Engram Raises $98M to Build AI That Never Forgets

Eight-month-old Engram secures $98 million from Sequoia, Kleiner Perkins, and General Catalyst to deploy a learned-memory layer that slashes enterprise token costs by up to 100 times.

Lead

What Happened

Engram disclosed the raise as the broader venture market continues to direct outsized capital toward AI infrastructure bets. The round is among the largest per-employee funding events in the current AI cycle and follows a founding date of October 2025, meaning Engram attracted one of the industry's marquee investor consortiums within its first eight months of operation.

The participation of Karpathy — whose credibility spans both foundational model research and consumer AI deployment — adds a technical validation signal that institutional investors typically weigh alongside financial metrics.

The Technology

Engram positions itself as the cognitive alignment layer for enterprise AI. Its core assertion is that most large language model deployments are inefficient because models must re-process organizational context from scratch on every query. Engram's approach stores and retrieves organization-specific workflows, terminology, and interaction history as a persistent memory substrate, allowing models to anticipate questions and deliver contextually richer responses without repeatedly ingesting the same background information.

CEO Dan Biderman, who completed a PhD in computational neuroscience at Columbia University before joining Stanford University's AI lab, frames the product through a neuroscience analogy: the way biological memory enables efficient cognition without re-learning foundational facts on each task.

Strategic Context

The funding arrives as enterprises grapple with AI inference costs that remain a primary barrier to broad deployment. Token pricing from frontier providers has fallen sharply over the past 18 months, yet high-frequency enterprise use cases — legal research, customer support, workflow automation — still generate substantial compute bills at scale.

Engram's client roster illustrates the demand. Microsoft (MSFT), the largest single backer of OpenAI and a central hub of enterprise AI adoption, is already a customer. So is Notion, whose productivity platform serves millions of knowledge workers, and Harvey, the legal-AI startup that has become one of the highest-profile vertical AI deployments among AmLaw 100 firms. A roster of this caliber, assembled before the company's first birthday, reflects how urgently buyers are seeking cost controls without sacrificing model capability.

The backing from all three of Sequoia, Kleiner Perkins, and General Catalyst simultaneously is unusual. Each firm has competing portfolio companies in the AI infrastructure space, suggesting Engram's technical differentiation was sufficient to override standard conflict concerns.

What Comes Next

With $98 million in fresh capital and a team of 13, Engram faces a clear build-or-hire decision. The company will likely need to expand engineering and applied-research capacity to support enterprise onboarding at the scale its client list implies. The central risk is replication: major cloud providers and frontier labs have the incentive and resources to internalize memory optimization natively into their model pipelines, potentially commoditizing the layer Engram occupies.

Outlook

Engram's $98 million raise underscores a maturing investor thesis: that the next frontier in enterprise AI is not model capability but economic efficiency. The startup's learned-memory architecture addresses a real and measurable pain point, and its early traction with Microsoft, Notion, and Harvey validates product-market fit at meaningful scale. Whether a 13-person team can build and defend a proprietary memory layer before hyperscalers close the gap will define Engram's long-term trajectory.