By Lina Irawan 
Malloy initiated this pivotal investment in 2024, acquiring the initial pair of NVIDIA GPUs. The immediate impact and observed benefits were substantial enough that he recently placed an order for two more, slated to arrive later this week. To fully integrate this powerful new hardware, the Acres team is meticulously threading new cabling through the office ceiling, a tangible commitment to connecting these machines directly to the workstations of his data science team. This direct connection bypasses the traditional reliance on cloud computing resources, enabling the team to train complex AI models directly on-site, within the physical confines of their own office.
"Having it on-prem is just a lot cheaper to train—and actually faster," Malloy states, articulating the core drivers behind his strategic pivot. This sentiment, seemingly straightforward, challenges the dominant narrative that the cloud is always the most cost-effective and scalable solution for startups. For Acres, a company of roughly 70 employees, the calculation proved different. The total cost of ownership (TCO) for cloud services, encompassing compute instance rentals, data storage, and crucially, egress fees for moving large datasets in and out of the cloud, can quickly escalate, especially for computationally intensive tasks like training large language models (LLMs) on vast geospatial datasets. By investing in their own hardware, Acres aims to gain greater control over costs, optimize performance, and potentially secure a long-term competitive advantage.
Acres is not an isolated case but rather an emerging exemplar of a growing trend. A select cohort of niche data companies and AI startups are increasingly opting to assemble their own GPU clusters outside the walled gardens of Big Tech cloud providers. This burgeoning movement represents a calculated gamble that owning their own compute infrastructure will provide a distinct competitive edge in the rapidly evolving AI landscape. The most prominent example is perhaps Andreessen Horowitz (a16z), the influential venture capital firm, which famously secured its own formidable GPU cluster. This strategic asset is then rented out to its portfolio startups, often in exchange for equity, effectively de-risking their investments and accelerating the development cycles of nascent AI companies by providing access to scarce and expensive resources. Individual startups, such as the video hosting platform Gumlet, have also publicly advocated for and implemented strategies to host their own hardware, citing similar benefits in terms of cost efficiency and performance.
The financial commitment involved in this strategy is substantial. Each high-end GPU can cost upwards of $25,000, and this figure doesn’t even account for the significant ongoing energy costs required to power and cool these high-performance machines. Furthermore, the global semiconductor supply chain, particularly for cutting-edge AI chips, has been under immense pressure. During periods of high demand and supply shortages, as witnessed in the past year, smaller companies often face considerable hurdles in acquiring these essential components, frequently enduring months-long waiting lists. Malloy’s ability to secure NVIDIA GPUs in 2024, a period still characterized by intense demand, speaks to either foresight, strong vendor relationships, or a willingness to pay a premium.
For Acres, however, the decision transcended mere financial calculations; it became a strategic imperative dictated by the very nature of their business. As a geospatial data intelligence company, their core operations involve processing, analyzing, and generating insights from massive and complex spatial datasets. Malloy emphasizes that for such specialized work, having an in-house GPU cluster simply "made more sense." The intricacies of handling petabytes of satellite imagery, LiDAR scans, and vector-based land parcel data demand a level of computational horsepower and low-latency access that on-prem solutions can often provide more efficiently than general-purpose cloud instances.
The current incarnation of Acres represents a dramatic evolution from its previous identity. Just a few years ago, Malloy was at the helm of a very different venture: AcreTrader, a Fayetteville, Arkansas-based farmland investment fintech platform. AcreTrader innovated by democratizing access to farmland investment, allowing investors to purchase fractional shares of agricultural land, much like buying shares of a stock. This model aimed to make a historically illiquid asset class more accessible. However, last summer, Malloy made a decisive move, selling off the "Trader" component of the business for an undisclosed sum. This strategic divestment allowed him to narrow his focus to a singular, increasingly compelling objective: data.
From its inception, AcreTrader had fostered a small, dedicated team focused on meticulously collecting and aggregating vast quantities of data to assist landowners in accurately pricing and evaluating farmland. This data repository was incredibly diverse and granular, encompassing everything from historical sale and lease records, detailed water infrastructure data, high-resolution LiDAR topography, and continuous satellite imagery, down to the precise depth of water wells in regions like Texas. Over time, the internal mapping and analytics stack developed by this team "became bigger than Trader could, very quickly," Malloy recalls. The reason for this exponential growth lies in the inherent complexity of land information; it is not only notoriously difficult and time-consuming to obtain from disparate public and private sources, but it also frequently requires sophisticated data engineers and specialized geographic information system (GIS) experts to parse, clean, and integrate into usable formats. This foundational data layer, initially a support function, gradually revealed its immense standalone value.
