Benedict Evans’ Spring 2026 slide deck frames AI as a platform shift comparable to the internet and mobile eras, tracing model commoditization and open deployment questions.
Key Takeaways
Evans tracks four versions of his AI deck since late 2024, showing evolving conviction: early uncertainty on scaling and business models, then growing signs of model-layer commoditization.
The deck uses historical platform-era framing: hardware, internet, mobile, cloud, and now AI, each producing distinct winner categories.
Annualized revenue figures in the deck use a 4-week sum multiplied by 13, a non-standard but common SaaS proxy for run-rate.
Coding agents are cited as a leading current use case for LLM-based systems, implying developer tooling is the near-term revenue concentration point.
Hacker News Comment Review
Commenters debate Evans’ telecom analogy for LLM labs: the stronger parallel may be cloud infra providers (AWS/Azure/GCP) rather than AT&T/Verizon, since model differentiation still exists unlike standardized radio protocols.
A recurring thread questions whether current large transformer architectures are inefficient “mainframe-era” systems, with the real gains coming from better inference harnesses and smaller specialized models rather than raw parameter scaling.
Consensus is that models will commoditize but timing and who captures app-layer value remains unsettled; Evans himself replied in thread acknowledging the commodity dynamic does not require standardization as a precondition.
Notable Comments
@benedictevans: Evans clarifies that capital required to build SOTA models today is “nowhere near enough to lead to a monopoly” and that standardization is not a precondition for commodity pricing.
@btucker: Summarizes Evans’ arc across four deck versions, noting the May 2025 edition pivoted toward model-layer commoditization as the central strategic question.
