A company that has been operating for less than a year, employs just thirteen people, and has never appeared on a technology industry ranking of any kind raised $98 million from some of the most respected names in venture capital on Tuesday. The amount it raised matters less than the problem it is trying to solve, and that problem is one that companies of every size are now scrambling to get under control.
Engram, founded in October last year by Dan Biderman and his wife Natalie, builds software that helps organisations dramatically reduce the amount they spend when running large computer programmes that process text, answer questions or handle complex tasks. The funding round was led by General Catalyst and included Kleiner Perkins, Sequoia and Andrej Karpathy, a co-founder of OpenAI who left to pursue independent research.
The investor group is significant. These are firms and individuals that have backed some of the most consequential technology companies of the past two decades. Their willingness to put nearly $100 million into a company with fewer staff than most small restaurants reflects the scale of the underlying problem.
To understand what Engram does, it helps to understand how large language programmes charge for their services. Rather than billing by the hour or by the question, these programmes measure usage in units called tokens, which correspond roughly to fragments of words. Every time a company uses one of these programmes to do something, it sends a package of context, instructions and background information along with its request. The bigger that package, the more tokens are consumed and the higher the bill. For a company running thousands of such requests every day, those costs add up with remarkable speed. Several large organisations have begun restricting how much their staff can use these tools precisely because the bills have become difficult to justify.
Engram proposes that most of what gets sent in those packages is unnecessary. When a programme is asked to help with a task, it does not need to be fed every document, email and note that might conceivably be relevant. What it needs is the specific information that is actually relevant to that particular task, retrieved quickly and precisely. Engram builds what it describes as a memory layer, software that sits between the organisation and the large computing programme it uses, figures out what information that programme actually needs in any given moment, and sends only that. The company claims that in some cases this approach can reduce the number of tokens consumed by a factor of between ten and one hundred, while matching or exceeding the quality of results.
Dan Biderman arrived at this problem through an unusual route. His interest in memory began during childhood, when he spent time with his grandmother who had lost much of hers. He would try to help her recall facts about family members, an experience that stayed with him. He went on to complete a doctorate in computational neuroscience at Columbia University and later worked at a research laboratory at Stanford, where he began thinking about what he calls the genius stranger problem: the phenomenon of sophisticated computer programmes appearing remarkably capable while knowing almost nothing specific about the people or organisations using them. Adding background information can help bridge that gap, but loading in too much information at once creates its own problems and drives costs up further.
“We’re trying to go beyond note-taking and build a layer of intuition that humans have, and current models don’t,” Biderman told CNBC. That framing positions Engram not merely as a cost reduction tool but as an attempt to give large computing programmes a form of organisational common sense.
Early customers suggest the pitch is landing. Microsoft, the note-taking platform Notion and the legal technology company Harvey are all listed among Engram’s signed clients. For companies of that scale, even a modest reduction in computing costs would translate into savings of millions of dollars a year. An order of magnitude reduction, if Engram’s claims hold up at scale, would be transformative.
Leigh Marie Braswell, a partner at Kleiner Perkins, described the underlying market opportunity plainly: “There’s this explosion of data, explosion of cost. Engram comes in and basically maps out your organisation and offers orders of magnitude cheaper output.” One of the other investors, Nimrod Bahat, put it in similarly direct terms: “The growing adoption of software agents in 2026 makes it clear that the industry faces a real cost problem. Users want more capabilities, there is a genuine shortage of computing power, and tokens remain expensive.”
Biderman noted that the cost problem had not been acute when Engram first launched. At that point investors were drawn primarily by the founders’ academic credentials. It was only when newer, more capable programmes were released at the start of this year that token consumption jumped sharply across the industry. Suddenly efficiency, which had been a nice idea, became an urgent commercial priority. The timing has worked in Engram’s favour.
The company plans to use the fresh capital to hire additional engineers and expand the computing infrastructure it uses to train and run its own models. With a valuation of around $600 million on the back of a single funding round, and a customer list that includes some of the most recognisable names in technology, the stakes for delivering on those claims are now considerable.
About the Author
Marcel Chidozie is a tech analyst and writer covering foreign news, fintech, and emerging technologies at TechRegard. Based in Nigeria, He's passionate about translating complex tech developments into compelling, accessible stories for diverse audiences. His work focuses on how technology shapes innovation across Africa and globally.
