Enterprise AI is Not a One-Time Purchase
Nimsumdimsum · reddit · 2026-07-15
From an enterprise AI practitioner's perspective, the author responds to Nadella's "you pay for AI twice" remark. They agree with the direction but argue the focus shouldn't just be on "whether the model learns from your corrections," but rather on whether enterprises carefully review terms of service and data usage policies before purchasing.
They emphasize that the bigger issue is the total cost: if a company truly wants to "own" its AI stack, it must bear the full costs of infrastructure, talent, maintenance, and security. In contrast, most companies simply want a sufficient, low-cost, low-hassle solution.
They also highlight a frequently overlooked design issue: in multi-tenant systems, different clients' models and data should not cross-contaminate or share hidden benefits. This should be an explicit architectural choice rather than a default behavior.
More from Companies & People
- OpenAI unifies ChatGPT, Codex and API release notes into one searchable page with RSS — craigsdennis · 2026-10-26
- diginomica: agentic automation hype is overdone, and 'The Hackening' of enterprise security has begun — jonerp · 2026-10-26
- Ben Bajarin: enterprise AI adoption mirrors early cloud, at magnitudes larger dollar size — BenBajarin · 2026-10-26
- An AI company built in five years now moves a mid-sized European GDP — felpix_ · 2026-10-26
- Microsoft offers mass buyouts to roughly a quarter of its workforce — doctorzook · 2026-10-26
- Fmr. OpenAI Design Head Ian Silber, Who Led ChatGPT and Codex, to Speak at SPC — soleio · 2026-10-26