Marengo

Data center development has a dirty secret: the hardest part isn't construction, it's everything before the shovel hits the ground. Site due diligence, feasibility studies, concept design, and FEED/permitting documentation can consume 10 to 12 months of a project's critical path. That's a year of carrying costs, a year of market uncertainty, and a year of compounded interest on land that isn't producing a single megawatt.

Marengo, a Y Combinator-backed startup, is attacking that bottleneck directly. The company calls itself "the engineering firm reimagined for speed," and its pitch is disarmingly simple: it compresses the pre-construction design cycle to 5-6 months, roughly half the traditional timeline, and claims a 50% cost reduction on feasibility studies. The AI isn't the product—it's the internal engine that lets a small team of licensed engineers move at a pace a traditional firm can't match.

The 10-to-12-month bottleneck Marengo is attacking

Traditional pre-construction engineering is a sequential, iterative grind. A site is assessed, a feasibility study is produced, concepts are drawn, and then—slowly—the design is refined through multiple disciplines. Each step depends on the previous one, and each handoff introduces delays. The industry has accepted this as normal, but Marengo's comparison table frames it as a fixable inefficiency.

The numbers are stark: traditional firms explore 3-4 concepts, assess trade-offs manually, and surface risks progressively. Marengo claims to explore 1,000+ concepts in parallel, use multi-objective optimization, and surface risks instantaneously. The result isn't just a faster timeline—it's a different kind of engineering output, one that starts with a broader search space and then narrows down to a preferred solution with data to back it up.

What '1,000+ concepts in parallel' actually means

It's easy to dismiss "1,000+ concepts" as marketing math, but the claim points to a real shift in how Marengo approaches design. Instead of a human engineer sketching a handful of layouts, the company's in-house AI tools generate and evaluate thousands of permutations across power, cooling, civil, access, and site configuration. The software iterates through combinations of building placement, HV incomer routes, cooling plant configurations, and structural grids, scoring each against constraints like setback limits, utility access, and constructability.

This is not generative AI producing pretty pictures. It's computational design applied to engineering constraints. The output is a shortlist of viable concepts, which licensed engineers then review and refine. The human doesn't disappear—the human gets to focus on judgment instead of repetitive drafting. That's why Marengo can claim to cut the feasibility study from 12-16 weeks to 6-8 weeks: the machine does the grunt work of exploring the solution space, and the engineer does the critical work of validating the results.

The discipline-by-discipline test fit

One of Marengo's most distinctive deliverables is the integrated 3D campus model that combines every discipline in a single view. The website shows a 500MW campus concept resolved against a real site, with the ability to hover over each discipline—power, cooling, utilities, access, civil, structural, permitting—and see it located on the site.

This is more than a visualization gimmick. In traditional workflows, each discipline produces its own drawings, and coordination happens in meetings where conflicts are discovered late. Marengo's approach forces integration early: the 33 kV transformer line is placed alongside the direct-to-chip liquid loops, the service spine carries water and fibre between building rows, and the graded platform is designed with retaining structures and attenuation from day one. The result is a "decision-ready package" that lets developers see the site before they commit capital.

Half the time, half the cost: the pricing signal

Marengo's website makes a bold claim: "50% lower time" and "50% lower cost" for feasibility studies, without cutting corners. The cost claim is unusual in an industry where engineering fees are often a small fraction of overall project cost, but for developers, time is money in a very direct way. A 6-month reduction in pre-construction can mean a data center comes online earlier, generating revenue sooner and reducing the risk of technological obsolescence.

The pricing signal is also a positioning statement. Marengo isn't selling AI software licenses; it's selling a service with a guaranteed outcome. The company says it can deliver the feasibility study in half the time and at half the cost, which is a much stronger promise than "we have a cool AI tool." It's a promise that can be benchmarked, and the website explicitly encourages potential clients to "measure speed, engineering quality and decision confidence against your conventional workflow."

Why licensed engineers still sign the drawings

For all the AI talk, Marengo is careful to emphasize that designs are "reviewed and validated by licensed engineers before you receive them." This is not a legal disclaimer; it's a structural necessity. Data center permits require stamped drawings, and no AI tool can replace professional liability. Marengo's model is to use AI to accelerate the work, but the final deliverable carries the same engineering authority as any traditional firm's.

This also addresses a common fear about AI-generated design: that it's unaccountable. Marengo states that "your IP is never used to train models" and that "you own everything we produce." These are reassurances tailored to enterprise clients who worry about proprietary site data and design approaches leaking into a shared model. By keeping the AI in-house and the IP with the client, Marengo removes a major adoption barrier.

From one site to a portfolio: the scaling play

The website's final section outlines a three-step path: prove it on one site, benchmark it against your conventional workflow, then scale it across your portfolio. This is a classic land-and-expand strategy, but it's particularly well-suited to data center developers who often have multiple sites in various stages of development.

Marengo's pitch is that early speed compounds. If a developer uses Marengo for due diligence on one site, then moves to feasibility and concept design, the accumulated knowledge and faster iteration become a competitive advantage. The company positions itself not as a one-off consultant but as a long-term engineering partner that can handle FEED and permitting design as the project matures. The "give us a site" call-to-action is a low-friction entry point: you don't need to commit to a full project, just share a location and let Marengo show you what accelerated engineering looks like.

Marengo's bet is that data center developers will trade the comfort of a familiar, slow process for the speed of an AI-augmented one. The evidence so far—a Y Combinator backing, a clear value proposition, and a working product that produces real deliverables—suggests the bet is worth watching.