Energy · Deep dive

Commonwealth Fusion Systems

MIT PSFC spinout building the SPARC compact high-field tokamak in Devens, MA and the 400MW ARC first commercial fusion plant in Chesterfield County, VA — raised ~$4B by July 30, 2026 (including a $1B pension/sovereign-wealth round and a $863M August 2025 Series B2 with Nvidia, Google, Khosla and Breakthrough Energy Ventures), with first plasma slipped from 2025 to 2027 and first grid power to the early 2030s.

emerging

The question that decides it: The bet is that a compact high-field tokamak beats every alternative on time and cost. **CFS wins if SPARC hits first plasma in 2027 and clears Q>10 by 2028; the Devens HTS magnet factory outputs the ~18 SPARC-class TF coils and the ARC set without a slip past 2030; ARC comes online in Chesterfield County before 2035 at a delivered $/MWh that beats a firmed solar-plus-4-hour-lithium stack (EIA models ~$126/MWh 2030 LCOS) and an SMR under decade-long NRC review; and it does all of this before Helion delivers Microsoft's 50MW in 2028 or China's CFETR proves a state-scale alternative in the 2030s. It fails if any one of first plasma, Q>10, magnet-factory throughput, ARC schedule, or grid-competitive LCOE slips a single year further** — the private-fusion capital base cannot fund another 2018-style multi-decade delay after ~$3B of dilution is already in the ground.

My take ★★ Very interesting ★ Interesting − Maybe ✗ Not interesting

Snapshot

Commonwealth Fusion Systems is the most-capitalized private fusion company on Earth. Spun out of MIT’s Plasma Science and Fusion Center on March 9, 2018, it is building SPARC (a compact high-field tokamak in Devens, MA, aiming to be the first magnetic-confinement device outside a government lab to net-produce fusion power) and ARC (a 400MW first-of-a-kind commercial plant in Chesterfield County, VA). It has raised ~$4B by July 30, 2026 — about 30% of all capital ever invested in fusion (Bloomberg, Jul 30, 2026): ~$50M seed (2018), $84M Temasek-led A2 (May 26, 2020), $1.8B Tiger Global Series B (Nov 30, 2021, still the largest private fusion round ever), $863M Series B2 with Nvidia (Aug 2025), and a $1B pension/sovereign round on Jul 30, 2026. It matters now because Google, Nvidia and Eni are pre-buying the electrons, SPARC first plasma has slipped from 2025 to 2027, and Helion is contractually obligated to deliver Microsoft 50MW in 2028.

Founding story

The seed is a 2015 MIT thesis. Whyte’s PSFC group — Mumgaard on Alcator C-Mod diagnostics, Sorbom on the ARC design paper, Hartwig on nuclear materials, Greenwald as elder statesman — argued that REBCO HTS tape wound into a full-scale TF coil would enable a net-energy tokamak ~40x smaller by volume than ITER. Smaller device, higher field; higher field, faster physics; faster physics, private-sector timeline. Mumgaard, an MIT translational fellow, spent 2015-2018 turning it into a company: a risk-retirement plan sequencing magnet before plasma, and a collaboration letting CFS work in MIT facilities while PSFC professors stayed. Announced March 9, 2018 with a ~$50M seed from Eni, BEV, Khosla, The Engine, Future Ventures.

How it works

SPARC is a compact tokamak — a doughnut chamber where D-T fuel is heated over 100 million degrees and confined by magnetic fields. What is new is the magnet. Tokamak performance scales roughly as the fourth power of field; ITER runs at ~12 tesla with low-temp niobium-tin. CFS wound REBCO (rare-earth barium copper oxide) tape into 16-pancake TF coils and on Sep 8, 2021 sustained over 20 tesla at Devens — the world’s strongest large-bore fusion-relevant magnet at the time (MIT News, 2021); peer-reviewed March 2024. Higher field means a smaller device you factory-build. SPARC needs ~18 magnets; ARC a similar set. Devens (announced Mar 2021) is an HTS magnet factory as much as a fusion R&D site — magnet throughput is arguably the gating constraint on ARC replication, not plasma physics. As of spring 2026, SPARC is ~75% complete with the first 20-ton magnet installed (Sorbom, 2026).

Product and business overview

Two devices, two products. SPARC is the Devens demonstration tokamak, designed to hit Q>1 (more fusion energy out than plasma heating in) and eventually Q>10, then run as a science/engineering testbed. It is not sold. ARC is the first commercial product: a ~400MW plant at the Fall Line Fusion Power Station in Chesterfield County, VA (announced Dec 2024, county permit approved 2025). Around those: HTS magnet manufacturing at Devens, and strategic partnerships — Google signed a 200MW PPA of ARC’s 400MW on Jul 1, 2025 (largest direct corporate fusion offtake ever); Eni signed a “billion-dollar” PPA (Latitude Media, 2025); Dominion Energy is a Virginia site partner.

