Introduction
While Anthropic's public offering dominated headlines, a different conversation was unfolding at the Slush startup conference in Helsinki. There, Fireworks AI co-founder Chenyu Zhao made the case that most businesses don't need to pay premium prices for frontier artificial intelligence. Instead, he argued, open-weight models can deliver comparable results at a fraction of the cost—provided companies are willing to customize them with their own data.
What Happened
Fireworks AI operates as an inference and fine-tuning platform for open-weight models. Rather than selling general intelligence like Anthropic or OpenAI, the company helps organizations such as Uber, Shopify, and Notion run smaller, specialized models on their own infrastructure. Zhao told the audience that open models now trail frontier systems by only three to six months, a gap that has narrowed significantly since the release of DeepSeek and other open alternatives.
The timing of Zhao's remarks was notable. Anthropic completed what was described as the largest IPO on record the same day, underscoring enormous investor appetite for proprietary AI labs. Yet Fireworks is pitching a different vision: one where companies own their AI stack rather than relying on a handful of dominant providers.
Fireworks has also faced recent setbacks. The startup disclosed two security incidents this fall. In the first, attackers accessed API keys and tokens for a limited number of accounts. In the second, intruders breached a cloud provider, moved into a development environment, and obtained customer data including billing information and some datasets. Notion said fewer than 1% of its workspaces had AI chat transcripts exposed, and the company has severed its connection with Fireworks. Uber said hackers stole personal information—including names, addresses, birth dates, and driver's license numbers—for roughly 150,000 drivers through a vendor identified by TechCrunch as Fireworks.
Why This Matters
The debate over frontier versus open models is reshaping enterprise AI budgets. Frontier models carry high token costs, and Zhao noted that spending becomes difficult to ignore once a company reaches 10,000 to 100,000 users and monthly bills climb from $10,000 to $100,000. At that scale, he said, it makes sense to evaluate open-source alternatives.
There is also a strategic concern beyond price. Microsoft CEO Satya Nadella warned in July that companies using proprietary models risk paying twice: once in fees and again in lost know-how. Fireworks positions itself as an alternative where businesses retain control over their data and models. Zhao described a "two-model world" in which frontier systems handle complex planning while open models manage routine tasks, shifting the conversation from cost per token to cost per task.
The security incidents add urgency to that pitch. Customers want assurance that their data won't be retained or exposed. Fireworks says it maintains a zero-data-retention policy for inference traffic, though it acknowledged that batch prediction jobs store data at rest and is moving those workloads to an online flow to limit retention.
Key Takeaways
- Fireworks AI helps enterprises run and fine-tune open-weight models, offering a cheaper alternative to frontier systems from Anthropic and OpenAI.
- Open models now lag behind leading proprietary systems by roughly three to six months, according to co-founder Chenyu Zhao.
- The company counts Uber, Shopify, and Notion among its customers, though Notion has disconnected following a security incident.
- Fireworks disclosed two security breaches this fall involving API keys, tokens, and customer billing data.
- Investors valued Fireworks at $17.5 billion in the summer, and the startup is reportedly in early talks for additional funding.
- Broader market activity includes large rounds for rivals Together AI and Baseten, plus Palo Alto Networks' $4.5 billion all-stock acquisition of Applied Compute.
- Nvidia has backed Fireworks since 2024, and Zhao said the chipmaker remains committed to open-source AI despite reports that Jensen Huang left the company last week.
Conclusion
Fireworks AI is betting that the era of universally expensive frontier models is ending for most enterprise customers. By combining open-weight models with customer-specific fine-tuning, the company says it can deliver faster, cheaper AI without locking businesses into a single provider's ecosystem.
That proposition faces real hurdles. Recent security incidents have shaken confidence among some customers, and rebuilding trust will require transparency and stronger safeguards. Still, with a $17.5 billion valuation and backing from investors including Nvidia, Fireworks is positioning itself at the center of a shift toward specialized, cost-conscious AI infrastructure. Whether enterprises embrace open models at scale may depend less on raw model performance and more on price, privacy, and who ultimately controls the technology.

Discussion
Join the conversation
Thoughtful reactions, questions, and follow-up ideas help shape the next story.