Max Levchin: Affirm targets $100B GMV doubling at 25% annual growth with AI writing 75% of code

Max Levchin: Affirm targets $100B GMV at 25% annual growth, with AI writing 75% of code

Affirm is targeting $100 billion in GMV, roughly double its current ~$50 billion, compounding at 25% annually — faster than the 20% it projected at its last investor forum three years ago, when it ultimately delivered 30%-plus. The company is simultaneously raising its profitability floor from 3–4% to 3.75% and above. Levchin framed the combination as a flywheel effect: a bigger company, a larger investor room, and a higher growth target than the last time it stood in the same building.

On TAM, the math leaves room. US revolving credit sits at roughly $1.3 trillion. A $100 billion GMV target is a meaningful but not dominant share, and Affirm is already live in the UK and Canada with additional markets announced.

Inflation and credit

On the macro, Levchin is cautious but not alarmed. During the last inflation spike, Affirm saw demand increase as consumers preferred not to pay upfront when prices were rising. Credit performance held through that cycle. The most recent quarter printed in line with expectations, with no visible deterioration in consumer credit. His standing rule: credit is job number zero — Affirm grows no faster than credit results allow, because losing credibility with debt investors is the one outcome he won't accept.

“Our CFO got on stage and said we're going to grow to $100,000,000,000 of GMV — we're just in the range of touching 50. We're going to compound at 25% every year and move our profitability target up. 75% of code last month was written by AI. It's not like a slow rise — it's rocketing. And it's not that people are writing less code; they're just that much more productive.”

AI and code

AI writing 75% of Affirm's code last month — up from 60% and still accelerating — is Levchin's clearest operational proof point. Engineers aren't writing less code or reviewing less; they're producing significantly more. Levchin describes it as the single biggest change AI has delivered across the business, ranking it well above translation, legal work, and other text-heavy functions where AI also helps.

The ROI is tracked weekly. A dedicated report, itself generated with AI assistance, monitors token spend by team, maps it to value produced, and flags conclusions downstream. Token costs are real — Levchin expects subsidy pricing to erode over time — but currently he describes the return as "undoubtedly very, very accretive."

A recent internal hackathon made the productivity shift concrete. 100 product managers, 37 projects, 27 hours — starting from a whiteboard and ending with shippable features presented to the company. Winners were graded not just on idea quality but on readiness to ship the next day. Most participants hadn't written code professionally since college.

Where AI doesn't work

Levchin's heuristic for evaluating AI vendor pitches: if you can describe the evaluation criteria the tool's makers are using, it's worth piloting. If you can't, it's "slop." Code generation works because the feedback loop is objective — does it pass the unit test, does it run? Agentic customer service works because you can ask customers whether they were satisfied and use that signal to improve. Underwriting model design and core infrastructure engineering are explicitly not candidates for AI replacement — the cost of a hallucinated underwriting model is real money lost, and an infrastructure error can cause a major outage. Those roles still require humans reviewing every line.

Depth as competitive strategy

Levchin's broader argument is that depth beats breadth, including in the AI era. His view: a computer science graduate who truly understands how software is made could become a 10,000x engineer rather than the 10x baseline AI is already setting. The engineers AI wants to learn from are the ones with real depth — and that's where the unattainable value sits. He applies the same logic to his own career: a prior startup outside payments underperformed, and Affirm became one of the world's strongest credit underwriters precisely because he kept going deeper into a domain he already knew.