Fintech fixes AI's Unit Economics Problem

Hey Fintech Nerds 👋

Huge week for me, in lots of ways. First day at school for our eldest, Tempo* launched, Nerdcon agenda launched. Phew.

Ramp hit $1bn run rate revenue (are the new fintech poster-child?). Brex hit $700m and launched in Europe. Figure, Gemini and Klarna all announced their IPOs. We then got Visa dropping its Agentic Commerce API’s to developers, 

What caught my eye? The Figure and Klarna IPO numbers. Analyzed in Things to Know.

Your Rant this week, why I think every Generative UI - AI start-up will embed finance.

PS. No Brainfood next Sunday. Need to focus on the new day job juuust a smidge 😅

Weekly Rant 📣

Generative UI Economics Crisis Will Drive them to Embedded Finance.

Every AI coding platform is burning cash at an unsustainable rate. The solution isn't cheaper tokens. It's becoming a fintech company.

UI Generation (vibe coding) apps, Vercel and Loveable, turn anyone into an engineer or a business. They generate functional, professional-grade websites, apps, and services in minutes. 

They’re now some of the fastest-growing companies in history, making record revenues. But they have a massive cost problem. This will drive the B2B era of AI to hunt for a new business model. And that business model will be embedded finance.

Claude summarized my argument as

Need some convincing? Let’s unpack.

AI-Generated Software and User Experiences are Displacing SaaS

Vercel, Loveable, Bolt et al, feel like magic. Take a screenshot of your favorite website, and it will make an almost perfect replica in seconds. It’s like having a personal web design agency that delivers immediately and never costs more than $20/month.

These companies represent some of the fastest-growing in history, surpassing $100m ARR in record time. 

What’s truly impressive is how large-scale enterprises have adopted these (and their code-gen brethren). They’ve simultaneously unleashed a class of solo-praneurs generating $x million ARR, and have been adopted by Fortune 500 and big tech. It’s rare I see inside a scale-up these days that is not using these tools as the default.

If history is any lesson, then maybe they follow the Amazon playbook: operate at a loss for a decade or more, while growing the top line, and investor patience will be rewarded.

How?

The working assumption had been that these AI companies could afford to operate at a loss, as the cost-per-token was falling so quickly; it would be a matter of time before these companies became wildly profitable. The chart from a16z below became Twitter canon for about 6 months whenever the topic appeared.

The catch?

As the incredible Ethan Ding explains

What’s the COGS issue?

Three things at once.

1. 99% of usage is on the newest models

2. “Reasoning models” use orders of magnitude more tokens

3. The labs and hyperscalers can’t bring new compute online as fast as demand is scaling

The usage is all on the latest model, for the most tokens:

Tasks are getting longer. If the cost per token is $0.001. That’s fine for 1000 token response, but prohibitively expensive at 100,000 tokens.

Hyperscalers can’t add new capacity fast enough. Demand is unlimited. Supply is constrained. It takes time to build a data center. Physics doesn’t care about your ROI calculation. Meta, Google, Open AI and others can allocate the dollars for Gigawatt scale data centers, but they can’t create new power stations and get the data center operational overnight. Today’s capex is 2028’s capacity. 

These companies are trapped. 

They have to keep delivering new revenue, getting new customers, and they have tons of competitors vying for those users. They also have a circus of VCs trying not to miss the next big thing AI-boom. So the show continues. If token usage continues to balloon

Their growth rate might be stalling too

This could be a sign of maturity, or it could be a sign the bubble is starting to pop. They need better ARPU, not just more users paying a subscription.

The COGS trap is starting to show up in the end of “unlimited” plans, and will force these companies (and everyone in AI) to develop more mature monetization approaches. 

People are looking for metered usage already.

Smaller models offer hope. There is a trend towards much smaller language models that are highly specialized (with reinforcement learning or “RL”). Apple just dropped a tiny model that can run in-browser. 85x faster and 3.4x smaller is promising. 

