In May 2023, Microsoft signed the first fusion power purchase agreement in history, committing to buy 50MW of electricity from Helion by 2028. In 2025, Google followed with a 200MW offtake agreement with Commonwealth Fusion Systems (CFS). In the same period, annual funding for the fusion industry exceeded $2.6 billion, with cumulative investment approaching $10 billion.
If you are an investor, these headlines can easily lead to one conclusion: fusion is almost here.
The problem is that this conclusion is off by at least an order of magnitude on the timeline. Fusion is not almost here. It has moved from “is this scientifically possible” to “can this be engineered economically.” The gap between those two stages is not a year or two. It is a joint constraint across fuel cycles, materials science, thermal engineering, and capital structure. This article is not meant to tell you which company is best. It is meant to give you a framework so that when you face any fusion pitch, you can judge whether it is telling the truth, telling half the truth, or blurring the key issue.
The most important weapon in fusion investing is one letter: Q.
Most pitches will tell you that they are approaching Q>1, or that they have already achieved Q>1. But Q means very different things in different contexts.
The first layer is Q_plasma, plasma gain. The numerator is the fusion power produced. The denominator is the heating power input to the plasma. This definition does not include magnet cooling, cryogenic systems, vacuum pumps, tritium handling, or control system power. When most companies tell the media that they have achieved gain, they usually mean this layer.
The second layer is Q_sci, scientific gain. For magnetic confinement, it is basically a stricter version of Q_plasma that includes heating efficiency and impurity radiation. For laser inertial confinement, the equivalent metric is often Q_target: fusion energy output divided by laser energy delivered to the target. NIF, the US National Ignition Facility and the world’s largest laser fusion experiment, announced ignition in 2022 at this level: 2.05MJ of laser energy delivered to the target produced 3.15MJ of fusion energy, giving a Q_target of roughly 1.5.
The third layer is Q_eng, engineering gain. Here the denominator starts to include the plant’s internal power consumption: magnets, cryogenics, vacuum, laser charging losses, pulsed power, cooling pumps, tritium circulation, and control systems. This number is usually much lower than Q_plasma or Q_target. If a company only reports Q_plasma and never talks about Q_eng, that is a due diligence question.
The fourth layer is plant net electric. This is the only definition that directly matters for investors. It subtracts all internal plant power consumption and asks how much electricity actually reaches the grid. NIF-level target gain would likely need to reach the hundreds before engineering net electricity becomes meaningful. No private fusion company has publicly disclosed this number, including the companies that have signed PPAs.
The relationship across these four layers is simple: when you move down one layer, the number usually drops by more than an order of magnitude. The first rule of fusion due diligence is therefore: never compare Q values across different definitions. If someone uses NIF’s Q_target to analogize a tokamak company’s Q_plasma, or uses Q_plasma to imply plant-level net electricity, they either do not understand the definitions or are deliberately blurring them.
The second intuition to adjust is that the bottleneck has shifted.
From the 1950s to the 2000s, the central problem was plasma physics: could the product of temperature, density, and confinement time, known as the triple product, cross the Lawson criterion? The Lawson criterion is a rough threshold for the combination of temperature, density, and confinement time needed for fusion reactions to sustain themselves and produce more energy than the external input required to maintain the plasma. Over the past few decades, large scientific devices have progressively answered this question: Europe’s JET, once the world’s largest tokamak experiment; Japan’s JT-60; Korea’s KSTAR, a superconducting tokamak; China’s EAST, another superconducting tokamak known for long-pulse operation; Germany’s W7-X, the world’s largest stellarator experiment; and ITER, the international tokamak project under construction in France with a target Q of 10. Together, these devices built the deepest physics database for magnetic confinement. NIF answered the feasibility question on the inertial confinement path through target ignition in 2022. Plasma physics is not “solved,” but the remaining questions are increasingly about optimization and route-specific challenges, such as disruption control, confinement scaling at reactor scale, and alpha particle heating, rather than basic feasibility.
The real commercialization blockers are four things.
First, tritium. D-T fusion uses deuterium and tritium as fuel. Deuterium can be extracted from seawater and is not the hard part. Tritium barely exists in nature, has a half-life of only 12.3 years, and global commercial inventory is extremely limited. A commercial D-T fusion plant must breed its own tritium by using neutrons to hit lithium inside the blanket, the structure surrounding the reactor core that both absorbs neutron heat and breeds tritium. This is measured by TBR, the tritium breeding ratio, which must be greater than 1. Paper neutronics calculations can reach 1.1 to 1.5, but real engineering eats into that margin: diagnostic openings, structural materials, neutron leakage, tritium retention and permeation, extraction efficiency, and decay losses. No device has demonstrated TBR>1 at power plant scale.
