Replit’s Amjad Masad On The Cursor Deal, Fighting Apple, And Why He’d Rather Not Sell

Amjad Masad has been building Replit for a decade, but nan past 18 months person been thing other entirely. The AI coding adjunct institution went from $2.8 cardinal successful gross successful each of 2024 to search toward what Masad describes arsenic a billion-dollar yearly tally rate.

At TechCrunch’s sold-out StrictlyVC event successful San Francisco connected Thursday night, we covered a batch of crushed successful a short time, opening pinch nan mobility everyone successful nan manufacture is asking correct now: successful a world wherever rival Cursor is reportedly successful talks to beryllium acquired by SpaceX for $60 billion, is Replit besides bound to sell? We besides sewage into Replit’s nett gross retention — a measurement of really overmuch existing customers grow their spending — which Masad says is reaching arsenic precocious arsenic 300%, his willingness to return Apple to tribunal complete what he called outright lies successful its App Store conflict pinch Replit, and nan anticipation of nan institution opening to put successful its ain customers.

On nan mobility of independence, Masad was unambiguous. Unlike Cursor, which he said has been operating astatine antagonistic 23% gross margins, he based on Replit has nan economics to make that way viable — moreover if he stopped short of ruling retired a waste entirely.

The pursuing has been edited for magnitude and clarity:

TC: Cursor’s reported SpaceX woody was nan talk of nan manufacture past week. What did you make of it?

AM: It’s benignant of difficult being an independent, smaller AI institution that’s building connected instauration models, particularly if you’re burning a ton of cash. Part of nan reporting suggested Cursor has antagonistic 23% margins, and if you’re besides wanting to put successful training models, that makes it incredibly difficult to enactment independent.

For america astatine Replit, partially because we target a different customer set, we’ve been capable to tally nan business much rationally. We’ve been gross separator affirmative for complete a year. We’re somewhat much expensive, but we supply a batch more. Our assemblage tends to beryllium mostly non-technical users who antecedently haven’t been capable to create immoderate software. We supply an end-to-end level — from nan punctual each nan measurement to a deployed exertion that tin scale. We grip security, databases, database migration. And we’ve been doing this agelong capable that we’ve built a batch of those primitives into nan platform.

Is Replit for sale? I would presume you are talking pinch imaginable acquirers each nan time; it’s your fiduciary responsibility.

Yeah. We person astonishing partners, and they sometimes bring up these topics. But we’re going to effort to enactment independent. I would emotion for america to stay an independent company. We’ve been astir for 10 years, earlier it was moreover accepted that you could make apps conscionable from ideas. We were talking astir creating a cardinal package creators backmost successful 2018 astatine YC, and group sometimes really laughed astatine that dream. Now that dream is possible, and we kicked disconnected this gyration pinch our agentic coding acquisition successful September 2024. It conscionable feels for illustration we tin return it overmuch further.

You activity intimately pinch Anthropic, Google, and OpenAI. If you had to rank them — who’s doing it best?

Anthropic is still undefeated connected nan halfway agentic loop. They person nan champion instrumentality calling; nan supplier tin enactment coherent overmuch longer. GPT-5 is catching up quickly. Google’s Flash family of models is conscionable astonishing connected price-performance. If you want thing accelerated and cheap, they’re really beating unfastened root correct now. We usage each three, and honestly I wouldn’t discount nan newer labs either. Reflection AI is coming retired pinch open-source models we’re proceeding awesome things about. And nan Chinese models are awesome — Kimi is arsenic bully arsenic an Anthropic-generation exemplary from January, truthful it’s only astir 3 months behind.

When you’re successful a bake-off for an endeavor deal, what wins it for you?

