What changed in the report's value-migration chapters between the original March 2026 edition (data as of February 28, 2026) and the updated July 2026 edition. A five-month view across the material of Chapter 5 (AI Value Migration) and Chapter 6 (The Value Layer Shift).
The value-migration chapters make a single argument: infrastructure captures the bulk of a revolution's value at peak, then commoditizes as value migrates up to the applications built on top of it. Chapters 5 and 6 anchored that argument in a snapshot — NVIDIA at $4.3T, Microsoft 13x Intel, Anthropic at a $14B run-rate. Five months of market data moved several of those anchors. This page holds the two editions side by side and asks a narrow question: what actually moved between February 28 and mid-July 2026, and did any of it bend the thesis? Every figure below is pulled directly from both editions of Chapters 5 and 6; forward or run-rate figures carried with a "~" in the source are marked est.
Chapter 5 and 6 use the Microsoft–Intel pair as the cleanest illustration of value moving from silicon to software. That ratio moved the most of any figure in the two chapters — and it moved in the direction that nuances the story rather than confirming it.
The nuance matters for how Chapter 5 reads. The chapter's claim was never "hardware loses" — it is that value migrates to the applications of intelligence, not simply away from silicon. Intel's re-rating and NVIDIA at ~$4.9T are the reminder built into the thesis itself: the infrastructure champion stays enormously valuable (utilities, telcos, and Intel all still exist), even as the relative share of value shifts up the stack. A 13x gap narrowing to ~6x doesn't refute the migration — it recalibrates how fast and how far the silicon layer is actually being left behind.
The layer-level figures that genuinely differ between the two editions. Every value in these columns is drawn from the February and July editions of Chapters 5 and 6; the "what moved" column reflects market movement, not editorial fixes (those are in the box below).
| Figure | March edition (Feb 28) | July edition | What moved |
|---|---|---|---|
| Hyperscaler Big-4 capex (2026E) | ~$600B projected | ~$700B projected est. | +~$100B The 2026 arms-race projection climbed as the year firmed up. |
| Hyperscaler Big-4 capex (2025) | $350B | $380B | +$30B Actuals for 2025 came in above the February projection. |
| NVIDIA market cap | $4.3T ($4,314B) | ~$4.9T ($4,900B); briefly >$5T | +~14% ~34x off its $144B base; now trading the #1 spot with Apple. |
| Microsoft / Intel ratio | 13x (Intel ~$230B) | ~6x (Intel ~$478B) | narrowed Driven entirely by Intel's re-rating; Microsoft ~$3.0T held. |
| Anthropic run-rate | $14B ARR (140x from $100M) | ~$47B run-rate (~470x) est. | ~3.4x The single largest revenue move in either chapter. |
| Anthropic valuation | $183B | ~$965B (Series H; confidential IPO filed Jun 2026) | ~5x Re-priced alongside the run-rate acceleration. |
| OpenAI ARR | ~$20B (end 2025) | ~$25B est. | +~$5B The fastest enterprise-software ramp kept ramping. |
| Netflix / AT&T ratio | 2.27x | 2.27x | flat A mature over-the-top layer; unchanged across the two dates. |
The pattern in the table is consistent with the chapters' argument: the fastest movers are all above the model layer — application and model-company revenue (Anthropic, OpenAI) and the valuations attached to them — while the pure-infrastructure ratios either held (Netflix/AT&T) or compressed toward the applications above them (Microsoft/Intel). The one figure racing upward on the infrastructure side is spend, not captured value: hyperscaler capex.
Chapter 5's central projection held across both editions: infrastructure spending grows ~3.7x to $500B by 2035 while application spending grows 116x to $1,550B, with the two lines crossing around 2031. The H1 market data — capex accelerating, model prices bifurcating — is consistent with that crossover, not a contradiction of it.
Two things are worth separating. First, the near-term infrastructure line is running hot: Big-4 capex rose from a $350B actual to a ~$700B projection for 2026, which the July edition notes "may delay the crossover relative to prior technology analogies." Second, the endpoint didn't move — $500B infrastructure and $1,550B applications by 2035, a 116x-versus-3.7x split, with a ~2031 crossover. Faster infrastructure spending is exactly what the thesis predicts at the peak of the build-out phase: every dollar into GPU clusters makes compute more abundant, drives token prices down, and expands the market the application layer will eventually capture. The forecast survived H1 intact; the July edition mostly tightened the two chapters onto the same numbers rather than changing them.
Three take-aways from holding the editions side by side — each one a place where H1 2026 reinforced rather than weakened the value-migration argument.
Even after Intel roughly doubled to ~$478B, Microsoft is still worth about six times as much. The re-rating narrowed the gap but did not close it — software remains multiples above the silicon it runs on, exactly where the layer framework places the value.
Hyperscaler capex is projected near $700B for 2026 alone, yet infrastructure is still modeled to grow only 3.7x to $500B of captured value by 2035. Infrastructure spend is racing ahead of infrastructure value-capture — the classic build-out signal that precedes every inversion.
The 2035 projection still has applications growing 116x ($13.3B to $1,550B) against infrastructure's 3.7x, crossing around 2031. Nothing in H1 2026 bent that curve; the run-rate surges at Anthropic and OpenAI are early evidence for it.
This page compares market reality between two dates — February 28, 2026 and mid-July 2026 — and deliberately keeps that separate from editorial corrections. The July edition also fixed several errors carried in the February edition of these two chapters. Those fixes are not in the "what moved" tables above, because they reflect copyediting, not the market:
Where a July figure is a run-rate or a forward projection carried with a "~" in the source — principally NVIDIA at ~$4.9T (briefly >$5T), Anthropic's ~$47B run-rate, OpenAI's ~$25B ARR, and the ~$700B 2026 capex projection — it is marked est. and should be read as the report's estimate rather than an audited figure.