After decades of being dismissed as perpetually 30 years away, fusion energy—power generated through nuclear fusion reactions—is increasingly viewed as a question of when, not if. Investor and public interest in fusion is driven in large part by surging electricity demand from the A.I. boom and the rapid expansion of data centers across the U.S. As Big Tech companies search for reliable, carbon-free baseload power, fusion has reemerged as a long-term solution with growing near-term momentum.

In September, Type One Energy, a startup backed by Bill Gates’ Breakthrough Energy Ventures, announced plans to develop a 350-megawatt electrical (MWe) fusion power plant in Tennessee in partnership with the Tennessee Valley Authority (TVA), the largest public power utility in the U.S. Type One Energy is one of dozens of companies racing to be first to bring fusion power into commercial operation in the U.S. Helion, a startup backed by Sam Altman, Peter Thiel and Reid Hoffman and valued at $5.4 billion, has set an ambitious goal of beginning commercial operation in three years. Startups such as Pacific Fusion and Proxima Fusion have each raised more than $100 million, while Commonwealth Fusion Systems recently announced collaborations with Nvidia and Siemens to apply A.I. to fusion development. Thea Energy, meanwhile, has completed a pre-conceptual design for a fusion power plant.

Despite the flurry of announcements, the central challenge facing the sector remains unchanged: demonstrating that fusion can generate electricity at a commercial scale and at a competitive cost. With global energy demand expected to at least double in the coming decades, industry leaders argue that fusion is not competing for a fixed slice of the energy market. “The market is big enough for all energy industries,” Matt Miles, senior vice president of marketing and external affairs at Type One Energy Group, told Observer. Fusion, he added, isn’t “fighting over the same megawatts” as existing sources such as fossil fuels, solar or wind. “Once the first fusion plant comes online, we can expect to see a dramatic capital influx,” Miles said.

How much will fusion energy cost?

Because no fusion power plant is currently in operation, both short- and long-term capital costs remain uncertain. Some estimates place costs as high as $8,000 per kilowatt (kW) by 2050. Still, modeling from fusion experts suggests that under favorable market conditions, even capital costs around $7,000 per kW could allow fusion to reach 100 gigawatts (GW) of capacity—roughly matching today’s U.S. nuclear fleet. In less favorable scenarios, where alternative energy technologies continue to scale aggressively, fusion costs would need to fall to less than half that level to achieve similar penetration.

A significant driver of fusion’s high costs is the extreme engineering environment inside a reactor. Reliable supplies of advanced materials, including tungsten alloys, silicon carbide composites, high-temperature steels and graphene-based coatings, are required to withstand intense heat and constant neutron bombardment. “In the race to unlock fusion and reinvent fission, graphene may be the missing material,” Kjirstin Breure, CEO of HydroGraph, told Observer. Graphene’s exceptional thermal conductivity helps prevent overheating, while its strength and radiation resistance protect reactor components from cracking and degradation, she explained.

Funding, however, remains the sector’s central constraint. Most of the capital invested to date in the fusion landscape is going into two primary technological approaches: Magnetic Confinement Fusion Energy (MFE) and Inertial Confinement Fusion Energy (IFE), said Christoph Frei, partner and head of energy at Emerald Technology Ventures. MFE, which uses powerful magnetic fields to confine superheated plasma, accounts for the majority of startup activity and has attracted roughly $30 billion in public and private funding. IFE, which relies on high-energy lasers to compress fusion fuel, accounts for approximately 20 percent of startup activity and around $7 billion in total investment.

Energy generation is only one piece of a much larger system. Grid infrastructure and energy storage pose equally significant challenges for emerging power technologies. In the U.S., the interconnection queue has become a major bottleneck, with more than 2.6 terawatts of generation and storage capacity seeking grid access. This represents more than twice the total installed capacity of the existing power fleet.

