AI Memory Startup Engram Raises $98 Million to Cut Token Costs for Enterprises

An 8-month-old artificial intelligence startup called Engram raised $98 million on Tuesday, drawing backing from some of the technology industry's most prominent investors as companies face mounting pressure to rein in runaway AI spending.

The round was led by General Catalyst, Kleiner Perkins, and Sequoia, and also included OpenAI co-founder Andrej Karpathy, who recently joined Anthropic, the company said.

Engram positions itself as a "learned memory" layer for AI systems, designed to recall organization-specific workflows and context in order to anticipate queries and return smarter responses at lower cost. The company claims its models can match or outperform frontier labs using up to 100 times fewer tokens — the unit of measurement used to price AI queries.

The funding comes as corporate America grows increasingly focused on controlling AI expenditure. New and more sophisticated models have proven costlier than earlier generations, upending the long-held assumption that greater scale would naturally drive costs down.

"You've got this explosion of data, explosion of cost," said Leigh Marie Braswell, a partner at Kleiner Perkins. "Engram comes in and basically maps out your organization and offers orders of magnitude cheaper output."

Despite its brief existence, the 13-person company has already signed on clients including Microsoft, Notion, and legal AI startup Harvey. Engram plans to use the proceeds to expand its computing capacity and grow its workforce.

The company was co-founded by Dan Biderman, who serves as CEO. Biderman holds a PhD in computational neuroscience from Columbia University and later worked at Stanford University's AI lab, where he began developing what he calls the "genius stranger model" — the observation that AI systems are capable but operate with far more limited memory than their apparent intelligence might suggest.

Biderman traces his fixation with memory to childhood, when he tried to help his grandmother, who had lost her memory, recall details about him and his siblings.

He acknowledges that Engram's models are not universally superior to those from OpenAI or Anthropic. "We're trying to go beyond this existing notetaking and build this layer of intuition that humans have, and current models don't," Biderman said.

The startup's name draws from neuroscience — an engram refers to a physical or biochemical trace of memory in the brain.

The raise underscores a broader shift in how enterprises are evaluating AI vendors, with token efficiency and cost containment rising alongside raw capability as procurement criteria. As frontier model providers continue to release increasingly expensive offerings, startups that can deliver narrower, cheaper, domain-specific performance may find a growing market among cost-conscious customers.

Disclaimer