The delivery timeline is the only metric that matters, until something ships wrong
There is one thing every data center operator, hyperscaler, and colocation provider agrees on: speed wins.
Billions of dollars in infrastructure investment are underwritten on 2027 and 2028 revenue. A facility that delivers capacity on time captures a tenant. A facility that delivers a month late may not. Hyperscalers will shop around until they find available power, cooling, and compute, and they will move to the next supplier without hesitation.
But speed alone isn’t enough. Hyperscalers impose stringent quality requirements on every component in the chain, including documented traceability for every assembly step, first-pass yield targets, factory acceptance testing, and zero tolerance for defects that make it to site. A cooling module that arrives on time but fails commissioning testing is worse than one that arrives late. A switchgear panel that ships with an assembly error delays the entire site energization sequence and puts the supplier’s next order at risk.
The result is a dual mandate that cascades down through the supply chain: deliver faster than ever before, with quality standards that have never been higher. And no part of the supply chain feels this vise more than the OEMs manufacturing the complex sub-systems; specifically cooling and thermal management, power generation, and power distribution; that make a data center work.
How the data center supply chain operates
A modern data center is assembled from hundreds of manufactured sub-systems, each flowing through a multi-tier supply chain before reaching a site.
At the top of the chain sit the hyperscalers and colocation operators. They are the customers. They set the specifications, the timelines, and the quality requirements that cascade down through every tier below.
Below them are the OEMs. They are the companies that manufacture the complex sub-systems a data center depends on. Cooling and thermal management OEMs build coolant distribution units, liquid cooling manifolds, precision air handlers, chiller packages, and heat exchangers. Power generation OEMs build generator sets and turbine packages. Power distribution OEMs build switchgear panels, PDUs, UPS systems, busway, and transformers. These are complex mechanical assemblies with dozens or hundreds of components, real sequencing requirements, and tight quality tolerances. This is where the highest-value, most complex manufacturing in the data center supply chain happens.
Feeding the OEMs are their Tier 1 suppliers. These are companies providing major sub-assemblies and components. Below them, Tier 2 and Tier 3 suppliers provide progressively more commodity parts, fabricated components, and raw materials: sheet metal, structural frames, busbars, piping, electronic controls, fasteners.
In some cases, OEM products are further assembled into larger deployable modules; including power skids, cooling pods, and complete modular data centers; by system integrators who combine products from multiple OEMs into shippable units. These modules then go to the site, where general contractors handle final connections and commissioning.
The OEM tier is the critical manufacturing layer. These companies are building the most complex, customized, engineering-intensive products in the chain at volume, under compressed timelines, to specifications that can change mid-project.
Why delivery timelines are compressing & and why OEMs are caught in the middle
Four forces are converging to compress delivery timelines across the data center supply chain.
First, the shift from site-built to modular construction has reset expectations. Highly modularized projects are achieving schedule reductions of 30 to 50 percent compared to conventional builds. Delivery timelines that once ranged from 24 to 36 months are now commonly 16 to 20. The most aggressive programs are targeting 12 months. Modular manufacturing is supposed to parallelize work — build the cooling system in one factory, the power system in another, the enclosure in a third, ship everything to site and assemble. But that parallelization only holds if every supplier hits its delivery window. A cooling module that arrives two weeks late delays the entire site assembly sequence.
Second, the technical complexity of what OEMs have to build is increasing across the board. In thermal management, GPU-dense racks pushing past 100 kW are driving a rapid transition from air to liquid cooling: new product lines, new assembly processes, new engineering challenges all ramping at once. In power distribution, the shift to higher-density configurations and modular power architectures is increasing the complexity of switchgear and PDU assemblies. In power generation, the proliferation of on-site generation means OEMs are fielding more product variants for more use cases simultaneously. Every category is getting harder to build at the same time that delivery timelines are getting shorter.
Third, the volume is simply unprecedented. Hundreds of data centers are under construction or planned globally. OEMs across cooling, power generation, and power distribution that were building tens of units per month are being asked to build hundreds. Production floors are scaling headcount, adding shifts, and standing up new lines while simultaneously trying to maintain quality and hit delivery dates that get shorter every quarter.
Fourth, upstream supply chain disruptions compound the problem. Lead times for critical components have stretched from weeks to months. An OEM can have a perfectly efficient production floor and still miss a delivery because a Tier 2 supplier couldn’t deliver a component on time. When that component finally arrives, the pressure to make up lost time falls entirely on the OEM’s assembly operation, which now has to build faster than planned, with less schedule buffer, while maintaining the same quality standards. Supply chain volatility compresses the window the production floor has to execute.
The result is an execution environment where OEMs are squeezed from every direction: shorter timelines from above, supply disruptions from below, increasing complexity in the product, quality requirements that never relax, and a production floor that has to absorb all of it. And the things that cause delivery timelines to slip on an OEM’s production floor are not the things most people talk about.
What actually causes a delivery to slip
When an OEM misses a delivery date, the root cause is the accumulation of small frictions on the production floor. Each one minor in isolation, collectively adding up to days and weeks of delay. Two of these frictions dominate.
Design changes that don’t reach the floor fast enough.
The products these OEMs build are overwhelmingly engineer-to-order. In cooling, roughly 80 percent of products involve custom design. In power distribution, switchgear and PDU configurations vary by site, by customer, by phase of deployment. Even in power generation, mounting configurations, enclosure specs, and control integration differ from order to order. A hyperscaler specifies an architecture, then changes requirements mid-project. These changes happen constantly, and they may happen mid-build.
