Maximum.com
A Miami-based fintech startup called Maximum emerged from stealth on Monday with one of the largest seed rounds in the industry’s history and a thesis its founder has been thinking about for years: the infrastructure that runs most American banks was built to remember things, not to understand them — and that distinction is about to matter far more than it ever has. Maximum’s emergence from stealth was announced alongside the funding on August 3, 2026.
The round totaled $30 million, led by CRV, with participation from Pear VC, Restive, Plug and Play Ventures, and Anthemis. Founded by Randy Fernando, who previously built and exited two fintech companies, Maximum is building what it calls an AI-native operating system for banks — a ground-up replacement for the legacy core banking systems that underpin more than 70 percent of U.S. banks today, according to Kansas City Fed core banking research on core banking market structure.
Why AI Agents Need More Than a Smart Layer on a Dumb Core
The specific technical problem Maximum is targeting is not widely discussed in startup fundraising announcements, but it is the right one to focus on. Legacy bank cores were engineered for batch processing: they collect transaction data throughout the day, then reconcile, settle, and update account balances in an overnight run. For decades that was adequate, because the humans who consumed the output of that process worked on the same schedule.
Autonomous AI agents do not. An agent tasked with scanning every transaction against the Treasury Department’s OFAC sanctions list, flagging suspicious activity, and routing alerts to compliance officers needs three things that batch-processing cores cannot provide in real time: a unified, continuously updated ledger it can read from; the ability to execute actions (send an alert, trigger a hold) rather than only generate recommendations; and a governed audit trail that meets regulatory standards. Research on agentic AI in banking architecture confirms that fragmented legacy cores fail on all three requirements. On a legacy core, the day’s full transaction data may not be available until the overnight batch completes. That means an AI agent running on a legacy core is, structurally, working from yesterday’s information.
“For decades, banks have attempted to serve the evolving needs of customers on infrastructure that was never designed for today’s world,” Fernando said in the company’s announcement. “The advancement of AI is creating a paradigm shift in financial services, and banks can’t keep pace by relying on legacy providers.”
This is what Fernando means when he calls Maximum an “operating system” rather than a “core.” The distinction is architectural, not marketing. A conventional core banking system is a system of record: it stores and retrieves. An AI-native OS is a control plane: it ingests, routes, executes, and audits — with AI woven into the data model and event stream from the start, not bolted on after the fact.
Who Controls American Banking Infrastructure — and What That Costs Banks
The competitive context for Maximum is a market structure that economists would recognize as textbook switching-cost oligopoly. Research from the Federal Reserve Bank of Kansas City found that Fiserv holds core contracts with approximately 42 percent of U.S. banks, Jack Henry with about 21 percent, and FIS with roughly 9 percent — meaning three vendors collectively control just over 72 percent of the market. About 20 smaller competitors share the rest. The banks in that 72 percent are not staying because they are satisfied.
Migrating from one core system to another has historically required 18 to 24 months and a dedicated internal team, on top of multi-year contract terms and early termination fees, Fernando said in an interview with Forbes. “It creates a system where it’s very, very sticky, and it’s very hard to move away.” The switching cost literature predicts exactly this outcome: when customers cannot easily change vendors, incumbents harvest their installed base rather than compete on improvement.
The result, visible in the day-to-day operations of community banks, is the kind of inefficiency that makes Fernando’s pitch immediate rather than theoretical. “Many banks have a team of humans that do upwards of four or five hours of manual reconciliation every day,” he told Forbes. “These are things that should be automated.” The reconciliation exists precisely because batch-processing cores cannot provide a continuously accurate picture of what the bank’s books look like — human reviewers close the gap between what the system recorded overnight and what actually happened during the day.