Business model and pricing

CFS sells the promise of future electricity through PPAs whose $/MWh terms are not public. Google’s 200MW and Eni’s roughly billion-dollar PPA are the two disclosed contracts, both contingent on ARC generating. Revenue today is DOE Milestone-Program payments (award 2023) plus partnership prepayments. Thesis: ARC ships to hyperscalers for whom firm carbon-free baseload has strategic value at a premium; subsequent units drop $/MWh via factory-learning. No public LCOE — the sharpest analytical hole: ARC must land below firmed solar + 4-hour lithium (EIA 2030 LCOS ~$126/MWh) and below SMR.

Traction over time

Market analysis

TAM is electricity, re-accelerating on AI data-center load. The Fusion Industry Association tracks 40+ private fusion companies with $8B+ cumulative capital by 2025. Structural forces: (1) hyperscalers signing firm-carbon-free PPAs at premium prices for 2030 climate targets (Google, Microsoft, Meta, Amazon all have fusion, SMR or geothermal deals); (2) SMR is real but slow — decade-long NRC review; NuScale cancelled CFPP in Nov 2023; (3) firmed solar + 4-hour lithium is the price ceiling — ~$125/kWh installed ex-China, 2030 LCOS ~$126/MWh. ARC must beat firmed solar-plus-storage and SMR at scale by the mid-2030s.

Competitive intel

Helion is the timeline threat: Microsoft-contracted, broke ground July 30, 2025, targeting 50MW in 2028 under real-penalty PPA. If Helion meters electrons in 2028, CFS is second-mover. TAE Technologies (~$1.5B, Google-backed) is the aneutronic wildcard. Tokamak Energy (UK, $335M by 2024) is the closest technical analogue — spherical tokamak, HTS magnets — and its September 2024 spin-out TE Magnetics targets £300M annual HTS-magnet revenue into medical diagnostics, storage and maglev, exactly the adjacency CFS has not shipped. Type One Energy attacks from the stellarator side; Zap Energy ($327M) from Z-pinch; Focused Energy and Marvel Fusion from inertial confinement; Realta Fusion and Xcimer hold DOE Milestone-Program awards. First Light Fusion abandoned its power-plant plan on March 3, 2025 and pivoted to amplifier IP — the cautionary data point. ITER and CFETR are state-scale competitors on multi-decade timelines; ITER burning-fuel now 2039, CFETR commercial post-2035. Behind all of this, firmed solar-plus-storage and SMR are the actual price ceiling.

History and evolution

Mar 9, 2018: MIT spinout, ~$50M seed. 2018-2021: SPARC TF Model Coil program at MIT PSFC. May 26, 2020: $84M A2, Temasek-led. Mar 2021: Devens campus announced. Sep 8, 2021: 20-tesla HTS magnet test — the risk-retirement moment that made the $1.8B B fundable. Nov 30, 2021: $1.8B B, Tiger Global-led — still the largest private fusion round ever. Dec 5, 2022: LLNL NIF Q>1 (industry inflection). Mar 2024: HTS magnet peer-reviewed. Dec 2024: ARC site chosen in Chesterfield County. Jul 1, 2025: Google 200MW PPA. Aug 2025: $863M B2 with Nvidia. Spring 2026: First SPARC magnet installed; ~75% complete. Jul 30, 2026: $1B pension/sovereign round; ~$4B total. Stumbles are schedule creep — first plasma 2025→2027, ARC from “late 2020s” to “early 2030s”, commercial power 2032-2035.

What people say

The case for. MIT press, Wilson Sonsini’s Series B write-up and successive investor releases (BEV, Google, Nvidia) land on three points: (1) the 2021 magnet test was the hardest engineering de-risking any private fusion team has completed; (2) MIT PSFC lineage brings a physics bench no rival matches; (3) the SPARC-to-ARC industrial architecture — HTS coils at Devens, plants built in factory-repeatable units — is a credible LCOE path critics call impossible. Hyperscaler PPAs and the Jul 2026 pension round are the market’s own signal CFS has cleared a bar most fusion companies have not.

The complaints. “Fusion is always 30 years away” has not retired: ITER, originally $5B, now $22-65B, first operation slipped to 2034 and burning fuel to 2039 (Physics World, 2024). Daniel Jassby (ex-Princeton PPL) argues tokamak fusion’s engineering and tritium-cycle burdens make grid economics implausible; Hirsch and Bezdek argue fusion as conceived costs 10x+ conventional fission. LPPFusion and the late Robert Bussard argue the tokamak line is the wrong geometry — Bussard told Google engineers in 2006 that fifty years of US fusion was directed politically not technically. CFS’s own schedule slipped 2025→2027, compounding against Helion’s 2028 delivery. ARC LCOE is not public. Tritium supply is unsolved. And quietly: the Jul 2026 round leaned on pension and sovereign capital not venture — venture may be out of appetite, and pension capital is less tolerant of another slip.

