The Power-Bound Era: Why the Bottleneck Moved to the Substation

For roughly a decade the question that gated AI capacity was "how many accelerators can you get?" Leading-edge silicon was scarce, allocation was political, and the firm that secured the most chips built the most compute. The constraint has since moved one layer down the stack — past the chip, past the rack, past the building — to the substation. What gates a frontier build now is how many megawatts you can energize, and when. That shift reorders everything downstream of it: siting, financing, procurement sequencing, even which companies are physically capable of competing at the frontier.

The mechanics live in the chapters this one points to; what the strategist and the engineer need in common is the mental model: power is the bottleneck, the gigawatt campus is the unit, and time-to-megawatt governs the rest. We trace the move from chip-bound to power-bound, lay out the global demand curve and the interconnection wall it runs into, identify long-lead electrical equipment as the gate, and close on the gap between an announced gigawatt and one under construction against a real energization date.

The shift shows up first in how capacity is sourced: an AI program now has two independent gates, energized power and accelerator allocation. Two years ago the bottleneck was a queue at a single foundry; today it is a queue at the utility, and the long-lead items — high-voltage transformers, medium-voltage switchgear, generator step-up units — now decide whether a site energizes on schedule. A program must secure both an energization date and accelerator allocation; whichever arrives last sets delivery.

The economics explain the reorganization. A gigawatt of energized AI capacity generates on the order of $12–13B/yr in revenue (SemiAnalysis, Jun 2026 — a contested, single-source figure) — which means getting 200 MW online six months early is worth roughly $1.2–1.3B in pulled-forward revenue, and a year of interconnection delay on a gigawatt campus is a multi-billion-dollar opportunity cost compounding against a depreciation clock that has already started ticking on chips sitting in a warehouse. When the marginal value of a delivered megawatt is that high, the entire program reorganizes around the one number that determines when megawatts arrive. Capital floods toward energized land, secured power, and platforms that bundle financing with grid-connected megawatts; "the deliverable megawatt becomes the thing capital underwrites" (Global Data Center Hub, 2026).

The bottleneck moved partly because the unit of compute grew. A frontier training run no longer fits in a hall; it fits in a campus, and the campus is now measured in gigawatts — a unit borrowed from utility planning, not from IT. A single tightly-coupled pre-training job wants to sit inside one power envelope to keep the back-end fabric short and synchronous, and that envelope has crossed from tens of megawatts (a large 2022 cluster) through hundreds of megawatts (a 2024–2025 build) toward the gigawatt and multi-gigawatt campus that 2026–2027 frontier programs are siting against. At that scale the binding constraint is whatever the local grid, the transformer market, and the interconnection queue will let you energize — and none of those scale at the pace of a silicon roadmap.

Compute roadmaps are now downstream of grid physics. NVIDIA can double rack power generation-over-generation — H100-class racks, HPE GB200 NVL72 at 132 kW nominal, GB300 NVL72 at 135 kW TDP / up to 155 kW peak / up to 142 kW facility basis, Vera Rubin VR200 NVL72 at 188 kW Max Q / 228 kW Max P / 330 kW facility basis, and Rubin Ultra-generation Kyber racks (NVL144) at the ~600 kW H2 2027 roadmap point on 800 VDC — but a 600 kW rack is only compute if a substation can feed it. The density ramp that makes each rack more capable also concentrates more of the bottleneck into the power chain: a hall that was power-bound at 40 kW/rack is dramatically more power-bound at 132 kW, because the same floor area now demands three times the megawatts behind it. → density mechanics in Chapter 1.1; the subsystem roadmaps that drive the ramp in Chapter 16.2.

