Product Designer

Fireworks AI seeks new funding, targeting a $30-40 billion valuation. Investor appetite for the infrastructure underpinning artificial intelligence continues to surge, as Fireworks AI seeks to raise new capital at a valuation between $30 billion and $40 billion, according to people familiar with the matter. This potential jump comes only months after the company's last fundraising round in July, which valued the startup at $17.5 billion. Founded in 2022 by veterans from Meta and Google's AI divisions, the San Francisco-based company has quickly established itself as a key player in the competitive world of AI infrastructure. Fireworks AI develops and rents out the computing systems and software tools that businesses use to run, fine-tune, and deploy both open-weight and custom AI models. The company distinguishes itself by focusing on what it calls "specialized intelligence," tailoring AI models to clients' proprietary data, workflows, and specific use cases, rather than building a consumer-facing chatbot to rival ChatGPT. Demand for AI infrastructure solutions has soared as companies seek to harness generative AI technologies. In July, Fireworks AI revealed that it had surpassed $1 billion in annualized revenue, serving more than 40 trillion tokens per day, with 95 percent of activity attributed to customer-specific models. The firm's clientele ranges from startups fine-tuning open models to larger enterprises integrating AI directly into their operations. Fireworks has become a magnet for top-tier investors. Its $1.5 billion Series D round this July was led by Atreides Management, Index Ventures, and TCV and included participation from Nvidia, Lightspeed Venture Partners, Bessemer Venture Partners, and Menlo Ventures. The prior Series C, in late 2025, brought in $250 million at a $4 billion valuation, co-led by Lightspeed, Index, and Evantic. The current funding talks come as Fireworks AI has been navigating the aftermath of a security incident. Earlier this month, the company reported discovering unauthorized use of internal credentials that allowed an intruder to access data from some customer accounts. Chief executive Lin Qiao, who previously led the PyTorch team at Meta, wrote on X that the company is committed to transparency and to turning the episode into "the most formidable security response in AI Infra." Fireworks declined to comment further on both the security matter and the fundraising process. Fireworks operates in a fiercely competitive sector, facing challenges from other model-infrastructure startups such as Together AI, Baseten, and Replicate, as well as from tech giants like Amazon Web Services, Google Cloud, Microsoft Azure, and Oracle. The company's plans for a new funding round, first reported last month by The Information, have not been finalized, and valuation targets may still change as discussions continue. Still, the staggering jump in valuation highlights the ongoing excitement surrounding AI infrastructure providers, even as operational and security challenges persist. Investors are betting that companies like Fireworks AI will remain critical to the booming adoption of artificial intelligence across industries. Technology

Fireworks AI is in talks to double its valuation to 40 billion dollars. Fireworks AI processes over 40 trillion tokens daily and crossed $1 billion in annualized revenue, up fivefold in a year, as Nvidia, Index Ventures and TCV back its rapid valuation climb. An inference startup most people outside AI circles have never heard of is closing in on a valuation that would put it ahead of companies that actually build the models. Fireworks AI is a Redwood City startup. It helps companies run AI models cheaply and fast, and it's now in early talks to raise new funding at a valuation between $30 billion and $40 billion, Business Insider reported on October 20. That's roughly double what the company was worth just three months ago. Back in July it closed a $1.505 billion Series D at a $17.5 billion valuation, led by Atreides Management, Index Ventures and TCV, with Nvidia joining as a follow-on investor. Founder Lin Qiao isn't new to this stack. She left Meta in 2022, where she led more than 300 engineers building and scaling PyTorch, the open-source framework that still trains and runs a large share of the world's AI models. She started Fireworks in Redwood City a month before ChatGPT launched, betting that whoever controlled the plumbing between a trained model and a paying customer would matter as much as whoever trained the model in the first place. That bet is paying off. Fireworks now processes more than 40 trillion tokens a day, up from roughly 15 trillion a year ago, and has crossed $1 billion in annualized revenue, about five times what it was generating last year. More than 95% of that traffic comes from models customers have fine-tuned on their own proprietary data, not off-the-shelf open weights. That's the part of the business Fireworks actually sells: leased GPU capacity plus the tooling to fine-tune and serve a model in production - not a frontier model of its own. Baseten closed a $1.5 billion Series F in June at a $13 billion valuation, barely five months after a $300 million round valued it at $5 billion. Together AI, another rival chasing the same enterprise inference workloads, closed an $800 million round at an $8.3 billion valuation after crossing $1 billion in annualized revenue of its own. The pattern across all three is the same: valuations are stepping up two, three, sometimes four times in a matter of months, far faster than revenue is actually growing underneath them. Fireworks AI closed a $1.505 billion Series D at a $17.5 billion valuation, nine months after a $4 billion Series C. The inference startup founded by ex-Meta PyTorch lead Lin Qiao has crossed $1 billion in annualized revenue, a fivefold jump year-over-year, and now processes 40 trillion tokens per day for customers including Cursor, Uber, and... - fireworks AI series D funding valuation - enterprise AI cheaper alternatives frontier models Here's the thing worth sitting with. Fireworks, Baseten and Together don't train frontier models, and they don't have the kind of proprietary data or model weights that create a durable moat. What they sell, fast and cheap model-serving, is increasingly commoditized, and the hyperscalers (AWS, Google Cloud, Microsoft Azure) all offer some version of it too. A customer who switches from Fireworks to Baseten loses very little. That's not a reason to assume these companies collapse. It's a reason their valuations should track revenue multiples closer to infrastructure resellers than to AI labs, and right now they aren't. Citadel's Ken Griffin has been the most prominent voice naming the mismatch directly. Speaking at the Future of Global Markets conference, Griffin said the AI boom's real payoff could take decades rather than years. The dream "could be 20 years... it can be 30 years" before the technology's value gets properly sorted between winners and losers, he warned, much like the dot-com crash separated Amazon from Pets.com. He wasn't talking about Fireworks specifically. But a 230% valuation jump in three months for a company that leases someone else's GPUs is exactly the kind of math his warning describes. None of this means the money is wrong to chase inference. Every major AI lab needs somewhere to actually run the models it trains, and that demand is real and growing. Nvidia followed its own Series D money into Fireworks as a strategic investor, rather than just selling it chips. That says the chipmaker sees the inference layer as a long-term fixture, not a bubble play it wants to avoid. But a strategic investor backing a company for supply-chain reasons is a different signal than a pure financial investor paying 40 billion dollars for a business that reported roughly 1 billion in revenue months earlier. Fireworks declined to comment on the specific terms of the talks, and Business Insider's report noted the round has not been finalized. If it closes anywhere near that range, Fireworks will be worth more than most publicly traded enterprise software companies, with a fraction of the public scrutiny. The rest of the inference sector will have a new ceiling to measure itself against. Join the discussion. No replies yet. Start the discussion. Janet Harrison has over 16 years experience in the financial services industry giving her a vast understanding of how news affects the financial markets, and an early adopter of blockchain technology and digital currencies. Janet is an active holder and trader spending the majority of her time analyzing blockchain projects, reports and watching new and upcoming projects and other initiatives in the industry. She has a Masters Degree in Economics with previous roles counting Investment Banking.

