Scenarios for 2030

Every chapter before this one decides something about a building you can draw. This one decides which world the building lives in. The same 200 MW campus is a brilliant asset in one 2030 and a stranded liability in another, and the difference is which of a few structurally distinct futures actually arrives. The scenario is the largest uncontrolled variable in any AI-infrastructure thesis, and it is usually smuggled in as an unexamined assumption.

Three axes frame 2030, and they move semi-independently. The demand-and-capital regime: supercycle, digestion, or bubble — distinct from the firm-level economics of Chapter 1.8; here it is the sector-wide structural lens. The architectural shape: centralized mega-campus versus distributed inference, the fork that decides where the megawatts go. The industry structure and geographic re-map: who owns the capacity and where it lands. From there the chapter turns to the power-supply endgame — CFE, SMR, fusion optionism, and the demand-side wildcards that could invalidate the whole demand curve — and closes on stranded-asset and systemic risk, the downside each scenario hides.

Start with the demand-and-capital regime, because it sets the boundary conditions for everything else. The three scenarios are not optimist/realist/pessimist moods — they are different beliefs about a single mechanism: does AI revenue grow fast enough to service the capital being deployed before the assets depreciate? McKinsey's own framing spans a $3.7T-to-$7.9T range of AI-capable capex to 2030 against a ~$5.2T midline — a spread wide enough that the constrained and accelerated cases describe genuinely different industries (McKinsey, 2025).

Supercycle is the case where demand keeps outrunning supply: reasoning and agentic workloads inflate tokens-per-task, inference compute compounds, and power — not chips, not capital — stays the binding constraint through 2030. In this world the scarce asset is an energized megawatt, time-to-power dominates every other lever, and the operators who locked firm power and long-lead gear early win. Digestion is the soft-landing case: the extraordinary base is absorbed, capex growth decelerates sharply (one widely-cited path: ~51% growth in 2026 falling to ~13% in 2027 and ~5% in 2028) without a crash, and the winners are the disciplined operators who did not over-commit at the top. Bubble is the case where the revenue gap wins: the spend behaves like a utility build-out while revenue still behaves like software subscriptions, and the mismatch forces write-downs, cancelled announcements, and a depreciation reckoning. David Cahn’s July 8, 2026 heuristic puts lifetime end-customer revenue at ~$1.5T for one annual capex cohort; Bain sized a ~$800B annual shortfall by 2030 even after AI-driven savings (Bain, 2025).

Nobody knows which scenario will arrive, so the decision that follows is which one your balance sheet is structured to survive. A supercycle bet is contracted, levered, and over-built on power; a bubble hedge is merchant-light, opex-flexible, and short on irreversible commitments. You cannot be optimized for both at once, and pretending you are is how operators get caught.

The second axis is architectural, and it is orthogonal to the demand regime: even a supercycle spans a workload continuum that pulls siting and design in different directions. Centralized mega-campus is the gigawatt-scale training factory — one tightly-coupled supercomputer chasing the cheapest firm power and the coldest climate, indifferent to user proximity, with the gigawatt campus as the unit of compute (→ Chapter 16.1). Inference splits by workload: frontier/model-parallel and batch inference fit the same dense liquid-cooled campus substrate as training; latency- or residency-bound inference fits regional sites; strict-locality inference fits edge or on-premises capacity.

The forecast that makes this fork decisive is the inference transition: by the end of the decade the majority of AI compute is inference, not training, even as frontier training concentrates into 1 GW-plus campuses that become standard (Bain, 2025) — and that crossover is not a distant event, since Deloitte's November 2025 outlook already forecasts inference at about two-thirds of 2026 AI compute (Chapter 1.3). If you build only for centralized training, you cannot serve every latency or residency need; if you build only a small distributed fleet, you cannot host frontier training or model-parallel inference. The design bases diverge by workload — density, redundancy, fabric blocking, and siting driver vary across the continuum — which is why this is a fork to decide at scoping, not a dial to tune later. Reasoning and test-time compute increase the decode-heavy share but do not by themselves determine whether serving is centralized or distributed (→ Chapter 16.3).