The demand side of the power-bound story is a wall of forecasts that broadly agree on direction even where they disagree on magnitude. Global data center electricity demand is on track to roughly double from ~485 TWh in 2025 to ~950 TWh by 2030, reaching about 3% of global electricity, with AI-specific load roughly tripling over the period (IEA, Key Questions on Energy and AI, Apr 2026). McKinsey's capacity lens lands at ~219 GW of global demand by 2030, ~70% of it AI (McKinsey, 2025). Goldman Sachs frames the same curve as ~+165% data center power demand by 2030 versus 2023 (Goldman Sachs, 2025). These are demand forecasts, not committed builds — the error bars are wide and the AI share is contested — but every credible curve points up and to the right at a slope the grid has never had to absorb.

That demand curve meets a constrained supply system, but the national queue statistic is supply-side context, not a load queue. As of the end of 2025, more than 2,060 GW of generation and storage was actively seeking grid connection in the US — over one and a half times the country's entire installed capacity, down about 10% from ~2,290 GW a year earlier on high withdrawal rates alongside relatively fewer new requests; LBNL names both contributors and says it is too early to measure the reforms' full effect, so the dataset does not rule out softer demand (LBNL, Queued Up 2026 edition). For the generation/storage projects built in 2025, LBNL reports a median request-to-COD duration around five years, up from under two years for the 2000–2007 generation/storage cohort. That is not a large-load service statistic. Data-center energization must be read from the named utility's tariff, studies, upgrades, agreements and milestones. One utility (CenterPoint) reported a 700% jump in large-load interconnection requests over a single summer, from 1 GW to 8 GW. The grid cannot say yes to load at the speed AI wants to consume it. → the interconnection mechanics, ISO/RTO pathways, and speed-to-power options live in Chapter 3.2; flexible/curtailable interconnection as a release valve in Chapter 15.8.

Behind the abstract "interconnection wall" sits a concrete supply chain, and within it a handful of components do the actual gating. Once accelerator allocation is secured, compare transformer delivery with the interconnection and construction milestones; whichever arrives last sets the site's energization date. Lead times for HV power transformers stand at roughly 128 weeks (about 2.5 years) standard, ~144–208 weeks for generator step-up units (SemiAnalysis reported 3–4 years for US GSUs by mid-2026), and up to ~60 months in constrained markets (Wood Mackenzie, 2025; SemiAnalysis, 2026) — up from 24–30 months pre-2020 and still stretching as orders surge. Medium-voltage switchgear, large breakers, and static VAR compensators are queuing behind the same constrained global manufacturing base. When contracted transformer delivery trails accelerator availability, the entire critical path is set by the electrical layer, and ordering long-lead power equipment before the design is frozen becomes a rational hedge rather than premature commitment.

In the chip-bound era you secured allocation first and worried about power later. In the power-bound era the order inverts: you reserve the interconnection slot and order the transformers first, then fit the GPU generation to the power envelope that equipment will deliver on its arrival date. Once accelerator allocation is contracted, manage the long-lead electrical bill of materials against the site schedule's critical path, and operators who treated it as a back-office facilities concern are discovering that a $283k server is useless without a $3M transformer that ships in 2029. → the large-load service and speed-to-power overview in Chapter 3.2; long-lead procurement in Chapter 2.3; on-site substation and MV distribution in Chapter 4.2; energy-supply strategy and PPAs in Chapter 3.4; on-site generation commissioning in Chapter 13.4.

None of the four levers is free, and the choice is rarely all-or-nothing. The real-world answer is usually a stack: a behind-the-meter bridge to cover the first 18–36 months, a grid interconnection underneath it for the long-run cheap power, flexible-load terms to widen the headroom the utility will grant, and an energized-land acquisition where speed is existential. The announced ~90 GW of behind-the-meter generation (Cleanview, mid-2026) signals developers' intended route around a slow queue; announcements are not committed capital or operating generation. Cleanview reports about 2 GW operating and 1.2% under construction within that announcement stock. Each lever trades a different currency: the queue trades time for cost, BTM gas trades carbon and capex for time, flexibility trades uptime determinism for headroom, and buying energized land trades dollars for certainty. Which lever to pull depends on which currency you are short.

