Does Devin have a moat?

Cognition raised $400M at $10.2B in 2025, more than $1B at about $26B in 2026, then $2B at $48B in September 2026. It bought Windsurf and brought about 10K NVIDIA GPUs online for training.

To cross it: Raise billions from top-tier investors, which in practice only the model labs and a few hyperscaler-backed rivals can do.

Goldman Sachs pilots Devin as a 'new employee'. Citi rolled it out alongside its 40,000 developers, and Mercedes-Benz, NASA and Santander are named customers. Banks don't hand their codebases to unknown vendors.

To cross it: Spend years running pilots and security reviews at regulated enterprises, with a clean incident record.

SOC 2 Type II since September 2024, VPC deployment, audit logs, custom IdP and fine-grained access controls on Devin Enterprise. It also has an AWS partnership for legacy migration deals.

To cross it: Get SOC 2 Type II, build a single-tenant VPC offering and pass procurement at large banks, which takes 12 to 18 months.

Cognition trains its own coding models (SWE-1.5, now the SWE-1.7 family) and serves them on Cerebras at up to 13x the speed of Sonnet 4.5. It also mixes in OpenAI and Anthropic models, and its founders include IOI gold medalists.

To cross it: Hire a frontier research team and buy GPU capacity to train coding models with their own agent harness, or just rely on lab APIs and give up on cost and speed.