Blackwell took cost per million tokens from 4.20 to 0.12. A 35x drop. Average inference hit 1.16 per million in August, the lowest of the year.
1
2
61
Sources: a16z on inference cost a16z.com/llmflation-llm-infe… | Deloitte deloitte.com/us/en/about/pre…
1
1
17
And: MIT NANDA (archived) web.archive.org/web/20250818… | Stanford AI Index hai.stanford.edu/ai-index/20…
6
The lesson I still use every day: when you're stuck, stop guessing and go talk to customers. The answer is usually sitting right in front of you.
1
1
7
We'd spent months guessing what customers wanted. This one conversation - maybe 20 minutes - mapped out what they actually needed.
1
1
8
What saved me wasn't a pivot, a new hire, or a clever strategy. It was one customer who sat me down and told me exactly what was broken.
1
1
13
Small businesses have the same problems as enterprises - just smaller budgets and zero technical staff. The difference is nobody's selling to them.
2
3
62
But that's exactly why the opportunity is real. The market too boring for everyone else is the one with the most room.
1
1
9
