When constraints in computing break, nobody predicts what comes next

Every major constraint in computing history eventually broke. When it did, nobody predicted what came next.

Storage got cheap — we didn’t just store more files. We got Spotify, Netflix, YouTube. Entire industries that couldn’t exist when every megabyte had a price tag.

Bandwidth got cheap — we didn’t just load pages faster. We got real-time collaboration, cloud computing, video calls replacing flights.

Subquadratic just shipped the first production LLM where attention scales linearly with context. 12M tokens. 52x faster than FlashAttention. 95.6% on RULER 128K — slightly ahead of Anthropic Opus 4.6.

If it holds, context just got cheap.

And the interesting question isn’t "what gets faster?" It’s "what gets built that couldn’t exist before?"


Originally posted on LinkedIn ↗