Compounding Waves — How Each Tech Era Built the Substrate, and the Skills, for the Next

Narrative

The VC and analyst record across 1995–2026 is most coherently read as a succession of substrate-and-application pairs, each of which required the prior wave's infrastructure to exist. Andreessen's 2011 'Software Is Eating the World' essay, republished and re-examined by a16z in 2024 and 2026, captures the moment when the internet wave's outputs — payment rails, broadband penetration, cloud economics — had cheapened deployment sufficiently to enable the software-defined wave. The essay's own numbers (application hosting costs falling from $150,000 to $1,500 per month between 2000 and 2011) describe a cost-curve compression that made SaaS viable, and that same compression, now applied to GPU inference, is the structural question animating the current wave. By February 2026, Fortune and Morgan Stanley were describing AI as eating the software layer itself, with unstructured data (over 80% of enterprise information) as the new automation frontier — the precise category that required the preceding decade of cloud-era tooling to make tractable.

Sequoia's David Cahn provided the most cited quantitative framing of wave-three's structural problem. His September 2023 analysis required $200B in annual end-user revenue to justify then-current GPU capex; by June 2024 that figure had risen to $600B, tracking Nvidia's revenue acceleration. Bain's sixth annual technology report (September 2025) independently sized the requirement at $2T in new annual revenue by 2030 against a 200-gigawatt compute demand. Both analyses raise the same structural question: whether the infrastructure investment precedes, and compels, the application revenue that justifies it — a pattern Gartner explicitly compared to dot-com, energy and telecoms corrections in October 2025, while simultaneously projecting that agentic AI could drive 30% of enterprise application software revenue ($450B+) by 2035. The tension between these positions — necessary correction versus deferred but genuine value — is unresolved in the published record.

CB Insights' annual venture data provides the clearest quantitative map of capital concentration. In 2024, AI represented 37% of venture funding and 17% of deals, both all-time highs, with all five of the year's largest deals going to AI infrastructure players. By full-year 2025, AI had reached 48% of total venture funding ($226B of $469B), with the six largest rounds (OpenAI $41B, Anthropic $32.5B, Scale $14.8B, xAI $12.8B, Databricks $5B, Aligned $5B) accounting for 49% of all AI capital. This concentration is consistent with the compounding-wave thesis — value accrues to the layer that controls the substrate — but it also constitutes the 'externalised harvest' pattern, in that hyperscalers who built their infrastructure during wave two now capture disproportionate returns from wave three without bearing the cost of the open-web corpus on which frontier models were trained.

McKinsey's 2025 State of AI survey and Bain's maturity framework converge on a productivity paradox that mirrors wave-two patterns. McKinsey finds 88% of organisations using AI in at least one function but only 6% qualifying as high performers with measurable EBIT impact, while Bain documents that early leaders achieved 10–25% EBITDA gains by embedding AI in specific workflows rather than distributing it diffusely. The skill categories emerging in wave three — data engineers, ML engineers, prompt engineers, AI ethics specialists and what McKinsey calls 'business–AI translators' — reproduce the pattern of each prior wave inventing its own technical intermediaries (webmaster, cloud architect, data scientist) whose scarcity briefly conferred wage premiums before the skills diffused. Gartner's projection that 80% of engineering workers will require upskilling by 2027, combined with the entry-level labour-market disruption documented in 2025 hiring data, suggests the transition cost is again borne by workers rather than the firms that benefit from the compounding returns.