Keep four columns in the capacity ledger. Announced: 100 MW. Financed: 40 MW. Energized in the first year: 20 MW. Productive service equivalent: 20 × 6/12 × 0.75 = 7.5 MW-years; multiplying by annual hours would describe service-capacity hours, not metered MWh. Under the six-month power slip, first-year productive service falls to zero, while funded equipment and its 25% residual stress remain exposed. In the first full year after release at 50% useful-service fraction, those 20 MW yield 10 MW-years, rather than the base case’s 15. Do not release the second 20 MW fit-out merely because the 100 MW announcement survived. Require executed power milestones, funded installation, an accepted workload/SLO trace and customer demand sufficient for the phase. In this fixture set the release hurdle at 12 productive MW-years per full year: 20 × u ≥ 12 means u ≥ 60%. At 50% defer; at 65%, with power and funding restored, 13 MW-years clears the hurdle. The 60% crossover follows from the opening ledger’s assumed 12 MW-year hurdle. Re-price the residual and financing in Chapter 1.8 and Chapter 2.5; this ledger controls phase release, not project valuation. → power-portfolio response in Chapter 16.5.

The most useful filter for reading the 2026 build-out is the distinction between an announced gigawatt and one under construction. Sightline Climate's early-2026 tracking put roughly 12–16 GW of US data center capacity announced for 2026 against about 4–5 GW under active construction — roughly one-third. Read that as one tracker's sample on one observation date, not a settled market total: SemiAnalysis rejects both the denominator and the construction figure, reporting that the top two hyperscalers' self-build construction alone exceeds 5 GW (SemiAnalysis, Jun 2026). On Sightline's count the remaining two-thirds sits in the announced stage with no visible construction progress despite typical 18–24 month build timelines, which means it is exposed to delay, repricing, cancellation, or indefinite postponement. The 'half of 2026 US capacity may slip or be canceled' headline (Bloomberg, 1 April 2026, on Sightline's estimates) is the reading SemiAnalysis's June rebuttal is aimed at. The headline-grabbing announcements and the steel actually being erected are two very different datasets, and conflating them is how capital gets allocated against capacity that never materializes.

The gap exists because announcements are gated by capital and ambition, both of which are abundant, while construction is gated by an energization date, which is scarce. A project can announce before financing closes; it can begin construction before power delivery is secured. Sightline’s unbuilt two-thirds therefore cannot be assigned to power alone: permits, financing, customer commitments and equipment can each stop the project. The market has responded by discounting announced capacity and paying a premium for energized, contracted, grid-connected megawatts that can actually carry load on a date — the deliverable megawatt as the asset.

The shift from chip-bound to power-bound reorders the entire decision hierarchy, and it changes who wins:

• Siting for frontier training and batch capacity is now power-first; latency-bound online and edge inference stays latency-first. The reordered criteria hierarchy puts speed-to-power at the top of the site-selection screen — a site with a four-year energization date is not a site, it is a liability with a deed. → Chapter 3.1.
• Financing underwrites megawatts, not buildings. Capital consolidates around energized land, secured interconnection rights, and long-term power offtake. The deliverable megawatt is the collateral. → the macro financing lens in Chapter 16.4.
• Procurement sequences backward from the energization date. You order long-lead transformers and reserve the queue slot before you freeze the chip generation, because the electrical layer is the critical path. → Chapter 3.2.
• Flexibility becomes a power-supply strategy, not just a reliability feature. Flexible-load studies model ~98 GW of aggregate US grid headroom that would otherwise require new generation (Duke Nicholas Institute, 2025) — a national potential rather than a site's offer, and one that trades uptime determinism plus curtailment cost for an earlier energization date. → Chapter 15.8.
• The competitive moat shifts from chip access to power access. When everyone can eventually buy the chip, the durable advantage is the gigawatt you can energize first — which is why the build-out's structure, geography, and supply endgame are the subject of the rest of Part 16.