Soaring Costs Prompt Fresh Interest in Open Source AI. Chinese Firms Are Way Ahead.
Investors tout a 'rebel alliance' of open-source startups across the AI stack
Price wars and political chaos at the frontier of AI have reignited the competitive battle over open source models.
Companies are reeling from the sticker shock of tokenmaxxing while Anthropic’s Mythos models are caught up in an export-control fight that has rattled the tech industry. That’s made a low-cost, more controllable alternative look more urgent.
But where it will come from, and who will make money from it, is far from clear.
American labs have ceded ground to the Chinese ever since DeepSeek’s breakout in 2025, with open models like Qwen and Kimi becoming the default foundation for startups around the world. Meta has pulled back from its once-touted open source strategy and others have steered away, spooked by the business-model questions and worries about security.
But a handful of Western companies, fueled by philosophical as well as financial fervor, are trying to win back ground in the open source arena.
Poolside, the $12 billion model maker backed by Nvidia, Bain Capital, and DST, has recently refocused the entire company around open source, co-founder and co-CEO Eiso Kant told us. He described the pivot as an ideological one — “the first and only ideological decision that this company has ever made.”
His argument is that there’s a dangerous concentration of power locked up in the three top model companies, comparing it to what foundational infrastructure like electricity would look like if only a few firms controlled the grid.
“We cannot have a world where every single company is going to be building upon intelligence that is either coming from only three companies in the world or is coming from Chinese players,” he said. This April, the company released its open source model Laguna XS.2, designed for agentic coding.
Geopolitics aside, Silicon Valley is buying into the belief that open source models will have an important role to play, with investors backing a growing group of startups across the AI stack that aim to capitalize on the opportunity. (See chart below.)
Michael Mignano from USV summed up the category with the Star Wars-inflected moniker “the rebel alliance.”
“For the past two years people have been model-maxxing and picking models at the frontier with little regard for how much they’re spending,” Mignano told us. “From startups’ CEOs to CEOs of big public companies, everyone is thinking about this.”
Other investors agreed that the industry was moving quickly to optimize on costs and diversify usage.
“No one wants all their token usage on a single company — especially the pricey ones,” Bill Gurley said to us over text. “Anyone not at the model layer wants this to happen.”