The Apollo Sandwich

IN 1959, Pillsbury was handed one of the strangest briefs in food manufacturing. NASA wanted food for astronauts, and its first requirement had nothing to do with taste. The food must not crumble: in zero gravity a floating crumb could drift into an instrument panel and short the electronics.

The second requirement was harder. With no medical help within a quarter of a million miles, the food had to be guaranteed as safe. Industry practice was to make a batch and test samples from it, which proved only that the samples were safe, and proved it after the batch already existed.

So Pillsbury and NASA worked from the other end, finding every point where a hazard could enter, controlling it there, monitoring it, and recording the evidence. The method became Hazard Analysis and Critical Control Points (HACCP) and it now underpins food safety worldwide. Even the sandwich you buy from the petrol station is safer for the race to the Moon.

A customer who could not tolerate failure paid to invent assurance. And everyone else inherited it.

Something similar is happening now, in a sector that did not exist when Apollo flew.

The customer who cannot tolerate failure

Globally, $492 billion of data-centre projects are under construction. In an AI facility, mechanical, electrical and control systems account for around three quarters of the building cost, with electrical systems alone about half. A substantial share of delivery risk therefore sits in complex, interdependent systems.

Redundancy creates a particular quality assurance challenge. A defect in one path can remain latent while the other carries the load. It may only emerge during integrated systems testing, or when a failure requires the standby path to take over. And by then much of the work may be concealed; once the hall is operational, access and rework become more difficult.

Critical facilities carry a different profile. The standards are exacting and compliance has to be demonstrated rather than assumed. A defect found in service is a failure against those standards, not an item on a snagging list. AstraZeneca claimed around £32m for remediating defects in dampers, ductwork and insulation at its Cambridge Discovery Centre, found after installation.

Cases like these reinforce that prevention is cheaper than cure. With consequential loss at stake, it suits both the client and contractor alike to buy an auditable record of what was installed, taken as it was installed.

The new norm

XYZ Reality offers one answer, putting engineering-grade augmented reality on site so both the client and contractors can compare installed work against the model before anything is concealed. Its turnover has more than doubled in two years as it has turned towards data centres, with projects as large as a 400 megawatt campus in the US.

Cumulus Digital Systems captures verification data directly from connected tools as work is done. In one study the company cites, trade partners on a hyperscaler project took between 25 and 75 days to submit their quality paperwork without it, and under a day with it.

These are different businesses with different products doing different jobs, but together they assemble an evidence layer between the design and what was actually built. The commitment to that service is shifting too. Earlier this month, Buildots signed an enterprise agreement with Norweigan data centre operator Green Mountain. What was once technology under trial is becoming an embedded way of working.

Whilst many of the projects these firms are involved in are overseas, a quiet revolution is under way closer to home. With west London and Slough forming Europe’s largest data centre cluster, similar principles are being taken in the UK, largely behind the veil of non-disclosure agreements. What starts as one client’s requirement tends to become the measure everyone else is judged against. It is a how a NASA food-safety method ended up governing your lunch. For the wider construction sector, a new approach to quality assurance is taking shape out of view.

Why we have never bought this before

We have known for a long time that independent inspection reduces defects. Yet over the last thirty years the clerk of works has quietly lost prominence. The role did not disappear because anyone disproved its value. As responsibility for checking passed increasingly to contractors, independent inspection became a visible cost with a return that was difficult to see. Grenfell put it back on the agenda, though largely in relation to fire safety and higher-risk residential buildings.

The Get It Right Initiative has regularly reminded us of the consequence. The cost of error that gets recorded is the smaller part; most of it hides in rework, delay and defects that surface long after handover. Everyone pays, and yet nobody sees the whole bill.

Digital assurance offers another way to provide and document scrutiny, but it runs into the same commercial problem. Cost plans in mature sectors are built around tight benchmarks, and there is rarely an established allowance for a new assurance line. Paying to check work that a contractor has already been paid to deliver can look like duplication. There is an irony in that objection, when independent scrutiny of cost is itself considered essential.

The easiest saving to put in a cost plan is a check that digital assurance replaces. But duplication is thinnest where budgets are tightest: the less assurance a project already carries, the less there is to take out.

Data centres escape that trap. Their scale and the consequences of failure give quality assurance a different place in the value calculation: it becomes a necessity rather than a discretionary addition. So the sector with the least need for a benchmark to justify buying digital assurance may be the one that finally creates one.

Three engines

That benchmark will not stay in the data hall. Risk assessments, method statements and near-miss reporting all began in high-risk work and are now unremarkable everywhere, and they did not spread by gravity. If digital assurance does travel, it will travel three ways.

1. Contractors carry the technology. A firm that has trained its people and integrated the systems across a data centre programme can deploy them across its wider portfolio, underpinned by a business case that is familiar. That route reaches wherever complexity and consequence already justify the spend.
2. Clients carry the standard. Data centre clients set the bar and often buy the evidence themselves rather than take the work on trust. A client that specifies what must be verified, and when, turns assurance from a contractor’s optional tool into a condition of delivery.
3. Elsewhere the regulator does the persuading. For higher-risk buildings, the golden thread and the gateway regime are changing what counts as acceptable evidence, and that is already visible. OpenSpace has a three-year agreement with L&Q, one of the largest housing associations in the country, covering its active development portfolio. The driver was partly the evidence obligations in the Building Safety Act. Housing did not arrive there because a hyperscaler proved the concept. It arrived because a regulator changed the value of evidence, and a client acted on it.

Public clients need not wait to be told. They can specify verifiable outcomes rather than thicker document bundles, and make assurance a normal line in the cost plan rather than an exception somebody has to argue for.

The building that waits longest

Which leaves a gap worth naming. A school has modest engineering, no capacity promise to break, tight historical cost benchmarks and, for now, no regulatory driver pulling digital evidence into its procurement. That is three locks and no obvious key.

HACCP did not spread because every food producer acquired NASA’s budget. It spread as critical checks became part of the production process and, over time, part of regulation. The construction equivalent is not a larger archive at handover, it is the ability to find a failure while it can still be corrected, and to show that the correction was made.

Data centres may help establish that capability. The real test is whether it can be made proportionate and routine on projects without a hyperscaler to pay for it. The buildings we worry about least are the ones that will wait longest. Which is fine of course …. until it isn’t.