A 13-person startup just got a $600M valuation because your AI bills are out of control

Engram exited stealth Tuesday with $98M in funding.

13 employees. $600 million valuation.

The math only makes sense if you understand the problem they're solving.

Enterprise AI budgets are hemorrhaging. Uber burned through its entire 2026 AI budget by April. Meta employees burned 73.7 trillion tokens in a single month before the company imposed centralized spending controls.

The root cause isn't token pricing. It's architectural.

Every time your AI agent processes a document, it rebuilds its understanding from scratch. A 70,000-word contract produces a KV cache exceeding 100 gigabytes of GPU memory. 250,000x inflation of the original document. Every single query.

Engram's fix: train a compressed memory offline on your organization's corpus. Load it at inference time instead of rebuilding context from scratch. They claim 100x fewer tokens with matching or better performance.

Microsoft is testing it inside M365. Notion is piloting it in custom agents. Harvey is exploring it for legal document environments.

General Catalyst, Sequoia, and Kleiner Perkins led the round. Andrej Karpathy angel invested.

Your CFO is about to ask why you're paying frontier model prices for tasks that repeat across the same institutional knowledge.

Audit your token spend by use case. If your agents interact repeatedly with the same documents, you're overpaying by orders of magnitude. The infrastructure layer that fixes this just got funded.

SOURCE: https://www.techtimes.com/articles/319106/20260625/enterprise-ai-token-costs-engram-exits-stealth-98m-claims-100x-cut.htm

VERIFIED: TechTimes, PR Newswire, Yahoo Finance, Calcalist

SIGNAL: Enterprise AI token costs have become a board-level crisis. Engram's $600M valuation on 13 employees signals that investors believe the cost structure problem is worth more than the models themselves.