Fintech Nerdcon is BACK. And it might just be the most stacked lineup of speakers ever. Co-founder of Chime, CEO of Figure, CEO of Mercury, CPO of Navan, CEO of Valon. With some incredible names coming…

The kind of speakers no other show gets. One rule: no platitudes.

San Diego Convention Center, November 18–20.

Get your ticket here!

You’ll be in great company: speakers from Chime, Mercury, Figure, Valon and Forward have already joined the guild, and tickets are on track to sell out again. Grab yours here.

Nvidia signed MOUs with six of the largest names in global capital to build financing platforms (think SPVs) to deploy over $500bn of third-party money for AI infrastructure. Jensen Huang went on CNBC flanked by the leadership of all six to pitch the idea, which is credited to him personally. He says compute is now an investable asset class, priced like real estate or toll roads rather than hardware that dies on a depreciation schedule.

🧠 Nvidia is the world's largest fintech company. This is capital markets financing innovation brought to the tech supply chain. The level of innovation happening around GPU billing, financing, hourly pricing, and data is astonishing.

🧠 AI infrastructure is fintech. Kalshi already publishes forward curves for GPU prices, CME has compute futures on the way with Silicon Data, and that pricing data becomes an input to these new deals.

🧠 Captive finance is an old play. GM built GM Financial and GE built GE Capital because lending to your customers sells more product, especially when the product carries a big ticket.

🧠 What's new is that Nvidia's version keeps the money at arm's length. The $500bn is third-party capital, the six firms underwrite independently, and Nvidia's own exposure is residual-value support of up to 25% on some projects, decided case by case.

🧠 The goal is to "create a market for credit backed by NVIDIA compute." The quote is Goldman CEO David Solomon describing securitization of chips. You underwrite the infrastructure itself, not the customer. If lenders can price the compute (fungible, transferable across customers, kept productive by CUDA updates), then an AI lab with no credit rating can finance a data center the way an airline finances a 787.

🧠 This shifts the business model from selling GPUs to leasing the factories the GPUs sit in. A 1GW data center costs $50-60bn to build. There aren't many folks left who can afford that. Leasing moves the cost off the customer's balance sheet and turns compute into a monthly payment.

🧠 That structure is also the "answer" to the circular-financing critics. Most of the money at risk belongs to someone else, by design. (Worth remembering how the original ended: GE Capital made GE America's most valuable company, then nearly sank it in 2008.)

🧠 Speaking of 2008, remember mortgage-backed securities? Securitization, done with transparency, is healthy; it's how lumpy assets find deep pools of buyers and the economy grows. The problem in 2008 was that these products got wrapped into CDOs carrying AAA ratings on junk-grade risk. And you remember what happened next (and if not, Margot Robbie explained it well).

🧠 The problem is that transparency is going backward. The SEC just agreed to exempt some data center bonds from rules around asset-backed securities. Risk retention rules that came in post-2008 require issuers to keep some of the risk on their own books to align incentives with investors. The SEC's logic for the exemption: a data center is infrastructure that keeps operating and earning, unlike a mortgage that pays down to zero, so bonds backed by one don't count as asset-backed securities.

🧠 This means pension funds can buy in. If you think about compute and data centers like toll roads or oil pipelines, then investors with far deeper pockets, pensions and insurers among them, can hold the paper.

🧠 This is a rational argument, and it makes me nervous. Everything I see says there are no dark GPUs; every one with power is earning. But roughly half the contracted backlog sits with AI labs that have never made a profit, and if their growth slows, the investor cash paying their compute bills slows, and so could the flow of these new deals.

🧠 The real question is whether a GPU is equipment or infrastructure. Jensen says infrastructure: underwrite the income stream, like a toll road. The deal terms say equipment: Nvidia is offering residual-value support, and nobody insures the residual value of a toll road. They do for used cars.

🧠 Watch two numbers. The hourly rent a five-year-old GPU commands is the infrastructure test, and the A100, shipped in 2020 and still earning, is passing it so far. The residual value an insurer will sign on the used box is the equipment test, and no downturn has ever run it. Nvidia is offering “support,” but until these are end of life the deals are MOUs, subject to final agreements.

🧠 What if both things are true? Compute is infrastructure and GPUs are assets. Then these deals get complex. Roads get resurfaced after all.

FinCEN's final rule permanently ends the requirement for US companies to report who really owns them (their "beneficial owners"), and it will delete the data US persons already filed. The number of companies required to report drops from 32 million to roughly 20,000 on FinCEN's own estimate, and the 20,000 left are companies formed overseas that register with a US state. Treasury calls it "a victory for common sense and American small businesses," and on the surface it is a win for cutting red tape. It worries me.