Who owns the 2030 capacity is a third axis. Three operator archetypes are consolidating, each carrying a different cost-of-capital and obsolescence exposure. Hyperscalers self-fund the largest share, financing from cash flow and carrying the depreciation debate on their own books. The top-four US hyperscalers alone have guided to roughly $730–760B of capex in 2026 (company guidance through the Q2 2026 prints; Amazon raised its guide to ~$220B on 2026-07-30, with CEO Jassy naming higher memory cost, not more megawatts, as a driver), with the Oracle-inclusive top-five cluster tracked at ~$750B+ (CreditSights). Neoclouds are the pure-play GPU landlords: capital-intensive, debt-financed, margin-pressured, and the most exposed to a utilization or residual-value shock because their entire asset is the depreciating silicon. Sovereign and enterprise buyers — nation-states pursuing compute independence (SK Telecom's 2 GW Vera Rubin DSX factory in Korea, first unit 2027, is the APAC instance), regulated industries pursuing data residency — are the fastest-broadening demand source, and the one least governed by pure unit economics. A fourth archetype is emerging in 2026: the energy/launch integrator selling time-to-megawatt — SemiAnalysis reads SpaceX's earnings-call plans as a path to ~10 GW of ground compute by year-end 2027 (with Microsoft the logical largest offtaker), a structure that is neither hyperscaler nor neocloud. Apply the announced-vs-under-construction test to every such figure before underwriting it as energized.

The geographic re-map follows from where firm power and permissive policy actually exist. Power-first training campuses migrate to stranded-generation and cold-climate regions; latency-first inference clusters stay in the metros. Sovereign demand and export controls re-map the map again: the (later rescinded) US AI-diffusion tiering, Gulf-state build-outs converting energy wealth into compute, and the empirical reality that residency is not control — a study of 775 non-US data centers found sovereignty depends on control-of-stack, not just where the concrete sits (arXiv, 2025). The consequence for a 2026 siting decision: the cheapest-power, fastest-permit jurisdiction may carry a geopolitical or export-control tail risk that a spreadsheet does not price. → grid integration in Chapter 15.8.

Every 2030 scenario eventually collides with the same wall: where do the firm, clean megawatts come from, and when? Global data-center electricity demand is projected by the IEA to roughly double from ~485 TWh in 2025 to ~950 TWh by 2030 (~3% of global electricity), with AI-specific load tripling (IEA, 2026). In the US, data centers move from ~4-5% of electricity today toward 9-17% by 2030 on EPRI's scenarios. The endgame is a ranking of supply options by the one variable that actually gates a build: time-to-megawatt, not headline LCOE.

Ranked that way, the clean-and-firm options that look best on a carbon spreadsheet are the slowest to energize. Firm, clean supply available now includes operating nuclear, hydro, and geothermal where resource geography and uncontracted capacity permit; restarts are project-specific future supply, with the 835 MW Crane Clean Energy Center serving Microsoft expected to return in 2027. SMRs are real but late: Google–Kairos targets first deployment in 2030 and a fleet by 2035 (Kairos Power, 14 October 2024), while Amazon’s Cascade targets operations in the 2030s and the wider X-energy ambition runs to 2039 (Amazon, 16 October 2025), and the 2030 dates are widely judged optimistic. Fusion is optionism: SPARC is the demonstration machine and ARC the generating plant, and Commonwealth Fusion targets grid electricity from its first ARC plant in Virginia in the early 2030s (CFS, Aug 2026) — so no one should underwrite a 2030 build on a fusion megawatt. That leaves gas-and-grid — behind-the-meter gas (~90 GW announced cumulatively by mid-2026, though only ~1 GW under construction) and host-utility grid service (project-specific tariff, studies, upgrades, agreements and milestones) — as the default bridge that actually carries the load this decade, with fuel emissions that remain in the inventory; CFE procurement and any separate offset claim do not erase the gas combustion. → speed-to-power mechanics in Chapter 3.2.

