Artificial intelligence startup Engram has raised $98 million in fresh funding as investors increasingly bet that the next frontier in AI is not simply building larger models, but making them more efficient.
The eight-month-old startup announced on Thursday that the funds come from a powerhouse group of investors that includes venture capital firms General Catalyst, Kleiner Perkins, and Sequoia, as well as AI researcher and OpenAI co-founder Andrej Karpathy, who recently joined Anthropic.
The funding round highlights a growing belief among investors that the next major opportunity in artificial intelligence may lie in helping companies reduce the costs associated with deploying AI at scale. "There's this explosion of data, explosion of cost," Leigh Marie Braswell, a partner at Kleiner Perkins, told CNBC. "Engram comes in and basically maps out your organization and offers orders of magnitude cheaper output."
The startup positions itself as a solution to a problem many companies are only beginning to confront. While powerful AI models from companies such as OpenAI, Anthropic, and Google continue to improve, they also require significant computing resources to process requests, especially when handling large amounts of organizational data and context.
Engram says it can dramatically reduce those costs by creating what it calls a "learned memory" layer for artificial intelligence systems. Rather than repeatedly processing the same information, the company's technology is designed to remember organization-specific workflows, institutional knowledge, and context, allowing AI systems to generate responses using far fewer tokens.
Tokens are the basic units AI companies use to measure and bill for model usage. Every question, response, document upload, and contextual reference consumes tokens, meaning businesses that rely heavily on AI can quickly face substantial operating expenses.
According to Engram, its models can match or outperform leading AI systems while using up to 100 times fewer tokens. While those claims have not been independently verified, the promise of dramatically lower AI costs appears to have resonated with both investors and customers.
Less than a year after its launch, the 13-person company says it has already attracted clients including Microsoft, Notion, and legal AI startup Harvey. The company's name comes from the neuroscience term "engram," which refers to a physical trace of memory stored in the brain.
Biderman told CNBC that his fascination with memory started during childhood while interacting with his grandmother, who had lost much of her memory. He would try to get her to recall facts about him and his siblings, an experience that eventually inspired him to pursue a doctorate in computational neuroscience at Columbia University.
He later worked at the AI laboratory at Stanford University, where he began studying what he calls the "genius stranger" problem in artificial intelligence. The concept describes AI models appearing extraordinarily intelligent while lacking a deep understanding of the people and organizations using them.
Adding more context can help, but doing so often increases costs and can overwhelm models with information. "We're trying to go beyond this existing note-taking and build this layer of intuition that humans have, and current models don't," Biderman said.