Regulators have begun saying something similar. The Office of the Comptroller of the Currency issued OCC Bulletin 2025-39 in November 2025 — a formal request for information specifically soliciting public comment on the challenges community banks face with their core service providers, including “contract negotiations and terms, fees, billing practices” and barriers to innovation. After the comment period closed in January, Comptroller Jonathan Gould told attendees at the American Bankers Association’s Washington summit in March that the commercial negotiating relationship between smaller banks and their core vendors was “very uneven.” That language, from the nation’s top bank supervisor, represents an unusual degree of federal validation for the market dynamic Maximum is pitching against. His remarks were confirmed by Banking Dive on core concerns.
How Maximum Plans to Actually Win — and Where Prior Challengers Failed
Fernando is not the first to identify legacy core banking as a disruption target. Thought Machine (backed by JPMorgan and Standard Chartered), Finxact (acquired by Fiserv in 2022 for $650 million), and Q2 have all pitched next-generation core modernization — and all, to varying degrees, ran into the same failure mode: building bespoke solutions for each new bank customer to generate growth metrics, accumulating technical debt in the process. “Other companies have tried to build modern cores,” Fernando told Forbes. “Those companies run into issues because you’re building bespoke solutions each bank that you try to onboard in order to show growth for your investors.”
Maximum’s stated approach to avoiding that trap is deliberate sequencing. Fernando plans to fully migrate one bank at a time — starting bank number one, completing the migration end to end, and only then moving to bank number two. No partial implementations, no bespoke additions to manufacture momentum.
The company has already attracted interest from larger community and regional banks, but as of its launch has no publicly named customers and no product in production — a fact Fernando acknowledged directly. That candor is part of the pitch: this is a long-game bet on infrastructure replacement, not a wrapper or add-on that can be sold quickly.
CRV’s conviction was visible in how fast the deal moved. Fernando initially set out to raise $25 million. After pitching Caitlin Bolnick Rellas, the CRV general partner who had led the firm’s seed investment in Fernando’s previous company, Power Finance, in 2022, a term sheet arrived within 24 hours — before Maximum had a finalized product, a complete team, or a company name. “Every generation of infrastructure eventually reaches a point where incremental improvements are no longer enough,” Rellas said in the company announcement. “We believe banking technology has reached that moment.”
Fernando brings a track record that gives the bet credibility. He founded Vault, a Portland-based retirement savings startup that Acorns acquired in November 2017. He then founded Power Finance in 2021 — a cloud-native credit card program management platform that Marqeta acquired in early 2023 for $275 million in cash. Several of the investors in Maximum — CRV, Restive, Anthemis, Plug and Play — backed Power Finance as well. That continuity is, in venture terms, an endorsement of the person as much as the thesis.
What Maximum Is Actually Up Against: Fiserv’s Counter-Move
Maximum is not launching into a static market. Fiserv, the largest of the three incumbent core providers, announced its own AI operating system for banks in May 2026. Called agentOS and developed with OpenAI and Amazon Web Services, it launched with six financial institutions co-developing the platform and two running agents in active pilots. Fiserv’s Co-President Dhivya Suryadevara described it as “the first place where banks can run Fiserv’s agents, build their own, and deploy from a curated set of partners — all under the same governance, identity, and audit controls.” Broad availability for agentOS was targeted for August 2026.
This is the competitive challenge Maximum’s architecture must answer directly. Fiserv’s agentOS is built to run “natively across Fiserv’s platforms — core, payments, issuer processing, and servicing.” In practice, that means the AI layer sits on top of Fiserv’s existing cores — platforms that still run batch-processing cycles and siloed data stores. The agents are new; the foundation they run on is not.
That gap is precisely what Fernando is betting banks will eventually be unwilling to accept. An agent that scans OFAC lists and flags suspicious transactions is valuable; an agent that can only do so using data from last night’s batch run is materially less valuable than one that operates continuously against a real-time ledger. The question Maximum must prove in production — with real banks and real migrations — is whether that architectural difference produces outcomes banks will pay to unlock.
