Silicon Valley Startups Build Cheap AI Models to Rival Chinese Systems

The artificial intelligence industry faces a shifting economic landscape as open-weight models developed in China gain global traction. AI firms.

Arcee AI and the 33-Day Pretraining Sprint

In late 2025, a little-known Silicon Valley startup named Arcee AI committed much of its remaining cash reserves to a high-stakes bet: building a highly competitive open-weight AI model. According to reporting by Kate Clark and Sam Schechner, the startup accomplished the goal following a 33-day pretraining run on a budget significantly smaller than those of the industry’s dominant labs. The resulting system is an economical artificial intelligence model that users can download and customize.

Open-weight models provide users with raw numerical values, known as weights, for each parameter stored within the system. This architecture allows developers to run models on specialized hardware and modify the underlying parameters with additional data through a process called fine-tuning. While some industry participants distinguish these from fully open-source models—which offer access to training code and complete transparency—the demand for customizable parameters remains high.

Venture Capital Hesitation and the Poolside Strategy

Despite mounting calls from policymakers and technologists for robust domestic open-weight options to counter foreign competition, securing financial backing remains a severe hurdle for homegrown pioneers. Every tier-one VC pretty much said no, Arcee Chief Executive Officer Mark McQuade told reporters, highlighting the reluctance of venture capitalists to fund free-to-use models that could potentially erode the established businesses of dominant players like OpenAI and Anthropic.

Arcee AI Builds Affordable Open Model to Counter Cheap Chinese AI

Other startups are pressing forward regardless. Poolside, an artificial intelligence company that released a new iteration of its open-weight model family called Laguna S 2.1 in July, sees an enormous market waiting for domestic alternatives free from geopolitical friction. There is a vast, vast degree of want for an American company producing the most-capable open-source artificial intelligence, said Jason Warner, co-chief executive and co-founder of Poolside.

Nvidia Backs Open Models Amid Growing industry Divisions

While traditional venture capital firms remain cautious, hardware giant Nvidia has emerged as one of the most vocal advocates and financial backers for open artificial intelligence. Led by Chief Executive Jensen Huang, Nvidia signed an open letter in July urging policymakers to avoid premature restrictions on open models. Nvidia has also constructed its own family of open models named Nemotron, formed an industry coalition to contribute to open research, and funneled heavy investments into startups including Reflection AI, Poolside, and Thinking Machines Lab.

Other industry leaders remain divided. Executives at Anthropic and OpenAI have expressed general support for open-weight systems while cautioning about inherent security trade-offs. Anthropic Chief Executive Dario Amodei noted in a blog post that open models carry risks of being misused for cyber or biological attacks. Meanwhile, U.S. companies like OpenAI have lowered prices on select high-end models to defend their market share, signaling the start of a broader reckoning over the future default status of open architectures.