On the same day Anthropic went public in the largest IPO on record, a competing vision for artificial intelligence emerged at the Slush startup conference in Helsinki: most businesses don't need expensive frontier models at all. That's the central argument from Fireworks AI, a four-year-old Bay Area company that runs open-weight AI models for enterprises including Uber, Shopify, and Notion. Co-founder Chenyu Zhao says companies are turning to platforms like his because frontier models are costly and, in most cases, unnecessary unless you need the very latest generalized intelligence.

Fireworks specializes in open-weight models that anyone can download, then helps customers tune them on proprietary data so they run faster and cheaper than offerings from Anthropic, OpenAI, and similar heavyweights. The company was assigned a $17.5 billion valuation in summer and is reportedly in early talks for another funding round. But Fireworks has also suffered two security incidents this fall: in one breach, hackers accessed API keys and tokens stored as secrets for a limited number of accounts, while in another, attackers breached its cloud provider, entered a development system, and stole customer data including billing records and some datasets. Uber says hackers took the names, addresses, birth dates, and driver's license numbers of roughly 150,000 drivers through a vendor TechCrunch identified as Fireworks. Notion says AI chat transcripts from fewer than 1% of its workspaces were exposed, and the company has since disconnected from Fireworks.

According to Zhao, open models now lag frontier models by "at most six months, probably closer to three," a dramatic narrowing from the year-to-18-month gap he would have cited a few years ago. He credits DeepSeek and "a waterfall of fantastic open models" for closing that distance. Zhao told the conference that frontier models remain a household brand, and before companies find product-market fit, using the best model is low risk—but once you pass 10,000 or 100,000 users and you're spending $10,000 to $100,000 monthly on tokens, it's time to think about open source. The company describes itself as a "specialized intelligence platform," letting businesses own their AI rather than being dependent on giant AI conglomerates.

The momentum behind open models reflects growing concerns that frontier labs absorb too much of their customers' know-how, even if only through the prompts used to get answers—a worry Microsoft CEO Satya Nadella raised in July. Zhao expects "a two-model world," with frontier and open models running in parallel: you want the best intelligence for high-level planning, but you don't need a frontier model for basic arithmetic. That's why there's so much interest in routing, and why people now think about cost per task, not just cost per token, he explained. The category is attracting massive capital: rival companies Together AI and Baseten have announced large rounds in recent months, and on Wednesday, Palo Alto Networks said it will buy Applied Compute—a startup founded just a year ago by former OpenAI researchers that, like Fireworks, helps companies train and run custom AI models on their own data—for $4.5 billion in stock.

Looking ahead, Zhao said he expects AI usage to reach "a completely different scale" in 10 years, requiring an incredible amount of capital, which is why these companies are going public. He acknowledged that every company is probably seeing an unprecedented frequency of attacks, and noted that a lot of the motivation behind Fireworks' incident was LLM jacking: hunting through providers for API keys and secrets to sell tokens on illicit marketplaces. The company promises zero data retention for inference traffic, and there's no indication that has been compromised, he said, though Notion's exposed data came from batch prediction, where you upload a dataset stored at rest—not a zero-retention product by nature. Fireworks is moving those services to an online flow so it keeps data only as long as needed. The open-model pitch hinges on a simple trade-off: accept a few months' lag in cutting-edge capability in exchange for lower costs, faster performance, and control over your proprietary intelligence. For enterprises with established user bases and rising token bills, that bargain is starting to look compelling—provided the platform handling their data can keep it secure.