Trajectory's Series A Is $45 Million Now and Was $40 Million in August, at the Same $300 Million Valuation

Trajectory, a three-founder startup that builds a post-training platform for open-source models, has raised a $45 million Series A at a $300 million valuation, TechCrunch reported on November 2. Sequoia led. In August, Dealroom wrote up the same round as $40 million at $300 million post-money, citing The Information.

The valuation matches. The cheque size does not, and neither article explains why.

Two reports, one round

TechCrunch lists Nvidia, Bessemer Venture Partners and Bain Capital Ventures alongside Sequoia. Dealroom's August list is Sequoia, Nvidia and Bessemer, with no Bain Capital Ventures. TechCrunch labels $300 million a valuation without saying pre or post; Dealroom says post-money.

Item	Dealroom, August 11	TechCrunch, November 2
Round size	$40 million	$45 million
Valuation	$300 million post-money	$300 million, unlabelled
Lead	Sequoia Capital	Sequoia
Others	Nvidia, Bessemer	Nvidia, Bessemer, Bain Capital Ventures

The simplest reading is that the round was extended by $5 million, with Bain Capital Ventures joining after the first report. That is the author's inference. If it was a $45 million cheque at a $300 million post-money from the start, Dealroom's figure, which rests on The Information, was low.

What the round pays for

Dealroom puts the preceding seed at $15 million at a $115 million post-money, announced about two months before the Series A, with Conviction, Bessemer and Radical Ventures among the investors. TechCrunch's piece does not mention it. Going from $115 million to $300 million is a step-up of about 2.6 times in roughly two months.

1. Seed (Dealroom)115 $M
2. Series A (both)300 $M

Source: Dealroom, August 11, 2026; TechCrunch, November 2, 2026

The company has about 18 employees, per TechCrunch, which works out to roughly $16.7 million of valuation per head and $2.5 million of new capital per head. It was founded about a year before the announcement by three Stanford classmates. Ronak Malde and Michael Elabd came from Google DeepMind, Malde after the company acqui-hired him and others from Windsurf, and Arjun Karanam worked on generative foundation models at Apple.

The product and the claimed customers

The platform handles post-training with continual learning. A business connects it to an open-source model and its own applications, the platform collects performance feedback and retrains the model on specialised tasks. It is now publicly available and has an interface for nontechnical users.

TechCrunch names Harvey, Rogo, Decagon, Mercor and Nvidia as customers. It gives no usage numbers, revenue or pricing, and no benchmark of a retrained model against the base model. For a company selling improvement, the absence of a before-and-after figure is the first thing a buyer will ask for.

Malde told TechCrunch the economy values someone who "worked on the job for 30 years", unlike a highly capable new hire who is always starting over. Elabd argued that keeping humans in the training loop is safer than autonomous self-improvement.

The competition and the comparison

TechCrunch lists Thinking Machines Lab, which makes the Tinker fine-tuning platform, Applied Compute, valued at $3.25 billion last month per the article, and Prime Intellect. Against Applied Compute, Trajectory's $300 million is about a tenth.

For how a quality-assurance agent company reported its own recall number, see Momentic's launch benchmark, and for how a hot round gets priced against revenue, see Kalshi at ten times revenue.

What would settle it

A Form D or a Delaware filing would show the amount sold and the investors. Until then the defensible figure is $40 to $45 million, and a reader citing one should say which outlet and which month.