Second, materials. D-T fusion produces 14.1MeV neutrons, which are extremely energetic. These neutrons continuously hit the first wall, the innermost structure facing the plasma; the divertor, the component that exhausts heat and impurities; and the blanket structure. This causes atomic displacement damage, helium bubbles, transmutation, embrittlement, and activation. Fission reactors have accumulated decades of irradiation data, but fusion neutron spectra are different, and existing irradiation facilities have not reached full reactor-relevant flux. Without material lifetime data, you cannot calculate plant capacity factor and maintenance cycles. Capacity factor, the fraction of time a plant actually generates power over a year, is a core driver of LCOE, the levelized cost of electricity over the plant’s lifetime. That is not plasma physics anymore.
Third, heat exhaust. The fusion plasma core is hundreds of millions of degrees, while the outer blanket and engineering structures must remain at manageable temperatures. The energy gradient between them has to be handled by the divertor, which sits in the vacuum vessel and removes waste heat and impurities. A tokamak divertor can see heat flux above 10 MW/m², comparable to spacecraft reentry heat loads, and it may need to handle that in steady or quasi-steady operation. Liquid metal divertors, detached plasma operation, and snowflake divertors are all being tested, but none has been integrated and validated at reactor parameters.
Fourth, capital structure. The first commercial fusion plant will require many billions of dollars in capital expenditure. That will not come from VC alone. It requires project finance, government loan guarantees, power offtake agreements, and EPC contracts. A 2025 CREO Syndicate report estimated that the first five 2GW-class fusion projects globally may require roughly $27 billion in project-level capital. No private fusion company is close to that financing structure yet. CFS and Helion’s PPAs are better interpreted as customer validation and future purchase options, not locked-in cash flow.
Fusion is not a single unified route. The differences between fusion approaches are larger than the differences between pressurized water reactors and sodium-cooled fast reactors in fission. You do not need plasma physics to understand them, but you do need one core trade-off: physics maturity versus plant complexity.
Tokamak is the traditional route. A toroidal vacuum chamber and strong magnetic fields confine D-T plasma in a ring. Tokamaks have the most experimental data, the largest physics database, and the most mature simulation tools. ITER’s Q=10 target is based on decades of scaling-law extrapolation, not guesswork. The downside is machine complexity: toroidal field coils, poloidal field coils, central solenoid, plasma heating systems, divertor, tritium blanket, and remote maintenance all have to fit into a donut-shaped machine. It is like a nuclear power plant and a particle accelerator combined. High-field approaches, such as CFS’s SPARC/ARC and Tokamak Energy’s spherical tokamak, use REBCO high-temperature superconducting magnets, a rare-earth barium copper oxide tape that enables compact high-field magnets, to reduce the scale of the machine. But higher field also means higher heat flux, higher neutron wall loading, and tougher materials requirements.
Stellarator is the tokamak’s steady-state cousin. It uses three-dimensional external coils to shape the magnetic field and relies much less on plasma current. This naturally avoids large current-drive requirements and disruptions. A disruption is one of the most dangerous tokamak failure modes: tons of plasma can lose control and hit the vessel wall within milliseconds. Stellarators avoid this and are more naturally suited to continuous operation. The trade-off is geometric complexity. Traditional stellarator coils look like twisted 3D sculptures, and manufacturing tolerances are tight. Modern optimization and supercomputing have made the coil shapes more manufacturable, with W7-X serving as the first large-scale validation, but assembly, maintenance access, and blanket integration remain difficult.
Laser inertial confinement follows a different physical path. Instead of using magnets to confine plasma for a long time, it uses powerful lasers to compress a tiny fuel target for nanoseconds, completing the burn before the plasma flies apart. NIF ignition proved the physics principle, but the way NIF works is far from a commercial power plant. NIF uses flashlamp-pumped glass lasers with efficiency below 1% and can only fire a few shots per day. A power plant would need lasers with more than 10% wall-plug efficiency, more than 10 shots per second, nearly a million targets per day, very low target cost, and precise target injection inside a debris-filled chamber. This is closer to an extreme precision manufacturing and automation problem than a continuation of a physics experiment.
Z-pinch is the most structurally radical route. It does not require large external magnets or a toroidal vacuum chamber. A strong current flows through a plasma column, and the current’s self-generated magnetic field compresses the plasma toward fusion conditions. Historically, the fatal problem was MHD instability: magnetohydrodynamic instability causes the plasma column to kink or pinch into segments under the interaction of current and magnetic field. Zap Energy’s approach is to inject sheared flow into the plasma and use the radial velocity gradient to suppress those instabilities. If this works, Z-pinch could have far lower plant complexity and capex than tokamaks or stellarators. If.
