Cognition closes $1B+ round at $26B valuation — Devin ARR at $492M, U.S. Army and Navy among enterprise customers
TL;DR: Cognition closed $1B+ at a $26 billion post-money valuation on May 27, 2026 — confirming the talks reported earlier this month. Co-leads: Lux Capital, General Catalyst, and 8VC. Other participants: Ribbit Capital, Atreides Management LP, Peter Thiel’s Founders Fund. The numbers underneath: Devin ARR now $492 million — up from $73M in June 2025 (a 6.7× jump in eleven months). Enterprise customer list: Goldman Sachs, Citi, Mercedes-Benz, the U.S. Army, and the U.S. Navy. Trajectory: valuation more than doubled from $10.2B in September 2025; combined enterprise ARR rose 30%+ in the seven weeks following the Windsurf close (December 2025). Cognition is now the most-valued pure-play AI coding harness company outside the frontier model labs — and the only one with confirmed Pentagon-tier deployments.
What closed
The reporting from Bloomberg, TechCrunch, SiliconAngle, and other primary sources confirms:
- Total raised: more than $1 billion
- Post-money valuation: $26 billion
- Pre-money valuation: $25 billion
- Co-leads: Lux Capital, General Catalyst, 8VC
- Additional investors: Ribbit Capital, Atreides Management LP, Founders Fund
- Date of close: Wednesday, May 27, 2026
- Status: officially closed and announced (not “in talks” — this is the close announcement)
The talks previously reported at $25B valuation closed slightly higher at $26B post-money. Time from initial report (around Apr 23, 2026) to close: roughly 35 days. That’s fast even for late-stage AI rounds.
The revenue trajectory
Devin’s ARR over the last twenty months:
| Period | Devin ARR |
|---|---|
| September 2024 | $1 million |
| June 2025 | $73 million |
| May 2026 (today) | $492 million |
That’s a 6.7× jump in eleven months from June 2025 to May 2026 — and a 492× jump in twenty months from the September 2024 starting point. The Windsurf acquisition (closed December 2025 at $82M Windsurf ARR) accelerated the combined growth: enterprise ARR rose 30%+ in the seven weeks immediately following the close.
For context, Claude Code’s ARR is reported around $2.5B+. Cognition’s $492M doesn’t match that, but it’s the only standalone harness business (i.e., not bundled inside a frontier model lab’s product line) above $400M ARR.
The enterprise customer list
The customer roster Bloomberg names is structurally important:
- Goldman Sachs — top-tier investment banking
- Citi — top-tier consumer banking + securities
- Mercedes-Benz — global automotive + industrial systems
- U.S. Army — Department of Defense
- U.S. Navy — Department of Defense
The Pentagon-tier deployments matter. Anthropic was excluded from the Pentagon’s May 1, 2026 classified-network AI procurement over its refusal to enable autonomous weapons and mass surveillance — a values stance that creates competitive opening for less-restricted vendors. Cognition appears to have moved into that opening. The exact deployment scope isn’t disclosed (likely classified), but Army + Navy as named customers signal Devin has cleared DoD vendor review.
For high-finance customers (Goldman, Citi, Mercedes), the read is similar: Devin’s autonomous-execution model — runs in isolated cloud sandboxes, opens its own PRs, requires explicit approval before sensitive actions — fits regulated-industry security models in ways that more chatty AI tools don’t.
The vertical-integration picture
Cognition is now the only vertically-integrated harness business in the AI Harnesses category:
- Model: Cognition’s proprietary SWE-1.5 (reportedly 950 tokens/sec, 13× faster than Sonnet 4.5 on coding tasks)
- Autonomous agent: Devin — runs in cloud sandboxes, opens PRs
- IDE: Windsurf — acquired December 2025 for ~$250M; now runs on SWE-1.5 with native Devin handoff
Every other major harness vendor sits downstream of Anthropic / OpenAI / Google models. Claude Code is Anthropic’s. Antigravity is Google’s. OpenAI Codex is OpenAI’s. Cursor is multi-model but doesn’t own its models. Cognition is the first to bundle model + agent + IDE in a single company at scale.
