tax rose 87.0% to £152.4m and the company, promoted to the FTSE 100 in June, now expects full-year adjusted profit before tax of no less than £380m, according to its regulatory news service (RNS) announcement. Mike Norris, chief executive, said the half was “significantly ahead of our expectations at the start of the year”.

Read on and the record is an American one. North America produced £118.5m of adjusted operating profit, up 141.3%, and now accounts for 62% of group adjusted operating profit before central costs, against 44% a year earlier. Norris credited “our growth with hyperscale, neocloud and enterprise customers”. The North American product order backlog stood at £6,936.3m at 30 June. The UK’s stood at £1,382.3m.

Then read the UK line slowly. UK revenue more than doubled, up 136.4% to £1,512.7m. UK adjusted operating profit was £26.4m, up 52.6% from £17.3m. That is about 1.7p of operating profit for every pound of revenue, down from 2.7p a year earlier. The RNS says why: gross margin “decreased by 846 basis points, driven by the increase in high-volume, lower-margin Technology Sourcing business”.

So here is the argument. The AI capex boom is real and the UK’s largest listed reseller is riding it well. But it is a hyperscaler story dressed as a channel story. The profit is made in America, shipping racks to companies that build data centers, and very little of it is trickling down to the UK mid-market. UK partners should stop reading Computacenter’s headline and start reading its margin line, because the margin line is the part that applies to them.

Trace where the UK growth came from

Even the UK segment is not a UK mid-market story. UK technology sourcing gross invoiced income (GII) rose 90.8%, which the RNS puts down to a higher proportion of AI data center products, and its gross margin fell 813 basis points as a result. The RNS then says where some of that work was done: “During the period, we completed large data center projects in the Nordics for leading European AI infrastructure companies and continued to deliver a large project in Eurasia.” The target market Computacenter names for its UK business is large corporate and public sector organizations.

The annuity lines tell the other half. UK managed services revenue grew 2.3% to £140.6m, and UK services gross margin fell 706 basis points, which the RNS attributes to “costs to improve efficiency and an increase in contract provisions”. Professional services, up 13.0% to £103.5m, was the healthiest UK line in the statement, and it is a people business rather than a hardware one.

Compare that with a reseller built on the mid-market. Softcat’s half to 31 January turned £2,008.6m of GII into £93.8m of underlying operating profit, according to its RNS of 18 March, with net new customers “led by our mid-market segment”. Computacenter’s UK arm turned £2,294.2m of GII into £26.4m. The periods and definitions differ, but a gap of more than three times is not an accounting quirk. Selling AI infrastructure at volume earns less per pound than selling licenses, security and support to mid-sized companies, and the second business is the one most UK partners are in.

Follow the money, if you can

Computacenter’s own answer is to go where the demand is. In January it completed the purchase of AgreeYa, a professional services firm in Folsom, California, for $120m (£89m). In June it bought Government Acquisitions Inc (GAI), a Cincinnati value-added reseller to the US federal government with about 90 staff and 2025 GII of about $390m (£289m), for up to $92m (£68m). Its capex guidance of £70–75m includes a new integration center in Atlanta. Group inventory stood at £1,260.4m at 30 June, against £316.8m a year earlier. That is what following the money looks like: a balance sheet most UK partners do not have, spent in a market most UK partners do not serve.

The first choice, then, is to follow it anyway: a US acquisition, a joint venture or a subcontract to one of the integrators the hyperscalers already use. It is open to perhaps a dozen UK partners with the capital and the appetite. For everyone else it is a way to lose money in a bigger currency.

Specialize where the hyperscalers will not go

The second choice is to go where Computacenter is not. Hyperscalers and neoclouds buy from a handful of integrators that can design, deploy and cable racks at scale, which is the capability the RNS describes in North America. A 400-seat manufacturer in the Midlands will not build an AI factory; it will rent inference by the token, buy Copilot seats and need someone to sort out its data, its identity and its security before any of that works. That is a services margin, and the RNS shows UK professional services growing 13.0% while UK hardware margin fell. The public sector, where Computacenter reports significant transformation work for a large customer, is the one UK market big enough to buy AI infrastructure directly, and even there the buyer is a large organization rather than a mid-sized one.

Or tell the board the truth

The third choice is the hardest to say out loud. Accept that a UK mid-market partner will grow at the mid-market’s pace, not at 71.6%, and say so before the board reads the headlines and asks why. The honest answer is that Computacenter’s growth is 1.7p-in-the-pound hardware shipped to people building data centers, and your customers are not building any. Softcat’s 27.3% growth in underlying operating profit, earned in the mid-market, is the more useful benchmark, and a demanding one.

Whichever you choose, put the margin line next to the revenue line in every board pack from now on. Chase Computacenter’s revenue line and you will end up with its UK margin and none of its American profit.