Craig Focardi of Celent, one of the banking industry’s foremost technology research firms, offered a measured read of the category when speaking with American Banker: AI-based core systems hold promise but will be evaluated against the near-term ROI of infusing AI into existing systems. That is the market Maximum is betting will shift — not toward the promise of a new architecture, but toward the measurable performance gap between agents running on stale batch data and agents running on a unified real-time ledger.
Does the Architectural Difference Actually Matter for Community Banks?
The answer for most community banks will depend on the economics of the conversion. An AI-native core is architecturally superior for agent deployment — but the migration from a legacy Fiserv or Jack Henry system takes 18 to 24 months, costs tens of millions of dollars in integration work, and exposes the bank to regulatory risk during the transition period. For a $500 million community bank, those switching costs represent a material fraction of its technology budget for several years. Fiserv’s agentOS, by contrast, could be layered onto a bank’s existing core in weeks, with agents delivering measurable reconciliation and compliance improvements before the end of a quarterly cycle.
Fernando’s counter-argument, articulated across multiple interviews, is that the AI-layered approach compounds the underlying problem. Every bespoke addition to a legacy core adds custom code on top of custom code, making the system progressively harder to maintain and more expensive to modify. Badri Sridhar, a managing director in FTI Consulting’s financial services practice who works with banks on core compliance issues, described the dynamic to Banking Dive: banks that write custom code on top of core systems to solve compliance problems “have custom code, maybe on top of other custom code, on top of more custom code. Years pass by, and that compounds. It becomes very hard to untangle.” AI agents that must navigate that accumulated technical debt will be constrained by it.
Maximum’s $30 million will fund product development, team expansion, and bank implementations. The company is starting with individual full migrations rather than broad distribution, which means its first test of the architecture will also be its most demanding one. The answer — whether an AI-native core produces demonstrably better agent performance in production — will determine whether the infrastructure bet that attracted CRV’s term sheet in 24 hours turns out to have been the right one.
Frequently Asked QuestionsWhat is the difference between an AI-native bank core and adding AI to a legacy core?
A legacy bank core is a system of record built on batch-processing architecture: it collects transactions throughout the day and reconciles them in an overnight run. When AI is layered on top, the agents operating in that environment must read from whatever data the last batch produced — which means they can be hours behind. An AI-native core, by contrast, maintains a continuously updated real-time ledger that agents can read from and act against at any moment. The practical difference shows up in time-sensitive compliance workflows: an AI agent scanning for fraudulent transactions or OFAC sanctions matches is considerably more effective when it operates on a real-time ledger than when it depends on overnight batch data.
Why do so many banks stay locked into the same core banking provider for decades?
Switching a bank’s core system is one of the most expensive and operationally risky projects a financial institution can undertake. Research from McKinsey has estimated switching costs for mid-sized banks above $50 million; for larger institutions, the figure can exceed $300 million. Migrations typically require 18 to 24 months and a dedicated internal team, on top of the multi-year contract terms and early termination fees that core vendors build into their agreements. For community banks with limited technology budgets, the disruption risk alone is often enough to keep renewing contracts with incumbent providers even when satisfaction is low.
Did the federal government take a position on core banking provider lock-in?
Yes. The Office of the Comptroller of the Currency issued OCC Bulletin 2025-39 in November 2025, soliciting input from community banks on their relationships with core service providers, including challenges around contract terms, fees, billing, and innovation barriers. After the comment period closed, OCC Comptroller Jonathan Gould stated publicly that the commercial negotiating relationship between smaller banks and their core vendors was “very uneven” — an unusually direct acknowledgment from the top bank regulator that the market structure is producing outcomes it intends to scrutinize.
Is Maximum’s product available for banks to use today?
No. As of its August 2026 launch, Maximum has no publicly named bank customers and no product in production. The company has attracted interest from community and regional banks and plans to begin with full end-to-end migrations, starting with one bank at a time. The $30 million seed round will fund product development, team expansion, and the initial implementations. Investors are betting on the architectural thesis and the founder’s track record rather than on demonstrated production results.