But now people will use even more intelligence tokens. A model that can generate an 8-second clip much faster leads to a model that can generate a 2-minute clip for the same price. We’re greedy creatures. Our token demand will continue to outstrip supply, and 99% of developers choose the weapons-grade mode every time. 

Monetization Options

There are three relevant business models and three options they have to increase their monetization maturity and average revenue per active user. 

1. Sell ads (probably not for B2B

2. Increase subscription price

3. Do payments/commerce (embed finance)

Vibe coding, UI generation and coding tools don’t seem right for ads. Enterprise businesses don’t like to give away their data to 3rd parties, nor are those tools historically a platform to serve ads into. Slack, Jira, Google Docs might use some enterprise data but they don’t serve ads there. 

Maybe ads work as a B2B2C model (hey app maker person, want to display ads to monetize your users?) But as a core business model, the audience feels wrong. And would any of these businesses have enough data to sell to make it worth their while, or move the needle on the ARPU?

There’s a separate debate to be had about the consumer market… In a future brainfood.

Power users break the subscription model. The obvious choice these platforms have is to increase the subscription price. There’s just one problem. If you offer a $200/month unlimited plan, around 5% of users will use an order of magnitude more tokens.

(Enterprise pricing may be an option for the ultra-power users/companies, and some of the code-gen tools like v0, cursor and cognition labs have made headway here, but enterprise users won’t fix the cogs issue for the mainstream, the fundamental supply of compute / demand for tokens math is still broken).

Do payments / embed finance. This is a natural fit for B2B, e-commerce pages need to get paid, vertical SaaS services might want to let their customers get paid. Marketplaces will want to lend to their business customer base. Fintech Apps will want to allow users to make payments.

Embedded finance uniquely solves the COGS trap: Take rates scale with customer success, not token usage. When a business built on your platform processes $100k/month in payments, you earn $3k whether that took 1,000 tokens or 100,000 tokens to generate. The revenue grows with the customer's business, not your infrastructure costs.

Embedded Finance will be how AI-Gen Software Monetizes

In the SaaS Era embedded finance was how vertical SaaS providers (e.g., software for retail, or restaurants) monetized. It’s now a meme that “Shopify is a Fintech company” because Shopify generated $5.22 billion in merchant solutions revenue in 2023, 73.94% of total revenue.

Remember this image from a16z in 2020? Every company did become a fintech company (well, almost). 

But how does this work for the vibe coding platforms?

Anish Acharya suggests 

1. Vibe-coding apps move from prototype only to production grade

2. Therefore they are positioned to be the “platform” for growth businesses

3. They’ll monetize through hosting & payments

Embedded finance now comes in multiple flavors, and everyone from Stripe* and Adyen through to specialists like Pipe or Lithic have some option for rolling your own Fintech stack.

The jury is still out on if the vibe-coding apps become production platforms for managing your business. But lets assume that some do, why wouldn’t they also embed monetization? The vast majority of these platforms already let you setup subscriptions pretty quickly, but I imagine the progression will follow embedded Finance:

• Today AI is in the subscription era. But we’ve reached the upper limit of that model.

• There’s a slew of AI focussed billing platforms (like Polar or Lava Payments) for AI. 

• It’s not a stretch to imagine these platforms selling the ability to accept payments. I imagine some are getting closer to this.

• Then from there the whole universe of embedded finance opens up. Want to use revenue based finance for your new subscription creatine gummies startup you made in lovable? Click. Done. 

This evolution is generational. We’ve seen it with the hyperscalers of the 2010 era.

A few months ago I was at dinner with the former head of payments from one of the big on-demand deliver companies. They said to me 

Payments are the heart of most at-scale companies. And as they mature they’ll increasingly bring more of that capability in-house. 

These platforms will chase embedded finance because the unit economics finally make sense. Instead of losing money on power users as they build their business, you make more money as they succeed with it.

The same will be true of the AI-First generation. The only question remaining is how you position yourself accordingly.

What does that mean for you?

It’s one thing to accept payments online. 

It’s another entirely to help someone else do that. 

Or to have the ability to resell payment acceptance. 