Magnetized target fusion (MTF) is a hybrid of magnetic and inertial confinement. The plasma is first magnetized, then rapidly compressed mechanically or electromagnetically. General Fusion’s approach injects plasma into a spinning liquid metal vortex and uses pneumatic pistons to compress the liquid metal wall. The liquid metal simultaneously acts as first wall, heat exchange medium, and tritium breeding material, which is an elegant engineering idea. The hard part is maintaining symmetry during sub-millisecond compression across hundreds of pistons, while also handling liquid metal corrosion and flow stability at high temperature.
Field-reversed configuration (FRC) is a compact toroidal plasma with no central column and a naturally linear geometry. Helion uses FRC for pulsed compression and direct electricity generation, where changes in magnetic field induce current directly and bypass the steam turbine. TAE Technologies uses neutral beams to sustain FRCs and eventually targets p-B11 aneutronic fusion. FRC’s appeal is small size and high beta, meaning higher plasma pressure for a given magnetic field. If direct conversion works, plant efficiency could go beyond conventional steam-cycle limits. The weakness is that FRC confinement stability and full energy accounting remain insufficiently validated, and Helion’s 2028 delivery commitment depends on repeatable, high-gain FRC operation.
p-B11 aneutronic fusion is the most attractive long-term route and the hardest near-term physics problem. A proton fuses with boron-11 and primarily produces three alpha particles, which are helium nuclei carrying positive charge and can in principle be converted electromagnetically into electricity. This avoids the tritium problem, greatly reduces high-energy neutron damage, and could remove the need for a steam turbine. But p-B11 requires temperatures more than ten times higher than D-T, and bremsstrahlung radiation losses, the energy lost when charged particles decelerate and radiate inside plasma, are extremely high. No lab has come close to the conditions needed for net positive p-B11 fusion. It is a long-dated option, not a 2030 grid story.
Once you understand the physical routes, the company landscape becomes easier. Fusion companies are not all racing on the same track. They occupy different positions in the stack and solve different problems. I divide them into four categories by credibility as power plant companies.
Tier 1: Full-stack power plant companies with real milestones. CFS is the only clear representative. It uses high-field tokamak physics, plans to prove net fusion energy with SPARC, its Q>1 validation machine, and then build ARC, a 400MW-class plant design. CFS has the most complete commercial signals: its SPARC magnet reached 20T DC field in 2021, it signed large offtake agreements with Google and Eni in 2025, and it selected a Virginia site. That does not eliminate risk. SPARC has not started operation. Q>1 is only the next gate, not proof of ARC economics. CFS still faces the tritium, materials, and divertor issues of the D-T route. It is simply the most transparent, best-funded, and most analytically tractable full-stack player today.
Tier 2: Differentiated routes with unproven core physics. Helion, Zap Energy, and Pacific Fusion fall here. Their common feature is that if they work, their plant architecture could be much cheaper than a tokamak. Helion’s pulsed FRC plus direct electricity generation could make plants far smaller. Microsoft’s PPA and more than $800 million of 2025-2026 financing are strong signals. But Helion’s 2028 delivery timeline means Polaris, its next-generation prototype, must prove net electricity, repeatable operation, licensing, and grid connection in less than two years. There is no precedent for that in nuclear energy. Zap’s sheared-flow Z-pinch completed a DOE-verified repetitive pulse milestone in 2025, with over 1000 plasma shots in three hours, and its Century platform is the first prototype to run repetitive pulses in a liquid metal environment. But it has not publicly disclosed triple product or Q progress. Pacific Fusion is pursuing pulsed magneto-inertial fusion and reportedly raised about $900 million, but with limited disclosure. These companies should be tracked, but valuation should reflect high-risk, high-upside physics uncertainty, not a 2030 grid assumption.
Tier 3: Long-horizon deep tech routes. Type One Energy, Proxima Fusion, Thea Energy, Tokamak Energy, and TAE Technologies belong here. Their physics logic can make sense over the long term, but multiple layers of technical de-risking and funding still sit between them and grid connection. Type One’s partnership with TVA is a strong signal because TVA is a real utility, and utility interest confirms that stellarator steady-state operation matters commercially. Proxima benefits from the German W7-X / IPP technical lineage and backing from RWE and the Bavarian government. Thea uses planar coil arrays instead of traditional 3D twisted stellarator coils, improving manufacturability. The near-term focus for these companies is not “what year can they generate electricity,” but magnet manufacturing, integrated device construction, steady-state operating data, and maintenance engineering.