The advent of increasingly sophisticated large language models (LLMs) proved to be a catalyst, igniting Malloy’s vision for new, intuitive ways customers could interact with the meticulously curated data his team had been building. With the new Acres beta platform, the cumbersome process of querying complex geospatial databases is transformed. A developer, for instance, can now 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." The system, leveraging its powerful AI and underlying GPU infrastructure, then rapidly combs through its extensive maps and proprietary data layers to surface viable sites, complete with detailed analyses. This natural language interface dramatically lowers the barrier to entry for accessing highly specialized geospatial insights.
Furthermore, Acres has forged a strategic integration with Hamlet, a public information startup. This partnership allows data center companies, for example, to analyze not only the physical attributes of potential sites but also the crucial political and regulatory landscape. They can assess whether local city and county governments are generally favorable—or conversely, less amenable—towards new development and large-scale infrastructure projects like data centers. This contextual layer of information, often overlooked in traditional site selection, provides an invaluable competitive edge.
This is precisely where the in-house GPUs become indispensable. Acres primarily works with geospatial data, which extends far beyond simple spreadsheets. It involves intricate vector and raster layers that precisely define the points, lines, and polygons underlying land ownership records, zoning maps, topographical features, and environmental datasets. Crunching this kind of high-resolution imagery and complex geometric data is extraordinarily computationally heavy. Bringing these powerful GPUs in-house allows the Acres team to train their specialized models and run sophisticated site-selection analyses not only significantly faster but also at a demonstrably lower cost compared to relying solely on cloud providers. Malloy, while declining to comment on the exact increase in his utility bills, light-heartedly admitted that "it uses some power," a testament to the immense energy demands of modern AI hardware.
Malloy’s excitement is palpable as he discusses these advancements. He genuinely feels that his team is operating at the vanguard of data science, pushing the boundaries of what’s possible. "We’re having breakthroughs in geospatial science with AI… We’re building things that there are no academic papers for," he exclaims. While such a claim might seem a touch hyperbolic to some, there is a profound truth embedded within it. The ambitious endeavor of combining granular, parcel-level land records, dynamic permitting data, and high-resolution satellite imagery at scale with the interpretive power of large language models is indeed still relatively nascent territory. The practical application of cutting-edge AI techniques to such a fragmented, complex, and high-value domain presents unique challenges and opportunities that often outpace traditional academic research cycles.
Despite the groundbreaking work and enthusiastic outlook, Malloy remains acutely aware of the inherent challenges, particularly in maintaining the blistering pace of change and keeping up with escalating demand. Acres only recently began rolling out its new generative AI search functionality to a select group of enterprise customers. The initial feedback, Malloy notes, has been a mix of awe and amusement, with customers expressing both expletives and laughter over the sheer amount of time they believe the new platform will save them.
Malloy harbors a specific concern rooted in past experiences. Historically, Acres (and before it, AcreTrader) has, at times, attempted to onboard customers too rapidly, potentially overextending its resources. With a lean customer support team of only five individuals, Malloy is committed to a more measured and careful approach to migrating customers onto the new beta platform. This caution is amplified by the fact that less than a year has passed since the company divested what was once its core business. The rapid transformation and the weight of pioneering new technology bring their own set of pressures.
"That definitely keeps me up—that we’ll get ahead of ourselves. We’ve done it before," Malloy concedes. This self-awareness, coupled with the strategic investment in foundational technology, positions Acres at a critical juncture. The journey from a fintech platform to a cutting-edge geospatial AI powerhouse, powered by its own humming GPUs, is a testament to Malloy’s vision and the growing recognition that in the age of AI, owning the means of computation can be as vital as the data itself. The challenge now lies in harnessing this immense potential responsibly, ensuring that innovation is matched by sustainable growth and robust customer support, thereby solidifying Acres’ place as a leader in the next generation of data intelligence.