GPUs, LLMs, and orchestration aren’t enough to make artificial intelligence run. Before the first rack can go into a data centre, months—sometimes years—of permits, environmental assessments, feasibility studies, and building approvals are needed. It’s an obstacle course that the Anglo-American startup Build promises to flatten with AI, announcing a seed round of $8.5 million (€7.4 million) led by undisclosed investors.

Build doesn’t build data centres or sell hardware. It’s software that automates the flood of documents and back-and-forth with local authorities required to get construction underway. The company claims its platform can cut bureaucratic workload by 95%, dramatically accelerating projects such as power lines, industrial plants, and, naturally, the facilities hosting the AI server surge.

Why paperwork stops bits

For those evaluating on-premise LLM deployments, bureaucracy is a silent adversary. While GPU specs, quantization, and temperature get the attention, the real bottleneck often sits upstream: finding a site, securing a power connection, complying with zoning rules. It’s no accident that the big hyperscalers have dedicated legal and real estate teams. Build tackles exactly that point, with an approach reminiscent of legal-tech: automated regulation parsing, guided form filling, compliance checks.

For on-premise AI, this has a double relevance. The first is obvious: if enterprises want to bring inference within their own walls, they need physical space and electrical power, and construction timelines directly impact TCO. The second is less discussed but critical: data sovereignty, a pillar of self-hosted AI, also depends on the ability to decide where and when to build. A platform that smooths relations with public administration could encourage local investment.

AI that builds AI

Build hasn’t released technical details about models or architecture, but the positioning is clear: apply generative AI to a domain rich in unstructured documents. This isn’t science fiction: legal teams already use LLMs to analyse contracts and regulations. Extending this to building and permitting procedures is a logical step, although it requires training on specialised corpora and careful hallucination management (a mistake on a permit can cost months of delay).

For infrastructure professionals, the news has a bittersweet taste. While Build’s promise is enticing, it also reminds us how software innovation often leaps ahead of concrete. But it’s also a signal: venture capital senses the enormous data centre hunger and backs those who remove friction. For on-premise stack planners, this is an early indicator of potentially greater availability of ready facilities down the line.

Beyond concrete

Build’s operation fits into a landscape where computational demand grows at triple-digit annual rates. Without an acceleration in construction, the gap between need and supply widens. For organisations exploring local LLMs today, the message is dual: on one hand, digital handling of paperwork reduces a major uncertainty factor; on the other, it shifts competition from the purely technological plane to the administrative one.

Those deciding between cloud and on-premise know that Total Cost of Ownership includes often-forgotten items: land costs, energy upgrade expenses, bureaucratic dead time. Tools like Build’s could compress those dead times, making initial CapEx more predictable. This is no detail: in projects where a GPU cluster costs tens of millions, every month of delay in the physical structure erodes expected ROI.

The startup is not alone. The theme of bureaucratic automation is emerging in parallel with the infrastructure AI race, and other initiatives are likely to follow. For AI-RADAR readers who analyse on-premise trade-offs, the lesson is plain: infrastructure is not just about silicon. Sometimes it takes a bit of AI to bring in more AI.