Fireworks pitches one-click loop, proof is thin. Fireworks AI pitches a train-to-deploy loop to replace closed models, but key speed and quality wins lack baselines and pricing. AI Engineer host Jatashri, who leads applied machine learning at Fireworks AI and goes by Jet, pitched a train-to-deploy loop that moves teams off rented closed models and onto tuned open models that stay on Fireworks for both training and inference. Stop renting intelligence was the hook. Jet framed three claims as current reality: open models are production ready, citing GLM 5.2 for coding; closing the quality gap depends on customization; and teams now want a continuous train, host, retrain flywheel instead of one-off serving. Fireworks says it serves about 30 trillion tokens and, pre-GLM 5.2, expected that to rise by about 5 trillion tokens per day, with more than 10,000 companies and about 200,000 requests per second in production. The talk cited Cursor's Composer 2 trained and hosted on Fireworks with 13x faster inference via its Fire Attention engine, Vercel with a 40x latency improvement, and GenSpark with a 12% quality gain after RFT plus dataset cleaning and evaluator work. None of those figures came with datasets, baselines, or costs. Fire Attention is presented as a custom CUDA kernel layer between model and GPUs, tuned per workload on input tokens, output tokens, and prompt cache hit rate, plus techniques like speculative decoding, disaggregation, and quantization. Jet said customers can train draft models for speculative decoding on the platform the same way they submit fine-tuning data. That helps latency and throughput if the draft model is good, but it ties speed gains to workload-specific kernel work that Fireworks controls. Training comes in three flavors. The Training Agent in the UI takes linked datasets and a task description, picks a base model, cleans data, and selects hyperparameters, then reports steps, accuracy and F1 and recommended next steps. Managed training via API or UI requires you to choose model, datasets and hyperparameters and click Create, with a dashboard for loss by epoch and an export to Weights & Biases. The Training API lets ML engineers define their own loop and loss. All three support supervised fine-tuning, reinforcement fine-tuning and DPO, except the agent does not support RFT today, including its data cleaning and hyperparameter help. Jet acknowledged teams often need help writing evaluator functions for RFT. The promised flywheel is one click from a new checkpoint to live deployment. Jet pointed to hotloading built for Cursor that swaps weights with no downtime, claims of numerical fidelity by keeping training and inference on the same stack to avoid cross-platform precision drift, and GPU supply via cloud and neocloud partners. Fireworks also says it has raised about $327 million and has about 200 people. What is still gated is pricing for the tuned kernels at scale, portability if you leave, and any independent reproduction of the Cursor, Vercel and GenSpark wins. Until Fireworks shows the evaluators, the data cleaning steps, and the latency measurements behind those 13x and 40x claims, the loop is an attractive operations story with the proof held inside the platform. (C) 2026 StartupHub.ai. All rights reserved. You may not republish this article in full without a license. Search engines and AI research tools may crawl and summarize for reference. Bulk reproduction or model training requires a license. See our terms. Discussion. Join the conversation below Editor, StartupHub.ai Daniel Singer is the editor of StartupHub.ai, a technology expert and thought leader on AI and its applications across sectors, from fintech and healthcare to developer tooling and consumer software. He writes and tests the tools covered here thoroughly and regularly, and built StartupHub.ai to give founders, operators and buyers a clearer read on what they are actually being sold.