While Anthropic's record IPO signals investor appetite for frontier AI, Fireworks AI co-founder Chenyu Zhao is making a sharper economic argument: most enterprises don't need expensive closed models at all.

At the Slush conference in Helsinki, Zhao pitched open-weight models—publicly downloadable AI systems that companies customize with proprietary data—as a faster, cheaper alternative to services from Anthropic and OpenAI. The economics are straightforward: open-weight models avoid the premium pricing of frontier labs while letting companies build defensible custom versions rivals cannot easily copy.

Fireworks specializes in this model. The four-year-old Bay Area company helps enterprises run and fine-tune open models on their own data. Zhao argues that most business applications don't require the highest-end generalized intelligence. More pointedly, he raised a data risk frontier AI labs don't advertise: even simple prompt interactions feed customer information into their systems.

Microsoft CEO Satya Nadella echoed this concern in July, flagging data privacy as a friction point in enterprise AI adoption.

But Fireworks' message faces credibility damage. The company disclosed two security breaches this fall. In one incident, hackers accessed API keys and authentication tokens for a limited number of accounts. A second breach compromised Fireworks' cloud provider, exposing customer data including billing records and datasets.

Notion reported that transcripts from less than one percent of its workspaces were exposed, prompting the company to sever its Fireworks integration. Uber suffered a larger theft: hackers stole names, addresses, birth dates and driver's license numbers for approximately 150,000 drivers through a vendor later identified as Fireworks.

The breaches create a problem for Fireworks' core argument. If open-weight models are sold partly on the security advantage of keeping data in-house, then losing customer data to cloud provider compromise undermines the pitch. Enterprises comparing Fireworks to closed frontier AI labs now have a concrete reason to question the company's execution.

Fireworks raised $17.5 billion in valuation this summer from backers including Sequoia and Benchmark. The company is in preliminary discussions for another funding round, though Zhao declined to comment.

Competition in this space is sharpening. Together AI and Baseten have announced major funding rounds recently. On Wednesday, Palo Alto Networks agreed to acquire Applied Compute for $4.5 billion in an all-stock deal. Applied Compute, founded a year ago by former OpenAI researchers, runs the same business model as Fireworks—helping companies train and deploy custom AI on proprietary data.

Zhao spent a decade at Google leading a 50-person engineering team on Google Cloud Vertex AI before co-founding Fireworks with Lin Qiao and Dima Dzhulgakov.