Pump.co

Somewhere in San Francisco, a CFO is closing 14 browser tabs

It is a Tuesday. A second-time founder is staring at her AWS bill. The number is bigger than last month. It is bigger than her runway model said it would be. Her infrastructure engineer is on vacation. Her CFO is on a flight. Her sales team is asking why marketing got a raise. Somewhere on the seventh floor of 1455 Market Street, an algorithm built by a company called Pump.co notices, quietly applies a savings plan, and trims four figures off her bill before she has finished her coffee.

This is, more or less, the entire pitch.

AWS pricing is a game played by giants

Amazon Web Services rewards scale and patience. If a company can promise to spend a few million dollars over three years, it gets a discount. If it cannot, it pays retail. Most startups cannot. Most startups, in fact, can barely promise to exist for three years. So they pay retail, and they call it "the cost of doing business," and they wonder why their gross margins look the way they do.

This was the inefficiency that nagged at Spandana Nakka after her first company, the fintech startup Sleek, was acquired by Snackpass in 2021. She had spent enough nights inside the spreadsheet of a young company to know that cloud spend was not just an expense - it was a tax on being small. The fix, she suspected, was not better discipline. It was better pricing power.

The Costco model, applied to compute

Nakka's bet, made formal in 2022 inside Y Combinator's Summer batch, was deceptively simple: pool the AWS commitments of hundreds of small companies into one giant commitment, claim the enterprise discount, and pass the savings back. Think of it as a buyers' co-op for compute - or, as the company prefers to put it, the Costco for cloud.

The mechanics are less folksy than the metaphor. Pump.co underwrites multi-year reserved commitments at AWS scale, then re-slices them into roughly month-long commitments that startups can actually live with. If a customer over-commits, Pump.co reallocates the unused capacity to another customer or sells it on the AWS Marketplace. The customer never sees the gears moving. They just see a smaller bill.

Fig. A - The numbers the founder put on a slide, and which appear, against the natural order of pitch decks, to be roughly true.

One platform, three jobs, none of them glamorous

Pump.co is no longer a single trick. The platform has split into a small constellation of products, each aimed at a different unglamorous corner of cloud operations. Together they form what the company calls, with engineering-flat affect, the "intelligent cloud."

The connecting thread is what the founder calls "no engineering effort." This is the line in the marketing copy doing the most work. Most FinOps tools require a team to interpret them; Pump is built to act. It reads, decides, and applies. The customer's role is to plug in an account and walk away.

Fig. B - A timeline assembled from public filings, blog posts, and the kind of half-confirmed details that founders enjoy dropping at parties.

The receipts, such as they are

A claim is only as good as the people willing to defend it. Pump.co's numbers are public-ish and the customers are real. The platform has signed more than 250 companies in roughly six months, including more than fifty Y Combinator portfolio companies and a steady drip of seed-to-Series-B teams that would otherwise be paying AWS retail. The average customer reports about 30% savings. The ceiling, advertised loudly and only occasionally hit, is 60%.

The partnerships do quieter work. Pump is a listed AWS Partner, which is less a stamp of approval than a structural necessity: the whole model rests on AWS treating the company as a counterparty large enough to negotiate with. The Techstars deal, announced in 2024, plugs a similar pipeline of pre-priced startups into the platform. The Y Combinator origin story keeps the customer acquisition cost roughly halved.

Make enterprise pricing the default

The company's stated mission is to democratize enterprise-level cloud cost optimization. Read past the corporate phrasing and what is actually being argued is more interesting: that cloud bills should not be a function of how good a startup's procurement team is, because most startups do not have one. The pricing power gap between a fifteen-person company and a fifteen-thousand-person company is, in this view, an accident of scale that software should erase.

It is a meaningfully different stance than the dashboard generation of FinOps. Tools like Cloudability and Vantage tell you where your money goes; Pump.co tries to put some of it back. The first is observation. The second is action.

Why this matters past the next funding cycle

The interesting question about Pump.co is not whether group buying for AWS works - it clearly does, at least at this scale - but what happens when the same mechanic is pointed at everything else. Cloud security commitments. Database licensing. AI inference. Anywhere a price tag scales with patience and volume, there is room for a co-op. Pump.co's roadmap, with Pump Secure and Pump Intelligence already in flight, is a quiet bet that the next decade of FinOps will not be about reports - it will be about agents that act on a customer's behalf, across the entire infrastructure bill.

There are risks. AWS could change its discount structure. A hyperscaler could build the feature in-house. A larger competitor could undercut on rebate share. But the model has a defensive moat that grows with usage: the more customers Pump.co aggregates, the bigger the commitment it can underwrite, the better the price it can negotiate, the more customers it can attract. This is the most boring kind of flywheel and, historically, the most reliable.

Back to the Tuesday

The CFO returns from her flight. The infrastructure engineer comes back from vacation. They check the dashboard. The AWS bill has, against all precedent, gone down. Nobody had to negotiate anything. Nobody had to write a script. Nobody had to read a 40-page reservations whitepaper. A small percentage of the savings will flow back to Pump.co, which is how the company gets paid. The rest stays in the startup's bank account, where it will quietly extend a runway that nobody will ever brag about extending.

That is the thing about good infrastructure. You only notice it when it stops working. Pump.co is betting it can become exactly that kind of invisible.

Where to find them