Everything About AI
NextFin News - Golub Capital is putting $5 million behind the machine that may replace its own analysts. The New York private credit manager, which oversees roughly $90 billion, has made an equity investment in F2 AI Inc. and, as part of the deal, secured access to F2's suite of AI agents to screen private credit investments, perform due diligence, underwrite deals and monitor portfolios using Golub's own data. The firms did not disclose the valuation implied by the investment.
The transaction is more than a vendor contract. By paying for equity rather than a subscription, Golub has turned its most valuable underwriting asset — decades of proprietary deal data — into the training ground for the very software that could displace junior underwriters. That choice captures the central tension of AI in finance: the institutions best positioned to profit from the technology are the ones whose people it threatens.
The Deal: Equity, Not a Subscription
F2 AI, an artificial intelligence developer specializing in credit, is receiving the $5 million investment from Golub Capital. Under the arrangement, Golub gains access to F2's AI agents across the full credit lifecycle — screening investments in private credit, performing due diligence, underwriting deals, and monitoring portfolios — trained on Golub's own proprietary data. The structure matters: this is not a software license but an equity stake that aligns F2's success with Golub's underwriting outcomes.
For F2, the Golub arrangement is validation from one of the largest and most conservative allocators in private credit. Golub Capital, founded in 1994, reports total firm assets of $90 billion and a direct lending platform staffed by more than 230 investment professionals, including 90 senior professionals. Since inception it has reviewed more than 30,000 transactions and completed roughly 2,700, amounting to approximately $200 billion in deployed capital. The firm works with more than 280 repeat sponsors, giving it one of the deepest proprietary deal-flow networks in middle-market lending.
F2 itself is a recent arrival with deep roots. The company spun out of Arc, the cash management and capital markets platform, with an oversubscribed $10 million equity round backed by nearly 50 of Arc's long-time investors, including NFX, Left Lane Capital and Y Combinator. Together with Arc, that brings combined capital raised to nearly $200 million. F2's CEO and co-founder Don Muir said the company's AI helps teams evaluate deals more than 60% faster.
"F2 is building the technology I wish I had during my days in private equity and private credit... Our AI doesn't just automate manual work; it strategizes, reasons, and executes in ways that legacy software can't."
Muir has compared F2's ambition to the market-data terminals that transformed public-market investing decades ago, arguing the same shift is now arriving in private markets. The name F2 is a deliberate signal: it is the keystroke for editing cells in Excel, the spreadsheet shortcut that generations of bankers have worn into their keyboards. Muir founded Arc during his second year at Stanford's Graduate School of Business with classmate Nick Lombardo, raising $181 million of private credit and venture capital to disrupt startup banking. The company's roughly 300 lenders on Arc Capital Markets did not just want the private credit deals, he said — they wanted the AI powering them. That demand became F2, which bundles Arc's AI underwriting technology into a standalone SaaS product. In the first half of 2025 alone, adoption of that technology grew tenfold before the spin-out.
Golub, for its part, has been moving toward more complex, equity-adjacent strategies that demand deeper diligence. In October 2025 the firm launched a GP-led secondaries strategy, committing more than $1 billion to invest in middle-market continuation vehicles — transactions that require more nuanced underwriting than a standard senior secured loan. AI that can read a data room, reconcile financials and flag inconsistencies is more valuable to a firm doing continuation deals than to one writing plain-vanilla unitranche.
Why This Is Structural, Not Cyclical
The easy read of the Golub-F2 deal is that it is another entry in the 2025-26 AI funding cycle — a wealthy financial institution chasing the technology of the moment. That reading is wrong. Three pieces of evidence point to a structural regime shift rather than a cyclical wave, and getting this distinction right determines whether the deal looks like a one-off experiment or the first of many.
First, the workflow being replaced has not changed in thirty years. Private markets recruit some of the brightest minds in finance and put them to work parsing data rooms, spreading financials and marking up investment memos — low-leverage work that burns out junior teams while adding little to investment judgment. Muir has argued that "the old playbook that's guided the industry for the last 30 years will be obsolete within the next 36 months." A technology that compresses a thirty-year workflow into three years is not a cycle; it is a regime change. Cycles revert; workflow revolutions do not.
Second, the moat is data, not models. General-purpose AI hallucinates and lacks context — an unacceptable failure mode in finance, where a single misplaced decimal can move millions. F2's answer is a deterministic Excel engine, a formula-level audit trail and a precedent library built from a firm's own historical deals. When Golub's agents train on Golub's data, they encode the firm's specific underwriting standards, sector preferences and risk tolerances. A competitor can license the same foundation model; it cannot license Golub's thirty-year deal history. That asymmetry is structural, and it compounds: every deal underwritten improves the agent, and every improvement makes the next deal faster.
Third, the economics of private credit demand it. Private credit has grown from a niche alternative strategy into a core institutional allocation. The five largest listed private credit managers now hold a far larger share of perpetual capital than the 35% they held in 2021; if the growth rate seen since 2021 continues, these firms will manage almost $5 trillion in permanent capital by the end of the decade. Capital at that scale cannot be underwritten by analysts working the way they did in the 1990s. The labor constraint is structural — experienced credit analysts are scarce and expensive — and AI is the only lever that scales judgment without scaling headcount linearly.
There is a cyclical overlay worth separating cleanly. The $5 million check arrives during an exuberant AI funding cycle, and venture capital for AI startups could cool if interest rates stay higher for longer or if AI returns disappoint investors. Valuations in the AI software layer are stretched, and a funding winter would hit F2's ability to raise its next round. But the adoption drivers — labor scarcity, deal complexity, and dry powder searching for yield — do not depend on the funding cycle. They are embedded in the industry's structure. A cyclical funding pause could slow F2's hiring; it would not send lenders back to manual financial spreading.
