Broadcom's Q2 2026 earnings call contained one of the clearest statements yet from a major semiconductor company about where the AI bottleneck actually sits. Management described the AI XPV platform — a joint vehicle with Apollo, Blackstone, and other investors deploying more than 20 GW of compute capacity through 2028, with a first tranche valued at $35 billion.
When asked what customers need to align before delivery, management's answer was unambiguous:
The call also stated that visibility now runs to 2028 and that AI infrastructure demand is distributing compute "through America at least" — meaning geographic expansion of power requirements beyond existing clusters.
WHY THIS MATTERS — THE ELECTRIFICATION CYCLE THESIS IS NOW OFFICIAL
Broadcom is not a grid company. It is a semiconductor company with over $700 billion in market capitalisation. When its management explicitly identifies power and power infrastructure as the pre-condition for compute deployment — ahead of chip delivery — the bottleneck thesis moves from analyst speculation to corporate earnings disclosure.
The transmission mechanism described in the earnings commentary names the exact companies GridReadiness tracks:
Eaton (ETN): switchgear, electrical distribution, power management, grid-edge infrastructure
Schneider Electric (SU): switchgear, electrical distribution, power management
GE Vernova (GEV): generation, grid, transformer, electrification demand
Siemens Energy (ENR): generation, grid, transformer, electrification demand
Quanta Services (PWR): transmission, substation, grid interconnect, utility-scale buildout
Schneider Electric is on this list. GridReadiness tracks Schneider France transformer lead times monthly — currently 18–26 months for HV units. The company Broadcom is calling a direct beneficiary of AI's power constraint is exactly the manufacturer whose production slots are available for France brownfield projects over the next 18–30 months.
THE $35 BILLION TRANCHE — WHAT "POWER FIRST" MEANS IN PRACTICE
The AI XPV platform's first tranche, valued at $35 billion and described as "currently being launched by Apollo," is structured specifically to deploy 20+ GW of compute capacity. The explicit statement that power must be aligned "before delivery" means this capital cannot be deployed until grid connections, transformer procurement, and power contracts are secured.
At $50 billion per GW (Jensen Huang's Computex estimate), 20 GW represents a $1 trillion infrastructure programme. The power infrastructure layer — grid connections, transformers, substations — represents roughly $15 billion per GW of that total, or $300 billion across the full 20 GW programme.
Every week that power is not secured is a week that $35 billion in committed capital cannot be deployed. This is the financial translation of "the power, the power shell."
THE EUROPE IMPLICATION — BROADCOM SAID "THROUGH AMERICA AT LEAST"
The phrase "through America at least" is important. Management explicitly acknowledged that distribution of AI compute will go beyond the United States. The US constraint — PJM grid queue at 7–10 years, Virginia effectively saturated, Texas at capacity margins — pushes 20 GW deployments toward markets where power can be secured faster.
France brownfield timeline: 18–36 months. France nuclear baseload: €50–70/MWh. EU second-tier transformers (Schneider, Efacec, Pauwels): 18–32 months. These are the numbers that answer "the power, the power shell" in a European context.
Efacec Portugal: 20–28 months · ANSI/IEEE 60Hz available
Pauwels Belgium: 24–32 months
France brownfield grid connection: 18–36 months
Virginia/PJM: 7–10 years
The bottleneck Broadcom is describing is exactly what GridReadiness tracks monthly.
Broadcom's Q2 2026 call is not an isolated data point. It follows Ashmore Group's May 2026 institutional analysis, Marvell CEO's Computex confirmation of bottleneck migration, and Choose France's June 1 commitments totalling €88 billion. The convergence is not coincidental. The AI infrastructure constraint is shifting from semiconductors to electrification — and that shift has now been confirmed by a $700 billion semiconductor company on a public earnings call.
GridReadiness tracks the physical infrastructure layer monthly. The data is at gridreadiness.com/data.