Outlook: the open question

For CFS to win, all at once: SPARC first plasma in 2027, Q>1 by 2028, Q>10 shortly after; the Devens magnet factory delivering the SPARC and ARC coil sets on schedule; ARC breaking ground late 2020s and commissioning before 2035 without a further multi-year slip; a delivered LCOE below firmed solar-plus-4-hour-lithium (EIA’s ~$126/MWh 2030 LCOS) and below NuScale/X-energy-class SMR; and the pension-and-sovereign-wealth base willing to fund a third and fourth ARC before the first meters electrons.

For CFS to lose: any one of first plasma, Q>10, magnet-factory throughput, ARC schedule, or LCOE slips another year. The private-fusion pool has crossed $4B and cannot fund an ITER-style extension. Helion delivering Microsoft’s 50MW near 2028 rewrites the narrative. A competent state actor — China’s CFETR, or a re-committed US DOE — reaching plasma at scale in the 2030s makes CFS’s lead over the public sector, its whole reason to exist, look narrower than the pension-fund term sheet requires.

How to attack it

CFS’s weaknesses are not physics — that team is the strongest in private fusion. They are commercial and structural.

Weakness one: the HTS magnet factory is more valuable than the plant it feeds. Tokamak Energy figured this out in Sep 2024 and spun out TE Magnetics targeting £300M annual revenue into medical diagnostics, accelerators, storage and maglev (WNN, Sep 2024). CFS has not disclosed a magnet-monetization strategy. A funded attacker builds a REBCO-magnet-first company — 7T+ research MRI vs Siemens/GE/Philips, and FCC-class accelerator dipoles — hires an ex-CFS or ex-Tokamak Energy magnet lead, ships revenue in three years. This attacks the reason CFS must spend $4B before booking a watt.

Weakness two: unsolved tritium supply. Global CANDU-derived tritium is measured in kilograms and depletes over the 2030s; commercial D-T needs breeding blankets no one has demonstrated at scale. An attacker builds a tritium and blanket-materials supplier — the “TSMC of fusion fuel” — and every future plant is a customer. Kyoto Fusioneering and Marathon Fusion are early.

Weakness three: schedule slippage. SPARC first plasma 2025→2027; ARC late 2020s→early 2030s. Every year compresses the IRR the Jul 2026 pension capital was underwritten against.

Weakness four: no public LCOE. Every hyperscaler signing an ARC PPA is buying a strategic option, not a hedge on grid economics. An attacker publishing a credible sub-$100/MWh LCOE — Helion is trying — reframes the category.

Adjacent-segment play

The obvious adjacent play is REBCO HTS magnets sold outside fusion. CFS built the world’s strongest large-bore fusion-relevant HTS magnet at MIT PSFC in Sep 2021 and has the Devens footprint. Buyers — 7T/11T research MRI (Siemens, GE, Philips), FCC-class accelerator dipoles, next-gen proton-therapy synchrotrons, superconducting drivetrains, SMES — pay premium unit economics for coils fusion needs anyway. Tokamak Energy proved the model: TE Magnetics spun out Sep 2024 targeting £300M ($394M) annual non-fusion HTS by decade-end (WNN, Sep 2024). CFS has not done this. It should.

Second: tritium and blanket-materials supply. Every D-T plant after ARC needs breeding blankets no one has built at scale. Kyoto Fusioneering and Marathon Fusion are early entrants; CFS is closer to the bill of materials than any private peer. Third, speculative: fusion-derived process heat for hyperscaler data centers — a cryogenic tokamak co-located with compute delivers electrons plus near-zero-carbon cooling. What does not generalize: neutron-as-a-service, medical isotopes, space propulsion.

Sources and further reading

• Commonwealth Fusion Systems Raises Another $1 Billion, Bringing Total Capital Raised to $4 Billion — PR Newswire / CFS release, July 30, 2026
• Fusion power darling Commonwealth Fusion Systems raises another $1B — TechCrunch, July 30, 2026
• Commonwealth Fusion Systems Raises $863 Million to Bring Fusion Power to the Grid — ESG Today, August 2025
• Google and Commonwealth Fusion Systems Sign Strategic Partnership Agreement to Commercialize Fusion Energy — PR Newswire, July 1, 2025
• Commonwealth Fusion Systems Raises $1.8 Billion in Funding to Commercialize Fusion Energy — PR Newswire, November 30, 2021
• MIT-designed project achieves major advance toward fusion energy — MIT News, September 8, 2021
• ITER fusion reactor hit by massive decade-long delay and €5bn price hike — Physics World, 2024
• Fusion firm Tokamak launches specialist HTS magnet division — World Nuclear News, September 2024
• Chesterfield approves fusion power plant permit — Virginia Business, 2025
• Microsoft agrees to buy electricity generated from Sam Altman-backed fusion company Helion in 2028 — CNBC, May 10, 2023
• Achieving Fusion Ignition (Dec 5, 2022 through 2025 ignition shots) — Lawrence Livermore National Laboratory