Everyone who’s selling embedded finance will be licking their lips. But the work to do is to figure out the model here. I suspect early on these companies won’t want to be a merchant of record, and this is what’s holding back a lot of development in the space.

I think there’s a gap in the market for people who are deep in the Generative UI and platform space, with payments experience, who can help square this circle.

Every Generative UI platform needs a head of payments or finance. 

And that head of payments has a lot of work to do.

ST.

4 Fintech Companies 💸

1. Olito Labs - The AI that keeps up with bank compliance

Olito helps banks keep track of all regulations that impact them, any changes in guidance, and impacts new guidance will have on their existing operations. Post exam it can also help understand any changes needed and help create plans to remediate any gaps or issues.

🧠This is Group Risk and Compliance (GRC) for the 21st Century. More than 50% of compliance jobs are below the waterline of the compliance iceberg. When you think compliance you think AML, underwriting. You don’t think, someone keeping track of new regulations and understanding how that changes your third party oversight or audit approach. AI is perfect for this type of job. Banks should love this. Similar in idea to Themis or Cable, but more broadly applicable.

2. Rollout - Plaid for Real Estate

Rollout helps real estate software companies and brokerages connect CRMs, transaction management and title systems under a single API. With rollout developers could build an AI-native Zillow or Rocketmortgage, or an AI agent real estate broker. 

🧠Every esoteric database and legacy system has some API layer over it. Each of these companies can unlock the creation of new SaaS businesses, and in the AI era, I’m fascinated to see what else might appear.

3. Munify - The SMB borderless account built on stablecoins.

With Munify users can create accounts with their national ID and open a US-dollar-based account, even without a residency. Users can hold and send multiple currencies (USD, GBP, AED, EUR), get virtual cards, send money across borders. Soon the service will add invoices and pay-ins 

🧠Munify looks like the Wise borderless account, but built on stablecoins not fiat. It’s aimed at freelancers, remote workers with a focus on the Egyptian diaspora. Egyptians find it particularly difficult to move USD in and out of their country via traditional rails. But can, legally, open US-based bank accounts as a foreign national, hold USD stablecoins, and transact with those simply. With a 100m+ population, services like Munify could make that entire population more economically active. 

4. Holo - The home buying platform for the UAE and Saudi

Holo is a single platform for finding and securing a mortgage on a property. The platform has more than 500 mortgage products, promises full fee transparency, and has a concierge team that will handle negotiations and paperwork on the buyers behalf. 

🧠I’ve seen so many pitches by banks to create a “Zillow plus mortgage” service, and they never pull it off. Holo went and built it. The middle east is in the middle of a migrant boom, as it looks to attract middle-class western talent with its low taxes and economic opportunity. With high taxes, low growth and stagnant wages across much of Europe, many freelancers are choosing places like the UAE to live from. This has created a real estate boom, where buying property can unlock visa’s for your whole family. Holo fits neatly into this story. The UAE is the #2 fintech investment location in the entire world and you’re sleeping on it anon.

Things to know 👀

1. The Figure IPO is playing on every bankers worst nightmare

Targeting $4.53 valuation, $29M profit on $191M. Can the founder of SoFi kill the bank lending business model? Mike Cagney helped build SoFi into a $4B lending giant. Now he's asking public investors to fund his plan to make lending obsolete. Is this just PR for a lending business or genuine threat? 

🧠 Here's how Figure could disrupt lending

1. Stablecoins create an alternative to deposits for consumers.

2. Tokenized lending creates an alternative to banks for lending

Consider that private credit already intermediates nearly 50% of lending in the United States. Doing this with tokens could get a lot easier than the manual, hard-to-reach, hard to use private funding markets today.

🧠 Figure is rebuilding banking with tokens

• Deposits become yield bearing stablecoins. Figure launched YLDS The first SEC-registered yield-bearing stablecoin. earns 4.5% vs. a 0.5% savings account.

• Balance sheet lending becomes private credit on steriods. Figure opened "Democratized Prime." A market place for investors to "buy" loans pooled from Figure's $13B in credit extended.

