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?" The supply of leading-edge silicon was scarce, allocation was political, and the firm that secured the most chips built the most compute. That era is over. The constraint has moved one layer down the stack — past the chip, past the rack, past the building — to the substation. The question that now gates a frontier build 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.

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

The clearest articulation of the shift comes from the people sourcing the capacity: as of 2026, capital is no longer the gating factor and chip supply is no longer the gating factor — the gating factor is the physical electrical layer (SemiAnalysis; Global Data Center Hub, 2026). That is a remarkable inversion. 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. The chip arrived; the power did not.

The economic logic that drives this is unforgiving. A gigawatt of energized AI capacity generates on the order of $10–12B/yr in revenue (SemiAnalysis, 2025 — a contested, single-source figure) — which means getting 200 MW online six months early is worth roughly $1B 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 GPU is now the easy part.

The reason the bottleneck moved is partly that 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. This is not rhetorical inflation. 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. When your unit of compute is a gigawatt, your 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.

The consequence is that compute roadmaps are now downstream of grid physics. NVIDIA can double rack power generation-over-generation — H100 at ~40 kW, GB200 NVL72 at ~120–132 kW, Rubin-class racks heading toward 600 kW 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 and power-boundedness rise together. → 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, for once, 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, Electricity 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 runs straight into a supply wall, and the wall is the interconnection queue. As of the end of 2024, roughly 2,290 GW of generation and storage was actively seeking grid connection in the US — about twice the country's entire installed capacity (LBNL, Queued Up 2025 Edition). The median project now spends about five years in the queue before reaching commercial operation, up from under two years in the early 2000s, and large-load interconnection waits in the densest data center hubs — Northern Virginia, Phoenix, Dallas — are running four to seven years. One utility (CenterPoint) reported a 700% jump in large-load interconnection requests in a single year, from 1 GW to 8 GW. The grid cannot say yes to load at the speed AI wants to consume it. That is the wall. → 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. The schedule-dominating long pole on most 2026 sites is not the GPU, not the building, not even the interconnection study — it is the high-voltage transformer. Lead times for HV power transformers stand at roughly 128 weeks (about 2.5 years) standard, ~144 weeks for generator step-up units, and up to ~60 months in constrained markets (Wood Mackenzie, 2025) — 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 the transformer is a four-year item and the chips are a four-month item, 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.

This reframes procurement sequencing entirely. 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. The long-lead electrical bill of materials is now the artifact that gates the whole program, 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 full long-lead procurement and substation engineering picture in Chapter 3.2; 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 ~82–101 GW of behind-the-meter gas (Cleanview, 2026) is the market voting with its capital that the queue is too slow — though only ~7 GW of it is actually under construction, which tells you the bridge is itself constrained by turbine lead times. 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. The discipline is to know which currency you are short.

The single most useful filter for reading the 2026 build-out is the distinction between an announced gigawatt and one under construction. Of the roughly 12–16 GW of US data center capacity announced for 2026, only about 4–5 GW — roughly one-third — is actually under active construction (SemiAnalysis; Sightline Climate, 2026). The remaining two-thirds sits in the announced stage with no visible construction progress despite typical 12–18 month build timelines, which means it is exposed to delay, repricing, cancellation, or indefinite postponement. Some analysts project that roughly half of the planned 2026 US pipeline may slip or be canceled. 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 reason the gap exists is precisely the power-bound thesis in action: announcements are gated by capital and ambition, both of which are abundant; construction is gated by an energization date, which is scarce. A project announces when the capital is committed; it breaks ground when the interconnection agreement and the long-lead transformers are secured. The two-thirds that announced but did not build are, overwhelmingly, projects waiting on power. This is why "the deliverable megawatt becomes the asset" — the market has learned to discount announced capacity and to pay a premium for energized, contracted, grid-connected megawatts that can actually carry load on a date.

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

• Siting is now power-first, not latency-first or cost-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. Curtailable load unlocks ~100 GW of grid headroom that would otherwise require new generation — trading uptime determinism 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.