The two daughters of Acres founder Carter Malloy press their faces against a glass window at the back of the office, attempting to catch a glimpse of the humming machines their father has been extolling—two high-end GPUs nestled in a dark corner.
In 2024, Malloy purchased those two machines from NVIDIA, and just recently placed an order for two more, which are expected to arrive later this week. He is also running new cables through the ceiling to connect the machines directly to his data science team’s computers, enabling them to train models on-site rather than renting cloud time.
“Having it on-premises is significantly cheaper to train and actually quicker,” Malloy states.
Acres, a small startup with only around 70 employees, is among an increasing number of niche data companies quietly building GPU clusters outside the boundaries of Big Tech, betting that owning their own computing resources will provide a competitive advantage. Andreessen Horowitz notably acquired its own GPU cluster, which it rents to startups in exchange for equity. Individual startups, such as the video hosting company Gumlet, have also mentioned hosting their own hardware. This hardware can cost over $25,000 per GPU, plus ongoing energy expenses. During supply shortages like last year, smaller companies often face difficulties obtaining these GPUs without waiting for months.
However, for operating a geospatial data intelligence company, Malloy asserts that having their own cluster simply made more sense.
It hasn’t always been like this. A few years back, Malloy was at the helm of a vastly different company—AcreTrader, a Fayetteville, Ark.-based fintech platform for farmland investment that allowed investors to purchase shares of fields akin to buying stocks. Last summer, he sold the “Trader” segment of the business for an undisclosed amount to concentrate on one thing: data.
From the outset, a small team at the startup had been gathering data to assist landowners in pricing and evaluating farmland—ranging from sale and lease histories, water infrastructure data, LiDAR topography, satellite imagery, to even the depth of water wells in Texas. Over time, the internal mapping and analytics system “grew much larger than Trader could handle very quickly,” Malloy remarks, as land information is not only challenging and time-consuming to acquire but often demands data engineers to sift through it.
As large language models became more advanced, Malloy envisioned novel ways for customers to engage with the data his team was meticulously gathering and cleaning. With the new Acres beta platform, a developer can type a plain-English query: Locate a 40-acre parcel mostly outside the floodplain, within three miles of sewage infrastructure, in a county renowned for rapid permitting—and the system sifts through its maps and data to uncover viable sites. Through Acres’ integration with the public information startup Hamlet, data center companies can also assess whether local city and county governments are favorable or unfavorable towards new development and data center projects.
Now, the GPUs come into play. Acres deals with geospatial data—not merely spreadsheets, but vector and raster layers that define the points, lines, and polygons underlying land ownership and zoning maps. Processing such imagery and geometry is computationally intensive, and bringing GPUs in-house allows the team to train models and conduct site-selection analyses more swiftly and cost-effectively, as Malloy notes, who refrained from disclosing how much his utility bills had increased, simply stating “it consumes some power.”
Malloy is enthusiastic as he discusses this. It feels to him as though his team is at the forefront of data science. “We’re achieving breakthroughs in geospatial science using AI… We’re creating things for which there are no academic papers.”
He might be exaggerating slightly, but there’s truth in the notion that combining parcel-level land records, permitting data, and high-resolution imagery on this scale with LLMs remains relatively uncharted territory.
The only thing Malloy appears concerned about is keeping up with the pace of change and demand. Acres only recently began rolling out its new generative AI search feature to enterprise customers, and Malloy states he has observed customers both exclaiming in surprise and laughing at the potential time it might save them.
Historically, Malloy notes that Acres has attempted to onboard customers too quickly. With only five people on the customer support team, Malloy intends to introduce customers to the new beta platform cautiously. Not to mention, it has been less than a year since Acres sold the former core part of its business.
“That certainly keeps me awake—worried that we’ll overreach. We’ve done that before,” Malloy stated.