Miles argues this constraint may also present an opportunity. A U.S. Energy Information Administration report found that utilities plan to retire 12.3 gigawatts of generating capacity in 2025, a 65 percent increase from 2024, with coal accounting for two-thirds of those retirements. In 2022, the Department of Energy identified 157 retired coal plants and 237 operating coal plants as potential sites for “coal-to-nuclear transition.” These locations already have transmission infrastructure and grid connections in place—assets that fusion plants will require. Type One Energy’s proposed Tennessee project, planned for the former Bull Run fossil plant site near Oak Ridge, exemplifies this strategy. “We will plug in to the existing grid,” Miles said. “We won’t need any re-shuffling of how the grids are operated.”

This past summer, an MIT report rattled the business community with its finding that 95 percent of enterprise A.I. applications fail to deliver the revenue growth companies expect. A newer Wharton study, released in October, reached a similar conclusion, noting that it is still “too early” for most large organizations to see measurable gains from A.I. Even so, long-term optimism remains high, with 88 percent of the Wharton study respondents saying their organizations expect to increase A.I. spending next year.

“The narrative that A.I. can’t deliver business impact is misleading,” Adam Gabrault, CEO of Solvd, a software and digital infrastructure firm, told Observer. In July and August, Solvd surveyed 500 U.S. CIOs and CTOs from companies with annual revenues exceeding $500 million and found that nearly 60 percent reported business benefits from A.I. in specific business departments, such as predictive analytics, customer support, HR and data management.

Companies that use A.I. effectively tend to align it with clear goals and follow long-established digital transformation practices. These approaches have guided successful tech adoption since the rise of personal computers and the shift to cloud computing.

“There’s a huge amount of pressure from all sectors, and all industries, to figure out how A.I. could be a change agent,” said Gabrault. The first step, he said, is tying A.I. to a specific objective—reducing customer churn, improving support or lowering costs. The “think big, take small wins” mindset applies here as well. Companies that see returns on A.I. don’t try to use it to “solve all problems,” Gabrault added.

Deploying A.I. on top of legacy systems and poor-quality data is often futile. An insurance company still relying on 30-year-old systems to write policies and manage claims, for example, will struggle to make any A.I. platform work. “To even get to a place of A.I. adoption, companies need to start looking at their data stack and how to make it A.I.-ready,” Gabrault said.

“This is where most A.I. initiatives actually die, not from bad algorithms, but from the unglamorous reality of messy data and systems that weren’t designed to share information,” Bakul Banthia, co-founder of Tessell, an A.I.-native enterprise data platform, told Observer. A.I. models run best on complete and consistent data, he said. While bridging data silos and cleaning up databases takes time, systems can be connected through APIs, and automated tools—with human oversight—can help improve data quality.

“Once you start building that kind of infrastructure, then we’re starting to see the acceleration of A.I. adoption significantly change,” said Gabrault.

Navigating governance and regulation

Governing A.I. is complex. As a new technology, it lacks a well-established framework, leaving companies to navigate largely uncharted territory.

“The only real answer is for companies to be thoughtful and ethical around how they use A.I. in their business and to continue to monitor and reform their governance,” Steven Pappadakes, founder and CEO of NE2NE, an automation and data integration company, told Observer.

Privacy and data protection should be top priorities, Pappadakes said. Building a strong relationship with an A.I. provider can help companies understand the technology and train internal teams. As new regulations emerge, staying informed is essential, he added.

Companies should also be aware that regulators like the SEC have lost patience with A.I.-washing—the practice of overstating a product’s A.I. capabilities. A.I.-washing can lead to legal consequences, fines and lasting reputational damage.

In the U.S., while federal regulators have been cautious about imposing broad A.I. rules, most states have already enacted or plan to enact some form of A.I. legislation. More are on the way. Companies operating in Europe face a more complex compliance landscape, with new laws such as the EU AI Act taking effect. “Those organizations that have that structure and framework ready are going to be in a much better position than those that do not,” Gabrault said.

Highly regulated sectors such as finance, banking and health care must involve strong compliance teams from the outset. These teams must vet A.I. projects, approve deployments and track new rules across local jurisdictions. Companies that plan for compliance early will be better prepared as new A.I. regulations inevitably emerge, Gabrault said.