🧠 Is this Peer-to-Peer 2.0 or the beginning of narrow banking? P2P lenders circa 2010 mostly went on to become specialist lenders and the “P2P” part died in favor of private credit funds. What Figure is doing is much more interesting:

🧠 Tokenization is a tech upgrade for private credit, that extends the trend.

• Bank lending requires massive balance sheets, regulatory capital, credit risk management. 

• Private credit needs complex contracts and lots of manual paperwork

• Tokenized credit automates most of this with software

• And many banks are launching private credit funds too

• This isn’t banks vs tokens. The opportunity is for everyone to adopt tokenization for efficiency.

🧠Cagney is quite happy to play to the fears of the banking lobby. In interviews: "We're contributing to deposit flight from traditional banks" and building tools to "disintermediate traditional capital allocators like banks and prime brokers."

🧠I prefer to see the land grab opportunity in Tokenization. It’s a way to abstract every lenders core ledger, and make it 24/7, instant and automated. Once that’s done the whole game of treasury and risk management changes from paperwork to software.

2. Klarna’s IPO is BNPL’s Gradulation to Consumer-Default

After years of speculation, one of Europe's biggest fintech companies is finally going public. Klarna just filed for their IPO. $14B valuation with a KLAR ticker and NYSE listing. Goldman Sachs leading.

Priced at $35-37 per share, aiming to raise $1.27B (34.3M shares) with the ticker KLAR on NYSE. Coming inwith Revenue: $823M (+20% YoY) but a $53M loss for the year. Their ace is their 111M active users 

🧠The obvious comp is Affirm. Affirm has 23M customers, similar ($876M) quarterly revenue, but just posted its first-ever GAAP profit ($58M). Trading at $26B valuation

🧠But they’re very different: Affirm: Higher margins, fewer customers, profitability, product innovation. Klarna: Massive scale, thinner margins, growth story, cross-sell op

🧠Cards are the next unlock. Affirm is CRUSHING, every earnings and innovated with their Affirm Card offering, which is now driving GMV growth. Klarna fast-followed and seems to be doing well with it. Klarna has a much higher existing customer base to sell into, but a lot of that is in Europe which isn’t a card-heavy market.

🧠 Is $14B the fair value for KLAR? Or are the bankers hoping it pops with comparisons to Affirm being right there? Pure Fintech IPO’s like Chime have been on a slide since their peak, and $CRCL is still lofty but not where it was a few months ago.

🧠 Can they achieve unit profitability with cross-sell? BNPL margins are brutal when so much of Klarna’s product is 0%. Klarna's new debit card hit 685k users in the US and just launched with Visa across 10 EU countries

🧠 Does Europe get its PayPal moment? 111M customers is an incredible foundation. A lot of this depends on the cross-sell. Early signs from the Klarna card are promising. Execution is everything.

🧠 BNPL, flexible cards, and consumer choice are the default choice for Gen Y and Gen Z consumers. You can see this in how successful Amex has been with similar “plan it” offerings driving growth in the affluent Gen Y and Z segments.

🧠What's surprising is how poorly traditional card issuers have been at fast-following this offering. Probably because it's P&L negative vs their current products. It’s hard to give full throated support to a less profitable product offering.

Good Reads 📚

1. Crypto credit cards might be the best credit cards period.

The birth of DeFi-based credit cards has some incredible user offers. Collateralized credit cards can give borrow rates as low as 4% (vs 20% on unsecured credit cards), and offer FX fees of 1% instead of 2%.

It’s critical to understand the nuance. 

🧠 With DeFi cards like EtherFi cash, you don’t offramp. ETH continues to earn 3% (via staking), and USDC continues to earn 9 or 10%. You then use your card to borrow against that vault. 

🧠It’s like a hybrid between a savings account and a credit card. “Put $10,000 in USDC into your EtherFi vault. It continues earning 9–12% natively around $900–$1,200 per year. Every dollar you spend earns an additional 3% or more cashback. That’s  another ~$300 if you cycle through the full balance.”

Tweets of the week 🕊

That's all, folks. 👋

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