AI as a Hardware Appliance Helps FIs Gain Cost and Operational Control

By Roy Urrico

The strategic benefits of AI focus on risk management, operational speed, hyper-personalization, and financial resilience. Despite this, AI is becoming a governance and operational risk issue, particularly in regulated industries, according to Chicago-based Go Abacus, which provides AI infrastructure for regulated industries.

In March 2026, Go Abacus announced the launch of the Go1, an all-in-one-box system solution designed to deliver secure, enterprise-grade AI within an organization's own infrastructure, without sending data to a third-party, cloud-based AI provider. It directly plugs into existing systems, delivering AI capabilities to thousands of users while maintaining data control and ownership.

David Moscatelli, CEO, Go Abacus.

“The Go1 will do for AI in financial services what the Mac did for personal computers — deliver a user-friendly, right-sized appliance that gives access to the applications transforming what is possible. We are taking the large AI data centers and fractionalizing them to small hardware devices that can serve thousands per appliance,” said David Moscatelli, CEO, Go Abacus, in the announcement. “In 15 minutes, any bank or credit union, with any level of IT team experience can have blazing fast, secure, high-performance local AI support for up to 2,000 users in one machine.”

Go Abacus’s Senior Associate, Operations Mariano Apodaca, and an expert on AI for regulated and legacy industries, sat down with Finopotamus for an in-depth conversation about the AI landscape in the financial services industry.

AI Banking Infrastructure

Mariano Apodaca, Senior Associate/Operations, Go Abacus

“Today everything is AI. But there is a really prominent distinction in how we are helping, particularly companies in regulated industries focusing on banking, credit unions, financial services,” Apodaca explained. The distinction falls between AI tools such as large language models (LLMs) which he describes as foundational models, and inference, the process of putting a trained machine-learned model to work.

“Where Go Abacus fits into this conversation is instead of being like, ‘Hey, you should just adopt this technology because it is revolutionary,’ we wanted (organizations) to think about it more critically about the scrutiny these particular companies are under with the regulatory bodies,” said Apodaca.

The question for financial institutions and other organizations is, “how to get the benefits of the technology of AI, but also have a secure environment to do that? So, we have that built out for our customers from top to bottom,” Apodaca maintained.

What this does, noted Apodaca, is it gives Go Abacus customers “full sovereignty” from potentially giving away personally identifiable information (PII), or any data that should not be leaving their organization to these foundational models. “But it also gives them sovereignty on pricing, and how they want to use those tools inside of their organization rather than opening the door,” he added.

Don't Lower the Drawbridge

Apodaca relayed that a customer described how their organization spent millions of dollars building up a digital fortress to protect their financial institution. “And these AI companies now want us to put down a drawbridge and let them have free access to all of our information. It is too risky for my business.”

GO Abacas puts its product inside of that fortress, noted Apodaca. Organizations using Go1 have full ownership and control of how AI is used and costs, as well as what information connects to it. “We're giving those companies the ability to have this technology without all the inherent risks.”

Apodaca also dove into the current status of artificial intelligence: “At the end of 2024, (AI) basically consumed all of the available information on the internet. This is where they've created LLMs used today. The only information that is left now is the information inside of these organizations that is proprietary and basically their trade secrets. And yes, they might be coming into these accounts and asking them to lower down that drawbridge.”

Apodaca continued, “Having that sovereignty and that security of their information and the way that you operate is so important to these financial institutions and really can be their competitive advantage.”

Providing Information Governance

Go Abacas also focuses on protecting the user from receiving inaccurate information, referred to as hallucinations. An AI hallucination is when a model definitively produces false, fabricated, or nonsensical information.

“Our technology removes hallucinations in the (financial institution) because of the way that we process that information,” said Apodaca. “We're pulling the direct correlation to the policies, procedures, and the regulation of that company, rather than assuming what that information means, which is what AI does today. It is a lot implied meaning.”

Information governance is also an important aspect of Go1. “Credit unions in our industry, they are getting audited. When they get audited, they need to be able to prove the technology they're using is accurate,” explained Apodaca. “We basically prepare those companies for that audit.” The preparation includes protection against hate speech and potentially using AI in a nefarious manner.

AI Cost Volatility Creates Issues

AI pricing is becoming harder to predict, administer, and operationalize, according to Go Abacus. “You may have recently heard in the news AI is now more expensive than your average engineer,” confirmed Apodaca. More companies are reporting despite the vast power of AI, the technology is actually more expensive than the humans it’s supposed to replace, according to a Prof G Media.

Financial institutions pay for what they consume, measured in text data portions called AI tokens, “So instead of these (AI) companies raising their prices dollar per dollar so that you can make a side-by-side comparison, they're actually taking more token costs to run similar prompts, whether that is cogen (co-generation), accessing information or doing research,” said Apodaca.

Apodaca also shed light on the difficult decisions organizations and CFOs are having when deciding on capital allocation for AI. That leads to another cost dilemma. Apodaca pondered, “What happens when CFOs spend $20 million, but they do not get a $20 million ROI? They look for areas that cut cost in the business.”

For financial institutions and organizations that are cautious about AI use because it can get extremely expensive, Apodaca suggests Go1. “We do not charge a consumption cost. We license the product to the customer at a per month baseline fee. And the more that the customer actually uses it, the more cost efficient it actually becomes. Because then you get to see that direct ROI in comparison to the foundational models. We just charge a flat fee rather than (saying), ‘Hey, you used it 10,000 times and we're going to charge you now 10,000 times.”

Risk Management

There is a steady march toward AI regulation and its potential impact hanging over financial institutions. On April 17, 2026, the federal banking agencies issued SR 26-2: Revised Guidance on Model Risk Management, updated supervisory guidance that officially superseded and replaced SR 11-7 for Model Risk Management at financial institutions. Although both share similar fundamental risk principles, SR 26-2 shifts the focus from rigid, prescriptive compliance checklists to principles-based, risk-driven institutional judgment.

Concurrent to the guidance update, the federal banking agencies issued a request for information (RFI) asking the industry for feedback on how financial institutions are utilizing AI—specifically generative AI and agentic AI models—so regulators can establish future supervisory expectations.

“(Federal agencies) do not want to overregulate (AI), but they do realize that when they do push real regulation down, they want it to be really distinct,” explained Apodaca. “So, they're doing an RFI out to the market right now.”

The results of the RFI, expected in the fourth quarter of 2026 or early 2027 will help define the rules for generative AI, suggested Apodaca. “As an organization, you can spend millions of dollars setting up AI today. What happens in December when those rules change, you are going to have to spend another couple million dollars to adjust those inside of your business and on top of all the additional costs that we are seeing today in terms of usage. This is the reason we meet with regulators monthly and sit with them to make sure we are being compliant far before those rules get put into place.”

While AI continues to dominant technological innovation and evolution, particularly in banking, Apodaca provided some additional food for thought. “There are some things with AI that cannot be undone. This is not like the internet where you can post things and take them down and they disappear. Once this information goes into these foundational models, it sits there forever and there is nothing you can really do about it."

Apodaca added, "So, if we are not very thoughtful about the technologies we are adopting today, you are inherently putting that information at risk forever. And companies really need to think of the implications of this into the future? if we can educate the market on that, and particularly our industry, I think that they will begin to think about which tools they select.”