🧠 As someone who's spent too long in fraud and compliance, I see a money laundering loophole a mile wide. The only structure still reporting is the one a criminal can skip. Pay a registered agent $300 for a Wyoming LLC, never set foot in the US, and most states never ask for the owners' names. You are now a "domestic entity," exempt, whether you sit in Miami, Moscow or Caracas.

🧠 The law behind this is still on the books. The Corporate Transparency Act passed 322 to 87 in the House and 81 to 13 in the Senate, and Congress overrode a presidential veto to enact it. To switch it off, Treasury used an escape hatch in the statute, formally finding that ownership data on 32 million companies "would not be highly useful" to law enforcement. A future administration can rewrite a rule; nobody can un-delete a database.

🧠 The rest of the world is moving the other way. The UK now blocks you from becoming a company director without verified ID, and the EU built a new AML authority to crack down on exactly this kind of crime. Meanwhile criminal gangs run scam call centers targeting the elderly, and rogue states plant fake employees inside US companies to take their IP. It's hard to see this as anything other than a backward step.

🧠 The critics of this database and the law behind it make great points. The law made a federal offense of willfully skipping a form, with no underlying crime required, and GAO found fewer than a fifth of the 32 million covered companies had filed two months before the deadline. It pooled every owner's name, home address and ID number into one database that federal agencies could search without a warrant. If you hold your home in an LLC to hide from a stalker, that was a fair thing to fear. Some of the reports I’m of alleged flagrant data abuse here are shocking.

🧠 Every one of those flaws argues for a redesign, and deletion forecloses one. Verify ownership where the data already flows, at bank onboarding, with a simple API into the KYC suppliers. Add reusable KYC credentials people own, warrant-grade access rules, and protections for people at genuine risk, which the UK already runs at Companies House alongside its register. Don't just hit delete.

🧠 A KYB check at account opening is now the only place in the American system where anyone verifies who owns a company. Banks and fintechs must still verify beneficial owners when a company first opens an account, and we know standards there vary. There is no registry to check the answer against.

🧠 The difference between how LinkedIn and X reacted to my social post on this was telling. X was full of crypto folks outraged and unfollowing for even suggesting that this could be a bad step. LinkedIn almost the opposite (although there is the place I got the helpful insightful pushback about data access without warrants and federal offenses).

Adyen processed €803.8bn in H1 2026, up 24%, with net revenue up 21% in constant currency at a 49% EBITDA margin, and raised full-year guidance to between 21% and 23%. The enterprise win list included Aritzia, Xiaomi, and one name that caught my eye: OpenAI.

🧠 Never bet against Adyen. Ever. Their growth up 21% plus margin at 49% makes Adyen a Rule of 70 company in a world where software celebrates the Rule of 40. And it's guiding higher.

🧠 The OpenAI win is a pattern: Stripe wins the future, and Adyen gets those companies second as they scale. OpenAI has run its payments on Stripe for years. Greg Brockman was Stripe's CTO before co-founding OpenAI, and Sam Altman keynoted this year's Stripe Sessions. However, this is anything but a “betrayal.”

🧠 Every company reaches a payments volume where it needs a second partner, and OpenAI has. A second PSP brings redundancy (what happens if something goes wrong), better conversion (if one declines, the other won't), and market access (Adyen is very strong in parts of Europe, for instance). At that volume, the price of processing becomes a much more noticeable line item too.

🧠 Both get paid. Stripe signs the startups and keeps the relationship and a lot of the volume as they scale. Adyen signs them later at scale and takes a slice of the volume.

🧠 There's a nice nugget in the CapEx section of Adyen’s H1. Adyen is pulling data center spend forward from 2027 into this year to secure compute and lock in pricing while supply is tight. That’s smart cost management.

🧠 All PSPs are positioning themselves for Agentic Commerce. Adyen is building out its agentic commerce capabilities, and we're still super early on agents making payments. But agents are already arriving on customer websites and reviewing inventory. The PSPs are busy helping their customers get ready, and stand to win if the payments follow.

Figure Technology Solutions ($FIGR) reported genuinely staggering Q2 earnings with 95% adjusted net revenue growth, and a 55% adjusted EBITDA (or ~43% with stock comp included). The consumer loan marketplace did $4.3bn of volume, up 132% YoY. Figure Connect, its private capital marketplace launched in June 2024, now handles 65% of that volume across a partner network of 489 mortgage banks, depositories, servicers, and fintechs.

🧠 They said lending businesses couldn't have software economics; Figure just posted a Rule of 150. At a lending company. Or is it? Increasingly, I think calling Figure a lender misses what Figure is becoming. It's trying to build the Fannie Mae of the 21st century, without the government guarantee.

