Pomegra
Sequoia led Trajectory's $45M Series A, valuing the 18-person startup at $300M, to let lawyers, doctors and other non-engineers retrain AI models.
Key Takeaways
- Trajectory closed a $45M Series A led by Sequoia Capital at a $300M valuation in November 2026.
- The round follows a $15M seed from May 2026, bringing total funding to $60M.
- Nvidia, Bessemer Venture Partners and Bain Capital Ventures joined; Harvey, Rogo, Decagon, Mercor and Nvidia are customers.
Lead
Trajectory, a San Francisco startup founded by three Stanford classmates, raised a $45M Series A led by Sequoia Capital at a $300M valuation, the company announced in early November 2026. Nvidia, Bessemer Venture Partners and Bain Capital Ventures also took part. The round arrives about six months after a $15M seed, so the company has raised $60M in total with 18 employees.
The product is a platform that automates post-training of AI models. A company starts with an open-source model, connects it to a live application, and the system gathers feedback on how the model performs and retrains it. The interface targets domain experts, such as attorneys and physicians, rather than machine learning engineers.
What Does Trajectory Actually Sell?
Trajectory sells a post-training platform that turns day-to-day expert feedback into model updates, without requiring a team of engineers. Chief executive Ronak Malde has compared foundation models to a very capable PhD student who is always starting a new job. The company's pitch is that the student should learn from each workday instead of starting fresh.
Customers listed at launch include Harvey, Rogo, Decagon, Mercor and Nvidia. The platform is now publicly available. Co-founder Michael Elabd has argued that keeping humans in the loop is safer than fully autonomous self-improvement, which sets the product apart from agent systems that rewrite themselves without review.
Who Is Behind It and Who Is Paying?
Three Stanford classmates started the company: Malde, Elabd and Arjun Karanam. Malde and Elabd previously worked at Google DeepMind, and Karanam worked on generative foundation models at Apple. The small team also includes hires from OpenAI, Stripe and Netflix.
The May seed was led by Conviction, with Bessemer, Radical Ventures and BoxGroup participating. Angels included Google's Jeff Dean and Stanford's Fei-Fei Li, along with founders of Notion, Dropbox, Hugging Face and Braintrust. Bessemer has now backed both rounds, while Sequoia arrived as the new lead.
Why Does a $300M Valuation Matter After Six Months?
The valuation matters because it shows investors pricing a company with no disclosed revenue on customer names and team pedigree. The company has not published revenue, contract sizes or retention figures, and the use of the new funds was not specified. A $300M mark on $60M raised is a steep step-up from a seed stage that closed only half a year earlier, though the seed valuation was not disclosed.
The customer list is the strongest evidence in the company's favor. Harvey and Rogo sell to law and finance, where expert judgment is the scarce input, and Mercor runs a marketplace of expert contractors. Nvidia appears as both investor and customer, a pairing that tends to flatter traction numbers because the buyer also has equity at stake.
How Does It Fit Into the Post-Training Market?
Trajectory competes in the crowded market for fine-tuning and reinforcement learning tooling, where the usual buyer is an engineering team. Its bet is that the person judging output quality, rather than the person writing training code, should control the loop. That reduces dependence on scarce ML talent, but it also shifts risk onto data quality, since expert feedback is noisy and expensive.
The strongest counterargument is that frontier labs keep improving general models quickly, which can erase gains from narrow tuning. If a base model's next release matches a tuned open-source model, customers may question the retraining spend. Trajectory's answer is that proprietary feedback compounds in ways a generic release cannot copy, but that claim has no public benchmark yet.
Outlook
Trajectory now has $45M in fresh capital, a Sequoia-led cap table and a public product, but it has not yet shown revenue. The next test is whether named customers expand usage and whether non-engineers can produce measurable model gains at scale. Competition from fine-tuning vendors and from frontier labs' own customization tools will shape how much of the $300M valuation holds up in a later round.
