a16zAI Eats the World: Benedict Evans on the Next Platform Shift
At a glance
WHAT IT’S REALLY ABOUT
AI as next platform shift, bubbles, products, and winners ahead
- Evans frames generative AI as a possible platform shift on the scale of the internet or smartphones, but stresses that history shows we can recognize magnitude without knowing which products or companies will define the era.
- He argues the core uncertainty versus past shifts is that we don’t know the “physical limits” of model capability or compute needs, making forecasting and roadmapping inherently vibes-based and bubble-prone.
- Usage data suggests a bifurcation: a minority uses ChatGPT intensely for coding/marketing/knowledge work, while many others understand it but “can’t think of anything to do with it,” implying productization and workflow integration are the real unlocks.
- Evans emphasizes that companies and industries will capture value differently—some will be structurally disrupted (like newspapers were by the internet), while others will mostly gain incremental efficiency.
- Competitive advantage may depend less on slightly better benchmarks and more on distribution, defensible product ecosystems, and cost/infrastructure control—creating strategic pressure especially for OpenAI and distinct questions for Google, Meta, Amazon, and Apple.
IDEAS WORTH REMEMBERING
5 ideasPlatform shifts are predictable in pattern, not in winners.
Evans argues you can know a shift is big while being wrong about which form it takes (internet vs web; smartphones dominated by Apple/Google rather than Nokia/Microsoft), so deterministic forecasts about AI’s end-state are unreliable.
Generative AI’s unique uncertainty is the unknown capability ceiling.
Unlike bandwidth or battery roadmaps, we lack a solid theory for why LLMs work and therefore can’t model limits; this fuels conflicting claims ("PhD-level agents" vs "not even close") and makes planning speculative.
A bubble is likely because transformative tech triggers overinvestment.
He expects “bubbly behavior” as rational actors overbuild capacity to avoid missing out, but warns that spare capacity can’t be easily resold if everyone overbuilds at once (pushing back on the idea that excess compute is safely liquid).
Compute cost may fall fast, yet total spend can still rise.
Evans notes per-unit efficiency improvements (orders-of-magnitude over time) can be overwhelmed by exploding usage—similar to late-1990s bandwidth forecasts—making demand vs supply bottlenecks hard to call.
The adoption problem is often ‘what do I do with it?’ not awareness.
Despite massive weekly active usage, many users don’t find weekly/daily tasks; Evans suggests the next wave is mapping AI onto specific jobs-to-be-done via products, not expecting users to invent workflows from a blank prompt.
WORDS WORTH SAVING
5 quotesChatGPT has got eight or nine hundred million weekly active users. And if you're the kind of person who is using this for hours every day, ask yourself why five times more people look at it, get it, know what it is, have an account, know how to use it, and can't think of anything to do with it this week or next week.
— Benedict Evans
We don't know the physical limits of this technology because we don't really have a good theoretical understanding of why it works so well, nor indeed do we have a good theoretical understanding of what human intelligence is. And so we don't know how much better it can get.
— Benedict Evans
Deterministically very new, very, very big, very, very exciting worlds changing things tend to lead to bubbles.
— Benedict Evans
So yeah, if we're not in a bubble now, we will be.
— Benedict Evans
People buy solutions, they don't buy technologies.
— Benedict Evans
High quality AI-generated summary created from speaker-labeled transcript.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome