Engram emerged from stealth on June 23 with $98 million in seed funding. General Catalyst and Kleiner Perkins co-led the round, with Sequoia Capital, Factory, Modern, Amplify Partners and Neo participating. Angel investors include Andrej Karpathy, Wiz co-founder Assaf Rappaport and roboticist Pieter Abbeel. Co-founder Jack Morris is the AI researcher best known for work on training-data extraction from language models. Notion, Harvey and Microsoft are named as launch customers.
The pitch is structural, not cosmetic. Most enterprise deployments today bolt retrieval-augmented generation (RAG) onto a frontier model and pay for the long-context tokens every time someone asks a question. Engram instead trains a compact, organization-specific memory representation in advance, then plugs it into a base model at inference time. The company claims that approach matches or beats frontier models on customer-specific tasks while using 1–10% of the tokens — a one-to-two-order-of-magnitude cost reduction that, if it holds at scale, changes the unit economics of enterprise AI.
The funding context matters too. Series A rounds for AI infrastructure startups are averaging around $52 million in 2026 — roughly 30% above non-AI peers — and seed rounds at this size remain rare. Engram is also entering a crowded category: AI memory and persistent context is now where Cognition, Letta, Mem0 and at least a half-dozen other startups are competing, on top of the in-house memory layers that OpenAI, Anthropic and Google have all rolled into their consumer products in the last six months. Customers like Notion and Harvey are credible signals; converting them to revenue is the real test.
A takeaway for learners: the cost frontier in enterprise AI has moved from "can the model do the task" to "how few tokens does it take." If you are studying NLP, the most useful work right now isn't pushing benchmark numbers up another point — it's understanding inference economics, retrieval, distillation and memory architectures. Those are the skills hiring managers at infra startups are actively short on.