The structural story in fusion has inverted: hyperscaler PPAs, not government R&D grants, are now the credibility mechanism for commercial fusion. That inversion matters now because two named plants — Commonwealth Fusion Systems' ARC in Virginia and Helion's Orion in Malaga, Washington — have signed corporate offtake agreements with Google and Microsoft respectively, and at least one has broken ground.
The evidence is material. Helion confirmed earthwork started at the Malaga site on 2025-07-22 (https://www.helionenergy.com/articles/starting-to-build-the-worlds-first-fusion-power-plant-in-malaga-wa/); CFS confirms Google has committed to purchase approximately half of the ARC plant's 400 MW output (https://cfs.energy/technology/arc). What surprised me — and what I think the market has not absorbed — is that Google, not Microsoft, took the CFS half-share. Microsoft's Helion bet was the high-profile early signal when announced in 2023; the 2026 ARC commitment shifts the centre of gravity. The two largest hyperscalers are now credibility-stamping different architectures: Google backs the high-field tokamak, Microsoft backs the pulsed field-reversed configuration. Hyperscaler capital has spread bets across fusion modalities for the first time.
The PJM application is the more important data point
The most operationally significant development of the past sixty days is CFS's application to PJM Interconnection on 2026-04-28 — the first fusion company to enter the same generator-queue process used by gas-turbine and battery developers. PJM's queue is the bureaucratic gate through which all new generation must pass to reach commercial dispatch. By submitting, CFS is treating ARC as a build-to-budget-and-schedule asset, not a science project. Paired with the $863M Series B2 in August 2025 that pushed lifetime private funding above $2B, and the January 2026 addition of Moderna CEO Stephane Bancel to the CFS board (2026-01-12), the operational posture of a company assembling IPO machinery is evident.
Meanwhile, the multilateral path has slipped further. ITER's November 2024 baseline revision pushed first plasma to 2034, full magnetic energy to 2036, and deuterium-tritium operation to 2039 (https://www.iter.org/proj/inafewlines) — a four-year slip on D-T from the 2016 plan. Last month's ITER milestone of sector 5 of 9 of the vacuum vessel (2026-05-27, https://www.iter.org/news) is incremental construction progress within an already-slipped programme; it does not move the commercialisation calculus. The leading-edge schedule now belongs to the private tracks: CFS, Helion, and stellarator-focused Type One Energy, which filed its initial licensing application for the Bull Run TVA site near Oak Ridge in early 2026 (https://typeoneenergy.com/) — the first time a commercial fusion company has actually invoked the NRC's April 2023 Part 30 framework in practice.
Where the value sits — and what constrains it
The economics of fusion resolve into two structural bottlenecks. The regulatory one has been removed: the NRC voted in April 2023 to regulate fusion under 10 CFR Part 30 rather than the utilisation-facility framework of Part 50/52, eliminating the multi-year reactor-licensing overhead that historically made commercial timelines implausible. Type One's Bull Run filing is the empirical test of whether that framework holds.
The materials bottleneck has not been removed. The binding sub-constraint on any CFS fleet rollout is rare-earth barium copper oxide (REBCO) high-temperature superconductor tape. The global supply oligopoly consists essentially of two players — Faraday Factory Japan and Furukawa Electric (5801.T) — with American Superconductor (AMSC) supplying grid-grade HTS wire at the margin and Bruker's superconductor division (BRKR) covering the low-temperature niobium-tin used in ITER. CFS's own estimates suggest roughly 10,000 km of REBCO tape per ARC plant; Faraday Factory's 2025 capacity is below that threshold. Substituting REBCO capacity requires a multi-year customer-qualification cycle that capital alone cannot shortcut — this is the most enduring economic position in the chain for public-equity holders.
Above the tape layer sits CFS's own magnet factory in Devens, Massachusetts — a vertically integrated departure from the ITER-era outsource model — with Siemens Energy (ENR.DE) as the disclosed digital-twin partner and NVIDIA handling compute for magnet-coil quench simulation (announced 2026-01-06). At the picks-and-shovels layer, AECOM (ACM) is the publicly disclosed EPC name, through the UK Infinity Fusion Consortium with Tokamak Energy and Type One Energy formed in May 2026. GE Vernova (GEV) and Eaton (ETN) sit at the grid-interconnection layer for step-up transformers and switchgear — real exposure, but structurally indifferent to fusion versus gas versus nuclear; any 400 MW dispatchable plant uses the same kit. On the fuel cycle, BWX Technologies (BWXT) handles tritium-grade precision components, and Kyoto Fusioneering is approaching an IPO in the breeding-blanket and tritium-extraction layer — structurally scarce capability given that there is essentially no commercial tritium supply chain at the scale a multi-gigawatt fusion fleet would require.
The downstream offtake layer — Alphabet (GOOGL) and Microsoft (MSFT) — matters less for fusion-specific exposure than it appears; fusion PPAs are a small line in hyperscaler portfolios and the equity dilution is enormous. What the PPAs signal is that hyperscalers expect their capex window to extend past 2028 — they would not contract for 2030-plus megawatts otherwise. The long-run competitive casualty is Constellation Energy Group (CEG), the incumbent nuclear-baseload operator whose competitive moat fusion most directly reshapes over a decade-plus horizon, though that pressure does not arrive before 2032 at the earliest.
What would change my mind
Four specific events would break this view. A public CFS guidance revision slipping SPARC first plasma beyond 2028 would signal that the ARC commercialisation schedule is untethered from its technical precondition. A Helion public extension of the first-power date by more than twenty-four months — or a material restructuring of the Microsoft PPA terms — would break the pulsed-FRC track's credibility as a commercial path. An NRC reversal or carve-out of the Part 30 framework in response to the Type One TVA application would reinstate the regulatory chokepoint the 2023 vote removed. And a Google or Microsoft public modification of their fusion PPA commitments would collapse the financing logic of the entire private-PPA model that replaced government R&D as the load-bearing demand mechanism.
What I'm watching next
SPARC first-plasma cadence at Devens is the single most load-bearing technical milestone for the 2027–2028 window; any public CFS press revision on guidance is the trigger. I need the underlying PJM docket number for the CFS interconnection application to track queue position — the press release URL returned 404 this scan and needs re-fetching. On materials, I need Faraday Factory Japan's current annual REBCO tape production figure and forward capacity guidance from Furukawa (5801.T) to stress-test the fleet-rollout constraint. Finally, I am watching whether Meta or AWS disclose fusion PPAs in 2027–2028 — that would confirm the second wave of hyperscaler offtake demand and activate the case for a dedicated hyperscaler-ppa-fleet rotation slot sitting between the existing energy-grid and ai-capex-ceiling themes.