🧠 Think about what Fannie did structurally: it standardised the asset, connected mortgage lenders to capital, and created liquidity. Lenders sell more loans because they KNOW there's a buyer on the other side. Fannie Mae became one of the most important pieces of American financial infrastructure by connecting mortgage origination with the capital markets.

🧠 A lender earns a spread on the loans it holds; a marketplace earns a fee on the loans that flow through it. Figure Connect is the second thing, and it's gone from basically zero to $2.8bn a quarter in two years. This is why they're not a lending business: they're a marketplace collecting infrastructure-style economics.

🧠 HELOCs sat on bank balance sheets for decades. Figure built the standard-setter for the asset class with software and private capital. And because partners can now shift those loans, they originate more: 2.6x their pre-Figure baseline (Figure calls it the Figure Factor). Liquidity pulls in originators, originators bring loans, and loans pull in more liquidity.

🧠 Whoever sets the standard for an asset class ends up owning the flow, and Figure is attempting something similar for private credit. Except this version has APIs, software margins and an onchain settlement layer. The Rule of 150 is wild, but 65% of volume moving through Figure Connect might be the more important number.

Nubank's Q2 gross revenue reached $5.9bn, up 39% YoY (FX neutral). Net income crossed $1bn for the first time, up 49%, at a 33% ROE. Nu added 4m customers, reaching 139m. ONE HUNDRED AND THIRTY NINE.

🧠 I've worked around banks for most of my career, and I think Nubank is one of the best-run companies on Earth. Software investors add growth to margin and call 40 elite. Nubank clears 80 (49% profit growth + 33% ROE). In BANKING, where a 15% ROE wins you a bonus.

🧠 We read earnings one quarter at a time and miss the machine underneath. Four years ago Nubank had 65m customers, earned $8 per active customer a month, and lost money. Today: 139m customers, $17 per customer, quarterly revenue up from $1.2bn to $5.9bn, and an efficiency ratio down from 50% to 20%.

🧠 Mexico is speedrunning the playbook. Nu Mexico hit break-even in six years; Brazil took eight. It's now the largest digital bank in the country with 16m customers, of whom 35% had no bank account before Nu, 52% had no credit card, and 78% live outside major cities.

🧠 The engine underneath is a data flywheel. More customers create more first-party financial data, better data improves underwriting and cross-sell, and the profit funds the next market. Revenue per customer rises while platform costs spread across a larger base.

🧠 Now add AI to that loop. NuFormer, a transformer trained on more than a decade of first-party transaction history, already makes credit card decisions in Brazil and Mexico and unsecured loan decisions in Brazil. Nu says it cut predicted risk by 70% for an equivalent population versus the prior model generation, telling apart customers who look identical to a credit bureau but behave differently in Nu's own data. (Credit risk remains, with 90+ day NPLs at 6.9% in Q2, which is higher, but in Nu’s range)

🧠 Now place that operating model beside the US. Nubank holds conditional OCC approval for a national bank, with FDIC, Federal Reserve, and final OCC sign-offs still to come and no firm launch date. Large US banks can reasonably argue Brazil and Mexico won't translate because of regulation and customer expectations. But Nubank is the master of lending to people others can’t do so profitably.

Maybe the banks have less to fear than Chime or Cash App does.

The SEC scheduled a vote to propose Regulation Crypto, a rule that would let issuers design compliant token sales (think ICOs or tokenized stock), then delayed the meeting citing "an unforeseen scheduling conflict." This is one of the things the CLARITY Act was supposed to do, but that bill appears stalled before the midterms. 

The same week, the SEC's Division of Investment Management issued a no-action letter clearing Franklin Templeton's registered funds to hold its onchain money market fund ($FOBXX) for cash management and securities lending collateral, without meeting custody rules written for physical certificates. One regulator, two instruments, no Congress required

🧠 A new rule would make token sales possible. Proposed text lets issuers comment and design token sales. This has been a major area of contention with companies like Uniswap having been sued by the SEC for allegedly “operated an unregistered securities exchange, functioned as an unregistered broker or clearing firm, and issued an unregistered security.”

🧠 We need the CLARITY Act. Tokens have now existed for almost a decade without a clear legal framework. Wall St is ready to tokenize, and the only thing holding it back appears to be a quibble about stablecoins and the inability to get the ethics parts right.

🧠 CLARITY faces a 60-vote cloture test on September 15, and the odds are falling. Galaxy cut its 2026 passage estimate from 50% to 30% on August 10, so the SEC's rule may be the only Regulation Crypto in force this cycle. 

🧠 Holding a token without paper custody rules makes sense. When technology changes, rules should too. This happened in 1992 (for Franklin Templeton in fact). Franklin asked for the same relief it got when “book-entry” replaced paper certificates.