Carry the combined shock through the power portfolio. Use the 100 MW announced / 40 MW financed / 20 MW energized phase ledger in Chapter 16.1, not another forecast of delivered capacity. If power slips six months while useful-service demand drops to 50% and the assumed residual falls 25%, suspend the next fit-out release, re-price equipment commitments and preserve only the power rights whose holding cost clears the downside case. For each grid, gas, nuclear or storage option, record contracted MW, earliest deliverable date, duration, fuel or energy dependency, permits, counterparty, termination payment and carbon boundary. Announced generation is not firm supply; storage needs recharge, and a clean-energy certificate is not a dispatch right. Release the next phase only after the power milestone, funding and accepted customer workload return together; the case’s 60% useful-service hurdle is one explicit management choice. An SMR or fusion target moving closer on a slide does not satisfy those gates. → operating flexibility offers in Chapter 15.8; financing consequences in Chapter 2.5.

The entire demand curve rests on assumptions that a single shock could invalidate — and the wildcards cut both ways, which is what makes them dangerous to ignore. On the downside: an algorithmic-efficiency step (a DeepSeek-style training- or inference-cost collapse) that deflates the compute needed per unit of value; an enterprise-adoption air-pocket (the MIT NANDA study found ~95% of enterprise GenAI pilots produced no measurable P&L impact on $30-40B of spend); or a regulatory/energy backlash that caps siting — no longer hypothetical, now that New York's Executive Order 62 (Jul 2026) has paused state permits for new ≥50 MW hyperscale sites for up to a year (→ Chapter 3.2). On the upside, the same efficiency gains can increase total compute under the Jevons rebound condition — cheaper inference begets more inference — and reasoning/agentic workloads are a structural demand multiplier that could keep the supercycle alive longer than the bears expect (→ Chapter 16.3).

The decision consequence is asymmetric. A pure-play, single-generation, debt-financed asset is fragile to every downside wildcard and only benefits from the upside ones if it is already full. A flexible substrate — over-provisioned floor loading and water, modal procurement, contracted-not-merchant revenue, a power deal that flexes — is the cheapest insurance against the wildcard you cannot predict: what it buys is the right to be wrong about the forecast.

Each scenario hides a different stranded-asset failure mode, and naming them is the point of the whole exercise. Stranded silicon is the obsolescence risk: the contested 2–3-year obsolescence bear case vs published 4–6-year useful-life estimates and 5–6-year book policies means, if the bear case materializes, a generational density step — the ramp from ~132 kW NVL72 racks toward ~600 kW Kyber-class racks — can leave current-generation fleets uneconomic against newer parts and uncompetitive against grid power before they are depreciated. Stranded substations is the power-delivery mirror image: an interconnection slot energized for a campus whose demand never materializes, or a behind-the-meter gas plant carrying carbon liability for a load that digested. Stranded balance sheets is the financial endgame: long-lived debt underwritten against short-lived silicon, the under-depreciation question that turns reported earnings into borrowed future write-downs (→ Chapter 1.8).

The systemic and societal constraints are the outer boundary on all three scenarios. Grid reliability is now a planning constraint, not a footnote — NERC issued a rare Level 3 alert (2026) after large-load loss events (~1.5 GW dropped across a six-fault, 82-second sequence in one July 2024 Virginia event), and ride-through is now a recommended essential action under that alert — not yet an enforceable, penalty-backed standard. Water, community social license, and the concentration of load in a handful of jurisdictions (Virginia alone projected at 39-57% of state electricity by 2030 — EPRI, Powering Intelligence 2026) are real ceilings that a demand curve cannot wish away. The deepest systemic risk is correlation: every scenario assumes the constraints relax independently, but a power slip, a residual-value shock, and a demand air-pocket are correlated — they tend to arrive together in a downturn, which is exactly when a levered, single-generation, merchant-exposed asset has the least room to survive them. The dual-use framing matters here too: the same failure modes that random faults trigger are attacker-induceable (→ Chapter 11.10).