Acres founder Carter Malloy’s two daughters press their faces to a glass window at the back of the office, trying to see the humming machines their father has been raving about—two high‑end GPUs tucked into a dark corner.
Malloy bought those two machines from NVIDIA in 2024, and just recently ordered two more, which should arrive later this week. He’s also threading new cabling through the ceiling to plug the machines straight into the computers of his data science team, so they can train models directly on‑site instead of renting time in the cloud.
“Having it on‑prem is just a lot cheaper to train—and actually faster,” Malloy says.
Acres may be a small startup of only about 70 people, but it is one of a growing number of niche data companies quietly assembling GPU clusters outside the walls of Big Tech, in a bet that owning their own compute will be a competitive edge. Andreessen Horowitz famously secured its own GPU cluster that it rents out to startups in exchange for equity. And individual startups including the video hosting startup Gumlet have said they are hosting their own hardware, too. This hardware can cost more than $25,000 per GPU, plus ongoing energy costs. During supply shortages like last year, it can be difficult for smaller companies to obtain them without months on waiting lists.
But to run a geospatial data intelligence company, Malloy says having their own cluster just made more sense.
It hasn’t always been this way. A few years ago, Malloy was running a very different company—AcreTrader, a Fayetteville, Ark.-based farmland investment fintech platform, in fact, that let investors buy slices of fields the way they might buy shares of a stock. Last summer, he sold off the “Trader” part of the business for an undisclosed sum to focus on one thing: data.
From the beginning, a small team at the startup had been hoovering up data to help landowners price and evaluate farmland—everything from sale and lease history and water infrastructure data to LiDAR topography, satellite imagery, and even the depth of water wells in Texas. Over time, the internal mapping and analytics stack “became bigger than Trader could, very quickly,” Malloy says, as land information is not only difficult and timely to obtain, but often requires data engineers to parse through.
As large language models became more sophisticated, Malloy envisioned new ways for customers to interact with the data his team was carefully pulling and cleaning. With the new Acres beta platform, a developer can type a plain‑English prompt: Find me a 40‑acre parcel that’s mostly outside the floodplain, within three miles of sewage infrastructure, in a county known for fast permitting—and the system combs through its maps and data to surface viable sites. Via Acres’ integration with the public information startup Hamlet, data center companies could also analyze whether local city and county governments are friendly—or not so friendly—towards new development and data center projects.
Enter the GPUs. Acres works with geospatial data—not just spreadsheets, but vector and raster layers that define the points, lines, and polygons behind land ownership and zoning maps. Crunching that kind of imagery and geometry is computationally heavy, and bringing GPUs in‑house lets the team train models and run site‑selection analyses faster and at lower cost, according to Malloy, who declined to comment on how much his utility bills had risen, apart from saying “it uses some power.”
Malloy is giddy as he talks about it. It feels to him like his team is operating at the frontier in data science. “We’re having breakthroughs in geospatial science with AI… We’re building things that there are no academic papers for.”
He may be overselling it a bit, but there is truth to the idea: combining parcel‑level land records, permitting data, and high‑resolution imagery at this scale with LLMs is still relatively new territory.
The only thing Malloy seems worried about is keeping up with the pace of change—and with demand. Acres started rolling out its new generative AI search functionality to enterprise customers just a few weeks ago, and Malloy says he has seen customers both swear and laugh over how much time they think it may save them.
Historically, Malloy says, Acres has tried to onboard customers too fast. With only five people on the customer support team, Malloy wants to move customers onto the new beta platform carefully. Not to mention—it’s been less than a year since Acres sold what had once been the core part of the business.
“That definitely keeps me up—that we’ll get ahead of ourselves. We’ve done it before,” Malloy said.