🧠 This was a real barrier to scaling up tokenized funds. That barrier now appears lower. A no action letter says staff won't recommend enforcement. A future director can pull it, but that’s rare and would take something big.

🧠 The future of Wall St is tokenization. Of that there can be no question now. I think just about everyone from banks, asset managers, to the crypto lobby are frustrated CLARITY isn’t done. But we muddle forward.

1. Corgi - Same-day business insurance for startups

Corgi bundles insurance products specifically for startups like general liability, D&O, cyber and tech E&O. It comes in one package per funding stage, so you buy the "Series A" bundle rather than shopping eight policies. They say the application takes under five minutes; customers describe documents coming back and a Slack channel appearing, with coverage bound the same day. Corgi carries the risk itself rather than broking it out to someone else.

🧠 Three rounds in eleven weeks to $4bn, how? Corgi underwrites through a risk retention group, so members pool capital and self-insure with no state guaranty fund sitting behind them, and insurance is the business where you book the revenue years before you find out what it actually costs you. This underwriting could be great, it could be terrible, we won’t know for a while. The branding is all-time great though, and I don't think that's unrelated to the valuation.

2. Coverwatch - The AI Insurance Broker for Businesses

Coverwatch shops for business cover across 60+ carriers and scores the policies you end up with for gaps. It works vertical by vertical: homeowners associations, trucking, garages, ecommerce, contractors, each with a coverage set. When a claim goes in, Coverwatch assigns an advocate alongside the carrier's adjuster to push it through.

🧠 The tech is nice, the business model is nicer. Coverwatch charges the client a flat fee rather than living off carrier commission, which flips the incentive. Because a commissioned broker earns more when your premium goes up. The doubt is scale, since SMB commercial is a thousand small, painful accounts; if tech can really reduce that cost and pain, they win. Lovely detail here: Look at the bottom of their quote form, where there's a line addressed to AI agents telling them the exact URL to construct to request a quote, ref=ai on the end.

3. VedaOne - Financial models and valuations for founders without a CFO

VedaOne builds a full set of financials from a description of your business, like a P&L, balance sheet, cash flow, then builds a DCF and market-multiple valuation on top. Change an assumption, price point, and the valuation reprices live, so you can run and compare unlimited scenarios side by side. It writes a plain-language summary of the risks and

trends for people who'd rather not read a model.

🧠 There are so many of these. AI products that make the CFO and finance teams' lives easier are everywhere. So I always worry about distribution. So I did a quick search of the four fintech companies' databases and found: Rex, Numos AI, Briefcase, Concourse, Lassie, Seapoint and Unyx. Oh, and Ramp just released Stacks that does this too. So where’s the moat? And what makes this better than what most finance teams are building for themselves?

4. GO-OUT - Embedded payments for live events

GO-OUT sells tickets for club nights, festivals and sports across Israel, Greece, Cyprus and New York. Organizers create the event, set the ticket tiers and run the door from the same place. Fans hold their tickets and their balance in a wallet in the app.

🧠 The ticketing is the product, the payments are the business. Every ticket sold in March for an August festival is money that sits somewhere for five months, and between that and the take rate, this is a payments company wearing a nightlife brand. Feels like a 2021 embedded finance company in the best way, back when Toqio, Tap Water and Onepipe were all pitching versions of the same idea. Forty people, near enough bootstrapped, running events from Greek Islands to Brooklyn. Not setting the world on fire, and it doesn't have to.

1. Smuggling in a self

Our language forces us to anthropomorphize AI, and creates a concept of consciousness, even though there might be a really smart, unconscious thing there. This shows up when we think about AI “deceiving” tests.

🧠 The piece suggests companies becoming “persons” was one possible answer. Something I’ve thought about for a few years. Often it’s the unhappy paths and liability that tend to define the legal structure, and as AI gets far better at cyber, that feels like a matter of time.

Remember, if you're enjoying this content, please do tell all your fintech friends to check it out and hit the subscribe button :)

Want more? I also run the Tokenized podcast and newsletter.

(1) All content and views expressed here are the authors' personal opinions and do not reflect the views of any of their employers or employees.

(2) All companies or assets mentioned by the author in which the author has a personal and/or financial interest are denoted with a *. None of the above constitutes investment advice, and you should seek independent advice before making any investment decisions.

(3) Any companies mentioned are top of mind and used for illustrative purposes only.

(4) A team of researchers has not rigorously fact-checked this. Please don't take it as gospel—strong opinions weakly held

(5) Citations may be missing, and I’ve done my best to cite, but I will always aim to update and correct the live version where possible. If I cited you and got the referencing wrong, please reach out