“Stop paying for intelligence. Start paying for what intelligence earns.”
Three numbers define the new regime. The US 10-year Treasury has moved above 5.2%, while Brent remains above $105. The financial and physical hurdle rates of the AI buildout are rising simultaneously.
Yet Nvidia has authorised another $150 billion of share buybacks, taking remaining repurchase capacity to $235 billion — one of the largest beneficiaries of the AI Supercycle can fund extraordinary investment while returning extraordinary capital to shareholders.
AMD is deploying capital differently: its $8.2 billion acquisition of World Labs moves it upstream from processors into spatial intelligence, simulation and Physical AI. AI is no longer one trade. It is becoming a contest between different forms of capital productivity.
top ten, unchanged
- United States — 94.4 (strong; financing drag rising)
- China — 91.1 (full-stack acceleration)
- South Korea — 88.5 (semiconductor supercycle)
- Japan — 84.9
- Germany — 83.2
- United Kingdom — 82.3
- France — 81.4
- Switzerland — 80.7
- Netherlands — 80.4
- Sweden — 79.1
Five Signals
the shifting economics of AI capital
-
A US 10-year yield above 5.2% fundamentally changes the economics of long-duration AI. Capital remains available, but its required return has moved sharply higher. The question is no longer whether AI can attract capital — it is which AI deserves it. signal: The question shifts from access to capital to worthiness of capital.
-
Nvidia's additional $150 billion buyback demonstrates that one of the most valuable layers of the AI stack has crossed from capital consumption to capital generation — a distinction that becomes enormously valuable as external capital becomes more expensive. signal: The most valuable AI layer now generates capital rather than consuming it.
-
AMD's acquisition of World Labs connects semiconductor architecture with spatial intelligence, simulation, robotics and Physical AI. The race is evolving from chips toward chips + software + models + applications. signal: The chip race widens into a full-stack capability contest.
-
Hyperscaler bond issuance has reached approximately $220 billion this year and could rise substantially again next year, while credit spreads are beginning to widen. The next AI stress signal may appear first in credit, not equities. signal: Watch spreads — stress may show in credit before equities.
-
France's Legrand raised its 2027–2030 organic-growth target to 6–8%, citing stronger-than-expected data-centre demand, and expects global installed data-centre IT capacity to rise from roughly 80 GW today to 180 GW by 2030. This is the transmission the GVI captures: AI capex → infrastructure → industrial orders → revenue → cash flow. signal: Data-centre demand converts into durable industrial revenue.
Three Second-Order Effects
how the AI trade reorganises
The AI trade splits into equity compounders versus credit consumers, with an emerging hierarchy: free cash flow → balance sheet → investment-grade debt → project finance → high-yield capital. As rates hold above 5%, the market will stop asking "How much AI exposure does this company have?" and start asking "Who is funding whom — and at what cost?"
Capital rotates from compute to bottlenecks. Every incremental accelerator creates demand elsewhere: memory → optics → networking → power → transformers → cooling → grid. These assets cannot scale at software speed, and scarcity creates pricing power. The next rotation is likely compute → bottlenecks → productivity.
M&A moves from consolidation to capability acquisition. AMD–World Labs is an early marker: the objective is acquiring a missing layer of the future stack — semiconductors → models, cloud → energy, software → agents, industrials → robotics. The next M&A cycle could become unusually cross-sectoral.
The Market Rotation Ahead
AI narrative → cash flow → bottlenecks → productivity
Phase I — Reduce duration
Long-dated, cash-burning AI stories become vulnerable as discount rates rise.
Phase II — Reward cash generators
Companies that can finance AI investment internally command a structural premium.
Phase III — Reprice bottlenecks
Memory, networking, power, cooling and grid infrastructure capture rents created by accelerating compute deployment.
Phase IV — Reward productivity
The ultimate rotation moves beyond technology toward companies proving AI raises margins, asset turns, labour productivity and free cash flow — where the Supercycle becomes an economy-wide productivity cycle.
five moves for the new regime
- Own AI cash flow — not AI narrative. Self-financing growth gains value as external capital gets more expensive.
- Watch credit before equities. Widening AI-related spreads may flag stress first.
- Move into scarcity. Memory, networking, power, transformers, cooling and grid can capture the rents of compute expansion.
- Expect cross-stack M&A. Companies will acquire missing capabilities rather than consolidate competitors.
- Raise the hurdle rate. At 5%+ sovereign yields, growth without cash conversion deserves a lower multiple.
from technology agenda to capital allocation
- Move AI decisively from the technology agenda to the capital-allocation agenda.
- Require every major investment to demonstrate: capacity → utilisation → productivity → cash flow → ROIC > WACC.
- Stress-test AI programmes against higher rates, higher power costs and lower utilisation.
- Ask explicitly which layers should be build → buy → partner → acquire.
- Measure competitive advantage through economic output per dollar of intelligence deployed.
For investors and boards: stop paying for intelligence — start paying for what intelligence earns.