This is why the $26B valuation. Investors aren’t paying for the harness; they’re paying for the stack. The closest analog is GitHub before Microsoft (when GitHub owned the developer-workflow layer but not the underlying tools), except Cognition also owns the models. At $492M ARR with this kind of vertical integration, the $26B mark prices Cognition at roughly 53× ARR — high but not absurd for a category leader with this growth trajectory.
What it means for the harness category
The AI Harnesses category we shipped earlier this week was structurally divided into:
- Big-tech harnesses (Claude Code, Codex, Antigravity, GitHub Copilot)
- Open-source (OpenCode, Aider, Pi, Goose)
- IDE specialists (Cursor, Windsurf)
- Autonomous specialists (Devin)
The Cognition close consolidates four of those positions — IDE (Windsurf), autonomous (Devin), and vertically integrated model + agent — into one company. The category is now structurally:
- Anthropic (Claude Code) — frontier model + harness, ~$2.5B+ ARR
- Cognition (Devin + Windsurf) — vertically integrated harness business, $492M ARR
- OpenAI (Codex) — frontier model + harness, ARR not separately disclosed
- Google (Antigravity) — frontier model + harness, ARR not separately disclosed
- Cursor + Windsurf-style IDEs — model-agnostic harness, lower-valued
- Open-source (OpenCode, Aider, etc.) — community-driven, not VC-valued
Cognition’s position as #2 — above Cursor on valuation, comparable to Microsoft’s GitHub Copilot business in revenue — is now durable.
What it means for Devin and Windsurf users
Practically little changes immediately. Both products continue to operate. The shared SWE-1.5 backend is already in production. Pricing remains as it was (Devin at $20/month consumer, $500/month team; Windsurf Pro at $20/month).
What does change structurally:
- Capital backing: Cognition is now well-capitalized for multi-year R&D commitments. Concerns about runway are moot.
- Enterprise adoption: With Goldman, Citi, Mercedes, and U.S. Army/Navy named as customers, every other Fortune 500 procurement team can point to peer adoption. Sales cycles get shorter.
- Product roadmap: Expect aggressive investment in deeper Devin + Windsurf integration through H2 2026. The “plan locally, execute in cloud” workflow is the clear competitive moat against Claude Code (terminal-only) and Cursor (IDE-only, no autonomous cloud execution).
- Multi-modal extension: With this capital base, expect Cognition to extend SWE-1.5 into adjacent domains (testing, security scanning, infrastructure-as-code generation).
The honest caveats
Two caveats worth surfacing:
Pentagon-tier customer claims are not third-party verified. Bloomberg’s reporting names the U.S. Army and U.S. Navy as customers, attributed to Cognition. Specific deployment scope, contract values, and DoD program offices aren’t disclosed. Treat the customer claim as Cognition’s own disclosure, not as a Pentagon-confirmed roster.
$492M ARR is the run-rate as of the funding close. ARR figures cited during venture rounds reflect annualized current revenue — they’re real but trail behind actual annual revenue by definition. The $73M → $492M jump represents 11 months of compounding; actual booked 2025-26 calendar year revenue will be lower.
What it changes for Pick Right readers tomorrow
If you’re a Devin subscriber, nothing changes about your service. If you’re a Windsurf user, same. What changes is the structural context: Cognition is now the second-most-valued AI coding company outside the frontier labs, sized comparably to mid-sized public software companies, and with confirmed deployments in regulated industries and DoD.
For broader context, see the Devin review, the Windsurf review, the best AI harnesses roundup, the Claude Code vs Antigravity comparison, and yesterday’s Cognition $25B talks coverage for the immediately-preceding reporting.
Sources
Related tool reviews
Questions or corrections? Email Pick Right. Want the full list? See all news.