Key points
- Render CEO Anurag Goel calls the company's AI bet a wager on unclear demand, saying there may be a valuations bubble but real value is being created.
- More than half of Render's top 30 customers by revenue now build LLM functionality into their apps, up from six months earlier.
- Render's long-running workloads attract AI customers from Cloudflare, Vercel and AWS Lambda, which execute for around five minutes.
Render bets the company on running AI agents
Anurag Goel spent years hunting for a problem he could stay excited about for decades, and Render is his bet that running AI apps and agents is that problem. Serverless platforms built for five-minute executions leave long-running workloads homeless, and his customers are asking for the pieces of an agentic runtime on one bill. He calls it a bet rather than a certainty, since the shape of agentic memory is still unsettled, and he argues real value is being created even if valuations are bubbly.
Takeaways
- Render bets the company on AI agents. More than half of Render's top 30 customers by revenue now build LLM features into their apps, up from none six months earlier. Goel calls making Render the best place for AI apps and agents a bet.
- Ephemeral serverless leaves long-running work. Cloudflare, Vercel, AWS Lambda and similar platforms are built for executions of around five minutes, which is very different from keeping something running for months. That gap brings AI customers to Render, which serves application developers rather than platform engineers.
- Customers want one agentic runtime and bill. Customers want workflows, observability, retries, an AI gateway, sandboxes and a shared file system from one vendor on one bill. Goel says the product's shape is still unclear, and no one has settled what agentic memory means.
- Bubble valuations, real underlying value. Goel accepts there may be a valuations bubble but argues real value is being created, unlike crypto. The rational response is to solve customers' problems now rather than wait for the cycle to resolve.
- Goel refuses founder myth-making. He says outsized startup successes need a series of unlikely things to come together, and he will not spread a falsehood about Render because it might help the company. Leaders earn trust by showing they face the same challenges.
- Render grew out of a notebook experiment. After about six months on healthcare, Goel built a small product at South Park Commons that spun up GPU-backed Jupyter notebooks and reached 10,000 data scientists. Building for them did not excite him, so he combined developer productivity with cloud infrastructure.
How Goel ended up at Stripe
- In October 2010 a friend asked Goel to build a shared video game rental platform, which needed on-site payments. Braintree refused him, saying it only gave payments to VC-funded companies.
- He read a Paul Graham post about two engineers building a payments product and signed up for early access. John Collison sent the invite in December 2010, when the company was still called /dev/payments.