Hardware is back and now it’s running AI

By Sean Murphy, an interview with David Moscatelli, CEO at Go.Abacus

There is a particular kind of embarrassment that anyone who has worked in a bank or any large institution will recognize. A customer sits across the desk, asking something simple, something the institution handles hundreds of times a week, and the person on the other side does not know the answer. David Moscatelli remembers it clearly: a customer in his office at a credit union, wanting to place a travel alert on a debit card before a trip to Spain. Moscatelli, then a college student working as a financial service advisor to pay his tuition, had no idea where to find the procedure. It happened again the next day, and the day after that.

Large institutions often have thousands of procedures, some of them that are very rarely used. Moscatelli recognized that the most valuable employee in the office might not have been necessarily the smartest person; it was often the person that just knew where everything was.

That experience, repeated across years in public accounting and financial services, became the inspiration behind Go Abacus, the Chicago-based AI infrastructure company Moscatelli co-founded in 2022.

At a big four accounting firm, he estimates that auditors spent eighty percent of their working time simply looking for things: contracts, working papers, or old invoices. At the credit union, policies existed but were incomplete, riddled with assumed knowledge, the equivalent, as he puts it, of a Martha Stewart recipe with half the steps missing. The information existed but getting to it fast enough to be useful was another matter.

What Moscatelli built, initially using basic natural language processing before large language models existed, was a way for financial institutions to ask a question and receive an answer, including the document itself. The technology has evolved considerably since 2018, when the idea first took shape, but the underlying logic has not: institutions hold enormous amounts of knowledge inside their walls, and most of the people working inside those walls cannot access it quickly enough.

The Go1, launched in March 2026, is the latest expression of that idea. It is a physical appliance: a box, roughly the size of a compact server and drawing no more electricity than a household refrigerator, that ships to a bank or credit union pre-configured with Go Abacus's custom-trained AI model. An Ethernet cable connects it to the institution's network. Within fifteen minutes, up to two thousand employees can begin querying it. No cloud. No third-party exposure. No data leaving the building.

“Hardware is back," Moscatelli says, with a degree of self awareness about how that sounds to an industry that spent two decades migrating everything upward. What the cloud gave in accessibility and scale, it also took in transparency and control. With the Go1, the AI is in the room, auditable, unpluggable, entirely owned.

The pricing model is unusual for a sector where AI costs have become notoriously difficult to forecast. Go Abacus charges a flat monthly fee per Enterprise subscription regardless of query volume, a structure Moscatelli contrasts explicitly with the usage-based models that dominate the market. He draws a deliberate parallel with Uber's early years in the United States: fares were kept artificially low to drive adoption while taxis disappeared, and the alternative has largely gone. OpenAI reported a loss of two and a half billion dollars last quarter, a figure he reads as intentional underpricing during their early customer acquisition phase. 

Institutions that build their infrastructure around those providers, he argues, are leasing a portion of a model rather than owning one. The difference, he suggests, will matter considerably more as usage scales.

For community banks and smaller credit unions, the proposition addresses a capability gap that has widened.

Large institutions carry the engineering resources to build and maintain private AI deployments; most smaller ones do not. A recent EY survey of banking institutions found that regulatory compliance and data privacy ranked as the two most-cited barriers to AI adoption across the sector. Large institutions carry the engineering resources to navigate those barriers and run private AI deployments; most smaller ones do not. The Go1 is positioned as the answer to that asymmetry: owned infrastructure, compliance-ready from day one, operable by an IT team of any size, at a fixed and predictable cost

What Moscatelli has created was something he could have used himself: a tool that made institutional knowledge instantly reachable, closing the gap between what a fifteen-year veteran knows and what a new hire can access in their first week on the job.suggests, will matter considerably more as usage scales.