The engineering team updates the CAD model. But how long does it take for that change to reach the operator assembling the unit on the floor? At most OEMs, the answer is days or sometimes longer. The change has to be translated into updated work instructions, which means a manufacturing engineer has to manually revise a document, get it reviewed, get it distributed. If the instructions are paper or PDF, they have to be reprinted and physically delivered to the station. Even companies using digital authoring tools often find themselves manually rebuilding instruction sets from updated models. This is a process that can take hours per assembly and still introduces the risk of human error.
Meanwhile, the floor keeps building to the old spec. The rework that follows costs materials, labor, time, and it creates quality escapes. An operator who built three units to an outdated spec before the revision arrived has produced three units that either need to be reworked or risk failing acceptance testing at the customer’s site. In a supply chain where a month’s delay means losing the customer and a shipped defect means losing the next order, neither time nor quality can be sacrificed.
New operators who can’t build at full speed on day one.
The data center supply chain is short hundreds of thousands of workers. But the labor challenge on an OEM’s production floor is different from the headline number. It’s not that the people don’t exist, it’s that the people who are there are new.
When an OEM scales from 50 operators to 150 to meet demand, the 100 new hires don’t arrive knowing how to assemble a CDU, a switchgear, or a generator set. The traditional approach is to pair each new hire with a senior technician for weeks of shadowing. The new operator watches, asks questions, gradually takes over tasks, and eventually builds independently.
This works when you’re hiring five people a year, but not when you’re hiring 50 in a quarter. There aren’t enough senior technicians to go around. The senior techs who are shadowing new hires aren’t building product. And every week a new operator spends in training is a week they’re not contributing to throughput, which means a week the delivery timeline absorbs. Worse, undertrained operators are a quality risk. An operator who learned by watching rather than following documented, verified instructions is more likely to make assembly errors that may not surface until commissioning testing, when they’re far more expensive to fix.
The labor shortage highlights the time it takes to turn a new hire into a productive, reliable operator. Every day of that ramp time subtracts directly from the delivery window and adds risk to the quality gate.
The gap on the production floor
Most data center OEMs are sophisticated engineering companies. They run product lifecycle management systems to manage designs. They use 3D CAD to model their products. Many have ERP systems tracking materials and schedules. What they don’t have is the layer between the CAD model and the operator’s hands.
Today, most OEMs bridge this gap with one of three approaches. The first is paper or PDF work instructions (manually authored, manually updated, manually distributed). The second is digital tools that let engineers create visual instructions from CAD models, which are better than paper, but still manual to update, and they don’t scale when designs change constantly across dozens of product variants. The third is tribal knowledge: the experienced operator knows how to build it, and the new operator learns by watching.
All three approaches share the same fundamental problem: they can’t keep up with the pace of change. And all three are quality risks as much as speed risks. Paper instructions can be outdated. Manually rebuilt digital instructions can contain errors. Tribal knowledge is unverified, undocumented, and impossible to audit.
The result is a production floor that’s always slightly behind engineering, building to specs that may be outdated, training operators on processes that may be incorrect. In a supply chain where delivery timelines are measured in weeks and quality failures are measured in lost customers, “slightly behind” compounds into missed dates and shipped defects.
What closing the gap looks like
The missing layer is software that sits between the CAD model and the operator, generating and maintaining work instructions automatically. This is what Dirac’s BuildOS does.
BuildOS takes a 3D CAD model — the same model the engineering team is already working in — and generates interactive, step-by-step 3D work instructions for complex mechanical assembly. When the model changes, the instructions automatically update with it. No manual rework, no redlining, no revision emails. Every instruction is traceable back to the current design revision. This gives quality teams the documentation and auditability that hyperscalers require.
For an OEM assembling cooling systems, switchgear, or generator sets, this changes the math on every delivery-timeline and quality problem:
When a hyperscaler changes a spec mid-build, the updated instructions reach the operator in minutes instead of days. The floor is always building to the current design. Rework caused by outdated instructions drops toward zero. The engineering change doesn’t adds minutes rather than days to the delivery timeline. Additionally, because the instructions are generated from the verified model, the quality risk of a manual transcription error disappears.
When 50 new operators arrive in a quarter, they don’t need weeks of shadowing. They follow model-based instructions at the station: visual, step-by-step, specific to the product variant they’re building. A new hire can build correctly on their first shift. Ramp time compresses from weeks to days. And because the operator is following verified, current instructions rather than interpreting tribal knowledge, first-pass yield improves at the same time throughput increases.
When a supply chain disruption compresses the build window, a floor running on automated instructions has the buffer to absorb it. There’s no time lost to manually updating documentation for the rescheduled build sequence. There’s no quality risk from rushing operators through an unfamiliar assembly. The production floor can flex without breaking.
The net effect is the difference between a production floor that can absorb change, scale headcount, maintain quality, and compress timelines without slowing down, and one that can’t. In a supply chain where the hyperscaler moves on if you’re a month late and doesn’t come back if you ship a defect, that difference determines who gets the next order.
The window
The data center supply chain is in a rare moment. Demand for manufactured sub-systems is higher than it has ever been. Product complexity is increasing across every category. Quality requirements are tightening. Delivery timelines are compressing while supply chains remain volatile and the workforce is new and inexperienced.
The OEMs that build the operational infrastructure to manufacture at speed and quality— not just the factories, but the production systems inside them — will be the ones that hyperscalers call first. The ones that don’t will watch their customers move on.
