Acres founder Carter Malloy’s two daughters press their faces to a glass window behind the workplace, attempting to see the buzzing machines their father has been raving about—two excessive‑finish GPUs tucked right into a darkish nook.
Malloy purchased these two machines from NVIDIA in 2024, and only in the near past ordered two extra, which ought to arrive later this week. He’s additionally threading new cabling via the ceiling to plug the machines straight into the computer systems of his knowledge science staff, to allow them to prepare fashions immediately on‑website as a substitute of renting time within the cloud.
“Having it on‑prem is just a lot cheaper to train—and actually faster,” Malloy says.
Acres could also be a small startup of solely about 70 individuals, however it’s one in all a rising variety of area of interest knowledge firms quietly assembling GPU clusters outdoors the partitions of Massive Tech, in a wager that proudly owning their very own compute can be a aggressive edge. Andreessen Horowitz famously secured its personal GPU cluster that it rents out to startups in trade for fairness. And particular person startups together with the video internet hosting startup Gumlet have mentioned they’re internet hosting their very own {hardware}, too. This {hardware} can value greater than $25,000 per GPU, plus ongoing power prices. Throughout provide shortages like final yr, it may be tough for smaller firms to acquire them with out months on ready lists.
However to run a geospatial knowledge intelligence firm, Malloy says having their very own cluster simply made extra sense.
It hasn’t at all times been this manner. A couple of years in the past, Malloy was operating a really totally different firm—AcreTrader, a Fayetteville, Ark.-based farmland funding fintech platform, the truth is, that permit traders purchase slices of fields the best way they may purchase shares of a inventory. Final summer time, he offered off the “Trader” a part of the enterprise for an undisclosed sum to deal with one factor: knowledge.
From the start, a small staff on the startup had been hoovering up knowledge to assist landowners value and consider farmland—every part from sale and lease historical past and water infrastructure knowledge to LiDAR topography, satellite tv for pc imagery, and even the depth of water wells in Texas. Over time, the interior mapping and analytics stack “became bigger than Trader could, very quickly,” Malloy says, as land data is just not solely tough and well timed to acquire, however usually requires knowledge engineers to parse via.
As giant language fashions grew to become extra subtle, Malloy envisioned new methods for patrons to work together with the info his staff was rigorously pulling and cleansing. With the brand new Acres beta platform, a developer can kind a plain‑English immediate: Discover me a 40‑acre parcel that’s largely outdoors the floodplain, inside three miles of sewage infrastructure, in a county recognized for quick allowing—and the system combs via its maps and knowledge to floor viable websites. By way of Acres’ integration with the general public data startup Hamlet, knowledge heart firms might additionally analyze whether or not native metropolis and county governments are pleasant—or not so pleasant—in the direction of new growth and knowledge heart tasks.
Enter the GPUs. Acres works with geospatial knowledge—not simply spreadsheets, however vector and raster layers that outline the factors, strains, and polygons behind land possession and zoning maps. Crunching that type of imagery and geometry is computationally heavy, and bringing GPUs in‑home lets the staff prepare fashions and run website‑choice analyses sooner and at decrease value, in keeping with Malloy, who declined to touch upon how a lot his utility payments had risen, aside from saying “it uses some power.”
Malloy is giddy as he talks about it. It feels to him like his staff is working on the frontier in knowledge science. “We’re having breakthroughs in geospatial science with AI… We’re building things that there are no academic papers for.”
He could also be overselling it a bit, however there’s fact to the thought: combining parcel‑degree land data, allowing knowledge, and excessive‑decision imagery at this scale with LLMs continues to be comparatively new territory.
The one factor Malloy appears nervous about is maintaining with the tempo of change—and with demand. Acres began rolling out its new generative AI search performance to enterprise clients just some weeks in the past, and Malloy says he has seen clients each swear and chortle over how a lot time they assume it might save them.
Traditionally, Malloy says, Acres has tried to onboard clients too quick. With solely 5 individuals on the client assist staff, Malloy desires to maneuver clients onto the brand new beta platform rigorously. To not point out—it’s been lower than a yr since Acres offered what had as soon as been the core a part of the enterprise.
“That definitely keeps me up—that we’ll get ahead of ourselves. We’ve done it before,” Malloy mentioned.