The Second-Order Trade: Golub Is Buying Distribution, Not Software
The first-order effect of the deal is obvious: Golub gets faster underwriting. The second-order effect is more interesting, and it is where the real value of the transaction sits. By taking equity, Golub has become a distribution channel for F2. A customer that is also a backer validates the product for every other lender in Golub's network — the 280-plus repeat sponsors the firm works with, and the broader private credit ecosystem that watches what the market's leading direct lender does.
Scott Peters, founding partner at RevTek Capital, which backed F2's spin-out, put it plainly: "F2's agentic AI is a game-changer. We're already using F2 to streamline our underwriting workflows and unlock faster, deeper diligence across our pipeline. This isn't hype—it's happening now. AI will fundamentally reshape how deals are sourced, underwritten, and monitored in private credit. F2 is leading that shift."
This customer-as-backer model short-circuits the traditional enterprise software sales cycle, which for financial institutions can run twelve to eighteen months through security reviews, procurement committees and pilot programs. An equity relationship aligns incentives and accelerates adoption. It also creates a network effect that a pure subscription vendor cannot match: every deal F2 underwrites at Golub improves the agents, which improves the product for the next customer, which generates more proprietary data, which improves the agents again. The asset Golub is really buying is not the software — it is a share of that data flywheel.
The precedent is visible elsewhere in the market. In February 2026, ToltIQ, an AI-powered due diligence platform for private markets, partnered with Intapp to integrate its diligence platform with Intapp DealCloud, connecting private equity and private credit firms' relationship data with what they learn during diligence. The pattern is consistent: the winners are not the firms with the best chatbot, but the ones that embed AI into the workflow where the data already lives. Golub's move goes one step further — it does not just embed the tool, it owns a piece of the toolmaker.
There is also a talent-market second-order effect. F2's pitch is that it lets deal teams deploy more capital "without adding headcount." For a firm like Golub, with 230 investment professionals reviewing thousands of deals, that translates into operating leverage: more deals per analyst, or the same deal volume with fewer junior hires. Over a full credit cycle, that operating leverage compounds into margin expansion — if the AI holds up.
The Strongest Case Against the Thesis
The bear case deserves a full hearing, because it attacks the foundation of the structural argument rather than its edges. AI credit underwriting is unproven through a real downturn. Most models now in production were trained on the benign credit conditions of 2021 through 2025, when default rates were historically low and liquidity was abundant. When the cycle turns and borrowers struggle, an AI trained on good times may systematically understate risk — not because the technology is flawed, but because the data it learned from was. Credit models do not fail in the abstract; they fail at the turning point, and the turning point is exactly what the training data lacks.
There is also a human-capital risk that compounds over time. If AI agents absorb the judgment-building work that junior analysts used to do — spreading financials, building models, writing first drafts of memos — the industry may find itself without a pipeline of experienced underwriters in five to ten years. Automation that looks efficient today can hollow out the apprenticeship model that produced the 90 senior professionals Golub now relies on. The firm that automates its training program may one day discover it has automated away its successors.
A third objection is model convergence. If every major lender licenses similar AI agents trained on similar foundation models, underwriting could converge toward the same answers. That would compress credit spreads as everyone identifies the same "good" deals, and it would amplify losses when the models collectively miss the same risks. Homogeneous underwriting is a systemic risk dressed up as efficiency.
The falsifying signal is specific and observable: if F2-vetted or AI-underwritten deals at Golub show default or loss rates materially above Golub's historical net loss rates through a downturn vintage — or if a major private credit lender publicly abandons AI underwriting after a bad year — the structural thesis weakens materially. Until that happens, the burden of proof sits with the skeptics. A single bad quarter would not be enough; the signal requires a full vintage cycle, because credit is judged over years, not quarters.
What to Watch
Short term (6-12 months): Watch for adoption announcements from other private credit managers. If two or more large lenders announce similar equity-for-access deals with AI vendors, the Golub structure becomes an industry template rather than an outlier. Also watch F2's own fundraising: a successful next round at a higher valuation would confirm that Golub's validation opened doors; a flat or down round would suggest the market sees the deal as a one-off.
Medium term (1-3 years): The key metric is not speed but loss rates. Investors should watch whether AI-underwritten vintages hold up through the credit cycle relative to human-underwritten ones. This is where the thesis is won or lost. Speed gains are easy to measure; risk selection is only visible in hindsight, and the hindsight will arrive with the next recession.
Long term (3+ years): The structural question is whether AI becomes the default underwriting layer for private credit. Muir's 36-month obsolescence timeline is aggressive, but the direction is clear. The laggards, he has warned, "will fall away, and before long, AI-powered investing will be the norm." The firms that own proprietary underwriting data will have the durable advantage; the firms that rent generic AI will compete on price.
Base case: AI underwriting becomes standard for middle-market private credit within three to five years, with equity-for-access deals like Golub-F2 as the dominant commercial model. Upside case: proprietary data moats consolidate the industry around two or three AI platforms, and the firms that own them capture outsized economics while their lender-partners earn better risk-adjusted returns. Downside case: a downturn exposes model risk, adoption slows, and AI remains a diligence aid rather than a decision engine — a productivity tool, not a paradigm shift.
The Golub-F2 deal is a wager on which future arrives first. Golub is betting that the institution that trains the machine on its own data will out-underwrite the one that doesn't — and that the junior analyst it no longer needs to hire will be replaced by something faster, cheaper, and trained on thirty years of its own best work. The rest of private credit is watching. If Golub's AI-underwritten deals outperform, the industry will follow. If they don't, the thirty-year playbook gets one more extension.
Explore more exclusive insights at nextfin.ai.