Tier 4: Supply chain companies that do not rely on selling electricity. Kyoto Fusioneering may be one of the most underappreciated categories. It does not build reactors; it builds fusion plant engineering systems and key components: tritium fuel cycle, blankets, gyrotrons for plasma heating, and test facilities. SHINE Technologies already has revenue from neutron sources and medical isotopes, and its Chrysalis facility received a $263 million DOE conditional loan, but that revenue comes from isotopes, not electricity. Treating SHINE as a fusion power company is a different thesis from treating it as a neutron-source and isotope company. HTS tape and magnets, cryogenic systems, pulsed power components, target manufacturing, remote maintenance robotics, and engineering services do not depend on one reactor company successfully connecting to the grid. These are the areas where meaningful revenue can appear within 3 to 7 years.
If a fusion company sends you a pitch deck, prepare these questions before opening it.
Gate 1, physics. “Which gain definition are you using? Q_plasma or Q_eng? What exactly is in the numerator and denominator? Are temperature, density, and confinement time achieved simultaneously? Do you have independent Thomson scattering diagnostics, neutron yield, and radiated power data? What is the shot-to-shot repeatability?” Thomson scattering matters because it is a standard way to infer plasma temperature and density by firing a laser through the plasma and reading the scattered light.
Gate 2, engineering. “What is your divertor or heat exhaust solution? What is the heat flux at reactor parameters? What is the dpa rate and predicted lifetime of first-wall materials under a 14MeV neutron spectrum? Do you have HTS magnet quench protection, mechanical fatigue, and radiation degradation data? For pulsed routes, what are the driver efficiency, repetition rate, chamber clearing time, and critical component lifetime?” Quench protection matters because a superconducting magnet can suddenly lose superconductivity and dump a large amount of stored energy into the magnet. If a company has not tested materials at reactor-relevant fluence, meaning cumulative neutron exposure, its capacity factor model is guesswork.
Gate 3, fuel cycle. “How is TBR calculated for the D-T route? Have you deducted losses from diagnostic openings, structural materials, and ports? Are tritium extraction efficiency, permeation losses, processing delays, and decay losses included? Where does the startup tritium inventory come from? For advanced fuels, what is the fuel supply path and what are the side-reaction neutron spectra?”
Gate 4, economics. “What is the FOAK plant CAPEX range? Where do the capacity factor and availability assumptions come from? Is the PPA binding or an MoU? Does it include price, capacity, penalties, delivery conditions, and grid interconnection approval? What technical milestone does the next financing round close, or is it merely financing the next building?” Capital path is one of the most overlooked but fatal issues in fusion investing. A FOAK fusion plant is likely more expensive than a single VC fund. Without government loan guarantees or major utility co-investment, companies can get stuck in the gap between “physics proven” and “engineering too expensive to build.”
Gate 5, regulation. “Does the jurisdiction have a licensing framework for fusion facilities? When will the US NRC’s Part 30 framework be finalized? What is the regulatory pathway for tritium inventory, activated materials, decommissioning, and transport? Will Agreement State compatibility affect siting?”
These questions do not require a nuclear physics background. Any vague answer is a place to discount valuation.
Fusion is not a race that is almost finished. It is more like a multi-level bridge being built at the same time. The bottom physics layer is mostly paved, but the layers above it, including fuel cycle, material lifetime, heat exhaust, capital structure, and regulatory path, can each block the whole project.
From an investment perspective, the right strategy is not to bet blindly on who connects to the grid first. It is to identify which layer each company is attacking. CFS has the strongest full-stack commercial evidence, but investing in it means accepting nuclear-engineering timelines and multi-billion-dollar follow-on capital needs. Helion and Zap have large upside from non-consensus architectures, but physics uncertainty must be priced into valuation. Stellarator companies can be revisited over the next three to five years; their steady-state advantage is real, but the risk profile is different from CFS.
For most investors who do not need to bet on a single route, the supply chain is the more reasonable entry point. HTS tape, cryogenics, pulsed power, tritium handling, target manufacturing, remote maintenance, and engineering services share one feature: demand does not depend on any single fusion company reaching the grid. Their revenue curves can also be supported by adjacent markets such as semiconductors, MRI, aerospace, defense, and advanced fission. Each area corresponds to a concrete de-risking milestone rather than a distant grid connection date.
Fusion’s investment value is real, but it is not “fusion electricity by 2030.” It is that the industry is becoming an industrial ecosystem with real R&D spending, real engineering procurement, and real supply chain demand. The layer you enter determines the shape of your risk-return function.