On prem GPU capacity - or decent enough devices for core engineering team - lends itself pretty nicely to local LLMs too. And you own the whole stack this way. Why pay premiums to Anthropic and fuel its trillion dollar valuation?