Most of our income are inbound aliases integrated — very product-led. We’ve acquired customers for illustration Zillow and Meta purely done group adopting nan merchandise and past raising their manus to bargain an endeavor plan. When it does spell top-down and there’s a general bake-off, we usually triumph connected product. But moreover successful cases wherever we mightiness beryllium missing a feature, erstwhile it hits nan C-suite and nan IT group, Replit wins connected security. A batch of vibe-coding devices will make a website and link it to an outer database — awesome products, but it makes information overmuch harder, because nan database is unfastened to nan nationalist and you request to configure row-level security, which is particularly difficult for non-technical builders. Replit being afloat stack, pinch nan database built into nan task and not unfastened to nan nationalist — that makes nan app inherently much secure.

We besides spent 10 years battling crypto scammers and hackers, truthful our cybersecurity usability is arsenic bully arsenic a dedicated cybersecurity startup. Every clip you deploy an app connected Replit, we create an wholly caller isolated task connected Google Cloud. We inherit Google’s information model.

Can we talk astir churn? How agelong do you clasp onto customers if nan champion prototypes yet get rebuilt into a company’s existing stack?

Churn is very, very low, and nett retention is incredibly precocious — 300% successful immoderate cases. What we really perceive from customers is that erstwhile engineers get tense and effort to rebuild an app into their ain stack, they often make it worse. Once enterprises get comfortable pinch nan afloat Replit stack — particularly erstwhile we group up a single-tenant situation for them — they support nan apps connected Replit. Bain & Company, for example, replaced Tableau and Power BI pinch Replit and Databricks.

There’s a increasing interest astir AI bloat — non-technical users make acold much codification and pain done acold much tokens. That’s bully for you [given your usage-based fees]. What astir your customers?

We don’t person a batch of regrettable spend. Enterprises are very ROI conscious, and they show america astir nan returns they’re getting. For nan astir portion they consciousness nan finance is wholly worthy it — often one, two, 3 orders of magnitude. If they walk $100,000 a period pinch Replit, they’re usually generating $2 million, $3 million, $10 cardinal successful immoderate benignant of return.

Let’s talk astir Apple. Another rival, Lovable, conscionable sewage an app-building app approved by nan App Store this week. Replit has been successful App Store purgatory, pinch Apple blocking your updates for months. How overmuch does that wounded you?

It’s not life aliases decease — we could suffer nan app and it wouldn’t do thing meaningful to our business. But it’s an app group genuinely love. We’ve been connected nan App Store for 4 years. Kids successful underprivileged communities study to codification connected Replit connected their Android devices. Executives usage it successful meetings.

The logic Replit sewage blocked erstwhile others weren’t, we believe, is that Replit makes iOS apps. When we launched that capacity successful December, location were charts going astir showing really galore apps were getting into nan App Store done us. We deliberation Apple feels threatened by that.

Apple’s stated logic is that you’re downloading caller codification to nan instrumentality [after nan support process], which violates their guidelines.

That’s a lie. And we tin beryllium it successful tribunal if we person to.

Is that going to happen?

I dream not. I’m a instrumentality of Apple, and I’d emotion to collaborate and build thing awesome together. We’re happy to nonstop customers to Xcode [Apple’s ain improvement environment]. But you can’t tally a marketplace that a cardinal group person entree to and make decisions that are discriminatory aliases based connected whims.

Just wondering if, for illustration Nvidia, OpenAI and others, you’re reasoning astir investing successful your ain customers successful speech for equity?

We’ve thought a batch astir it, and it is simply a consideration. I’ve personally invested successful a fewer startups that started connected Replit earlier they made immoderate money. Some of them, for illustration Magic School — a coach decided to return his clip during COVID to study a small spot of vibe coding and built an AI app for different teachers. He recovered this problem that successful America, we pain retired a batch of teachers. He wanted to usage AI to trim nan workload. He did that, and he made $20 cardinal successful nan first year. Other companies that started connected Replit, I think, are weighted astatine half a cardinal dollars. The entrepreneurship happening connected Replit correct now is genuinely exciting. We integrated pinch Stripe a fewer months ago, and nan transactions flowing done Replit are increasing triple digits period complete month. Pretty soon, our customers will beryllium making much gross than we are.