Trajectory Raises US$ 45 Million to Democratize Continuous AI Model Training

Trajectory announced on Monday (November 2) a $45 million Series A funding round led by Sequoia Capital, with participation from NVIDIA, Bessemer Venture Partners, and Bain Capital Ventures. The company, founded approximately one year ago by three Stanford classmates with backgrounds at Google DeepMind and Apple, reached a $300 million valuation — a figure that, while far from the $3.25 billion valuation of competitor Applied Compute, signals investor confidence in an underexplored layer of the AI stack: post-training.

The problem Trajectory seeks to solve is a bottleneck shared by virtually all modern AI models, whether produced by OpenAI, Anthropic, or open-source projects. Foundation models do not learn from their mistakes without additional training. Companies adapting these models for specialized fields — law, finance, medicine — typically get around this limitation through fine-tuning, but that practice requires internal research teams or expensive consultants, known in the industry as forward-deployed engineers (FDEs).

"Foundation models are like a very smart PhD student who always starts a new job," says Ronak Malde, co-founder and CEO of Trajectory, in an interview with TechCrunch. "But what the economy values is someone who has worked at the job for 30 years." Trajectory's answer lies in what it calls "continual learning": instead of relying on researchers to build new models from scratch, Trajectory allows domain experts — lawyers reviewing contracts, financial analysts valuing companies — to feed the system feedback on model performance and automatically retrain it.

The platform, launched publicly on the same day as the announcement, allows teams to start from an open-source model, connect it to their applications, and collect feedback on how the model performs in real situations. A central feature of the platform, called "Continual Grading," asks experts to rate model outputs on specific tasks — reviewing a particular type of legal contract or valuing a company using a specific methodology — and uses that grading to adjust the model in single-production training runs, using the GS1 algorithm. Engineers trigger the retraining steps, keeping human specialists in the loop.

"Keeping humans in the loop makes this safer than fully autonomous self-improvement," says co-founder Michael Elabd. "Autonomous self-improvement could produce unintended or even dangerous consequences."

Existing customers of the platform include Harvey (the AI legal startup founded by OpenAI's former CTO), Rogo, Decagon, Mercor, and Nvidia itself — a list signaling significant traction in an still-emerging market. The company, with 18 employees, has also recruited researchers and product designers from OpenAI, Stripe, and Netflix, reinforcing the thesis that the user interface is the true differentiator.

"Why does design matter so much for a company focused on AI training?" asks Arjun Karanam, the third co-founder. "We built an interface that makes post-training accessible to non-technical customers. The interface is the bridge between specialists who know what 'good' looks like and models that need to learn it."

The post-training market is becoming increasingly competitive. Thinking Machines Lab, founded by OpenAI's former CTO Mira Murati and developer of the Tinker fine-tuning platform, is a direct competitor. Applied Compute, which reached a $3.25 billion valuation just last month, and Prime Intellect, which helps companies build and train their own AI agents, complete the competitive landscape.

NVIDIA's participation in the round is particularly revealing. The company has consistently invested in AI infrastructure startups whose products expand the number of workloads suitable for their chips — not only for massive pre-training, but also for fine-tuning. As Value Add VC observes, NVIDIA "has increasingly taken stakes in AI infrastructure startups whose products expand the number of workloads suited to its chips, a pattern also visible here."

The round follows a $15 million seed round announced in May 2026, when Trajectory introduced itself as a research lab and product company focused on continual learning. With the new $45 million, the company now has $60 million in total funding and an 18-person team to scale the platform.

What makes Trajectory particularly interesting is the timing: in a market saturated with capital for foundation models, money is shifting to infrastructure below the pre-training layer. The $3.25 billion Applied Compute valuation is the clearest signal, but Sequoia itself, by leading Trajectory's round, is betting that the next big AI innovation won't come from bigger models, but from models that learn better — and faster — from end-user feedback. "Every company building on AI is facing climbing token bills, and models still miss critical context," the company states on its website. "The best AI won't only come from researchers in labs — it will come from the experts closest to the work."

The question is: in a market where Applied Compute is worth $3.25 billion and Thinking Machines Lab carries the name of Mira Murati, can Trajectory defend its position with just 18 employees and $60 million? Or will the next wave of AI investment continue to push money upward, toward foundation models, while post-training remains a niche for smaller startups?

Sources: TechCrunch, Runtime Wire, Value Add VC

✓ Independent sources cross-checked and verified before publishing