Pomegra
Trajectory, a San Francisco AI startup roughly a year old, raised $45 million from Sequoia at a $300 million valuation and launched its post-training platform on November 2, 2026.
Key Takeaways
- Sequoia led the $45M Series A at a $300M valuation; Nvidia, Bessemer and Bain Capital Ventures joined.
- Total funding reaches $60M; Harvey, Rogo, Decagon, Mercor and Nvidia are named customers.
- The platform lets non-research staff retrain open-source models on production feedback.
Lead
Trajectory announced a $45 million Series A on November 2, 2026, valuing the company at $300 million. Sequoia led the round, with Nvidia, Bessemer Venture Partners and Bain Capital Ventures participating. The financing brings total capital raised to $60 million, which implies about $15 million came before this round. The company did not disclose the source of that earlier money. The announcement coincided with the public release of a self-serve platform for post-training AI models on a company's own production data.
What Does Trajectory Actually Build?
Trajectory builds software that adjusts an AI model after its initial training so it performs better on a specific job. The platform connects to a customer's business applications, collects feedback on how a model performs real tasks, and feeds that feedback back into the model. The company calls the method "Continual Grading" and pairs it with an algorithm it names GS1, which it says lets very large models learn from deployment data.
The pitch targets fields such as law, medicine and engineering, where domain experts know what a good answer looks like but rarely write training code. Users supply task-specific judgments, and Trajectory's engineers manage the retraining steps. Co-founder Michael Elabd has argued that keeping humans in the loop makes the process safer than fully autonomous self-improvement.
Who Is Behind It and Who Is Buying?
Three Stanford classmates founded the company: Ronak Malde (CEO, formerly at Google DeepMind), Michael Elabd (formerly at Apple) and Arjun Karanam (formerly at Google). The team numbers 18, with recent hires from OpenAI, Stripe and Netflix.
Named customers include Harvey, Rogo, Decagon, Mercor and Nvidia, which use the platform to post-train open-source models. Pricing was not disclosed. Neither were revenue figures, so the $300 million valuation rests on the investor roster and early logos rather than reported sales.
How Should the Valuation Be Read?
A $300 million price for a company about a year old, with 18 employees and undisclosed revenue, is a bet on a category rather than on current financials. The $45 million check is about 15 percent of the post-money figure, assuming the valuation is post-money, a detail the announcement did not specify. That is a conventional dilution level for a Series A, so the price is high mainly because of the stage.
The named customers are a mixed signal. Harvey, Rogo, Decagon and Mercor are fast-growing AI application companies with strong incentives to tune models cheaply. Nvidia appears twice, as investor and as customer, which makes its logo less clean evidence of independent demand. The announcement describes some of these customers as testing the approach, so the extent of paid production use is unclear.
Who Else Is Chasing Continual Learning?
Trajectory enters a field with better-funded rivals. Thinking Machines Lab offers a fine-tuning product called Tinker, and Applied Compute and Prime Intellect also work on continual learning and post-training. The common argument is that frontier labs ship general models while the value in enterprise use sits in specialization on private data.
The counterargument is that the labs themselves can offer tuning as a feature, and that open-source model quality shifts quickly enough to strand specialized pipelines. Trajectory's defense is workflow: a tool that non-engineers can operate and that keeps retraining tied to live feedback. Whether that is a durable product or a feature that larger platforms absorb is the open question for the next two funding cycles.
What Comes Next for Trajectory?
The company says the money goes toward infrastructure for continual machine learning, research on its optimization algorithms, and hiring deep-technical engineers. Expect the 18-person team to grow quickly, since the round gives it a larger runway relative to headcount than most Series A companies carry.
The metrics to watch are conversion of pilot customers into paid contracts and whether the platform works on models outside the open-source families it supports today. A disclosed revenue figure at the next raise would test the valuation more directly than this round did.
Outlook
Trajectory secured $45 million at a $300 million valuation from Sequoia and strategic backer Nvidia, taking total funding to $60 million. The product addresses a real gap between general AI models and specialist work, but it faces funded competitors and possible absorption by larger platforms. Revenue disclosure and customer retention will determine whether the valuation holds up.
