The next constraint on artificial intelligence may not be a better model or a faster chip. It may be the ability to deliver enough reliable electricity to the data center running them.
That pressure is pushing technology companies toward an industry many investors once considered too slow for Silicon Valley: nuclear power. Large technology and cloud companies are exploring advanced reactors and small modular reactors, or SMRs, as potential sources of round-the-clock, low-carbon electricity.
At the same time, several publicly traded advanced-nuclear developers have experienced sharp stock declines. An August 2026 Financial Times analysis, citing S3 Partners, estimated that short sellers had made about $2.1 billion betting against NuScale Power, Oklo and NANO Nuclear Energy. That figure is an estimate of trading gains tied to three companies—not an audited measure of realized profit and not a verdict on nuclear energy as a whole.
The tension is useful. AI creates a credible need for firm power. Advanced nuclear offers an appealing technical vision. But the distance between a promising reactor design, a signed agreement and commercial electricity delivered to a data center remains enormous.
Data centers need electricity continuously. AI training clusters and inference systems add specialized accelerators, dense racks, high-speed networking and intensive cooling. The issue is not only total annual energy consumption. A data center also needs dependable capacity at the right location, with sufficient transmission and predictable delivery.
Wind and solar can provide low-cost clean electricity, but their output varies. Batteries can shift energy across hours, yet long-duration coverage remains expensive at large scale. Natural-gas generation is dispatchable, but it produces emissions and can conflict with corporate climate commitments.
Nuclear plants generate firm, low-carbon power. For cloud and AI companies trying to secure electricity around the clock, that combination is difficult to ignore.
Small modular reactors are designed to produce less power per unit than conventional gigawatt-scale nuclear plants. “Modular” refers to the ambition to manufacture more components in controlled factory environments and assemble standardized units at a site.
In theory, smaller units could reduce the amount of capital committed before a project begins earning revenue. Developers could add modules as demand grows, locate reactors nearer industrial loads and learn through repeated construction.
But modularity does not automatically guarantee low cost. The economics depend on standardized designs, repeat orders, manufacturing scale, licensing, financing and construction performance. A first-of-a-kind project can still be expensive because it must establish a supply chain, workforce and regulatory path.
“SMR” also describes a wide range of technologies. Some designs use water-cooling methods related to today's reactor fleet. Others use liquid metal, gas or molten salt. Their fuels, temperatures, waste streams and commercial maturity differ significantly. Treating them as one uniform product hides the engineering detail that matters.
There is genuine progress, but commercial readiness varies.
NuScale has a US Nuclear Regulatory Commission-certified SMR design, an important regulatory milestone. Certification is not the same as an operating commercial fleet; projects still require customers, sites, permits, financing, construction and operation.
Oklo is developing fast-fission powerhouses and pursuing licensing and site work. NANO Nuclear is developing microreactor concepts and has announced studies and agreements, including work related to potential AI data-center power. These companies are at development stages where engineering progress and future commercial expectations can exist alongside limited current revenue and substantial spending.
Outside the United States, small reactors have operated or connected to grids in countries including Russia and China. That means it is too broad to call every form of SMR technology unproven. The narrower and more accurate statement is that many of the specific designs attracting US AI-related investment have not yet demonstrated repeatable commercial deployment at scale.
Energy announcements use terms that sound similar but represent very different levels of commitment:
A partnership announcement can be strategically meaningful while remaining years away from electricity. Readers and investors should ask which milestone has actually been reached, which obligations are binding and who carries the development risk.
Many advanced US reactor designs require high-assay low-enriched uranium, known as HALEU. The US Department of Energy defines it as uranium enriched to between 5% and less than 20% uranium-235. That higher concentration can allow smaller cores, longer operating cycles and different reactor designs.
The constraint is supply. DOE says most advanced US designs need HALEU, but it is not yet available at commercial scale. The department estimates domestic demand could reach 50 metric tons per year by 2035 and is supporting enrichment, deconversion and demonstration pathways.
A reactor cannot operate on investor enthusiasm. It needs qualified fuel fabricated to exact specifications, transported securely and available on schedule. Fuel availability can therefore determine whether an otherwise viable project reaches operation.
Nuclear projects face demanding safety and environmental reviews for good reason. A developer may need approvals related to the reactor design, construction, operation, security, emergency planning, water use and environmental effects.
The site must also work as infrastructure. A reactor serving a data center needs transmission or a suitable direct connection, cooling, physical security, qualified operators and a plan for spent fuel and eventual decommissioning. If the data center and reactor are built on different schedules, one asset may wait for the other.
Grid interconnection is another source of delay for many energy projects. Locating generation near a large load may reduce some transmission constraints, but it introduces complex regulatory and commercial questions about who owns the assets, how backup power works and how electricity is delivered when either facility is offline.
AI enthusiasm can compress a decade of infrastructure risk into today's company valuation. The story is attractive: data centers need huge amounts of clean power; a developer has an advanced reactor; therefore rapid revenue appears inevitable.
Infrastructure does not move on software timelines. Reactor development requires capital, regulatory work, fuel, supply chains, sites and construction. Early-stage companies may spend heavily for years before generating recurring electricity revenue.
That does not make them fraudulent or their technology useless. It means a company's market value can become disconnected from the probability, timing and economics of deployment. A falling stock price can reflect changing expectations, dilution, execution risk or speculative excess. It does not, by itself, prove that the underlying technology failed.
The likely answer is not nuclear instead of every other technology. Large AI campuses may combine grid power, renewables, storage, demand management, natural gas and, eventually, advanced nuclear. The right mix depends on location, timeline, climate goals, reliability and cost.
Small modular reactors deserve serious attention because the need they address is real. AI infrastructure, industrial electrification and aging grids are increasing demand for dependable low-carbon power.
They also deserve scrutiny because energy projects must work in the physical world. Software can be updated after deployment. A nuclear plant must prove safety, financeability, constructability, fuel security and operational performance before it becomes part of the grid.
The strongest case for advanced nuclear is not that it will power every data center next year. It is that a successfully licensed, repeatably manufactured and commercially competitive reactor could become one valuable component of a broader clean-energy system.
AI may accelerate that future by creating customers with enormous, persistent power demand. It cannot eliminate the engineering steps required to reach it.
I help businesses design and build dependable digital products, backend platforms and technology infrastructure with a focus on what works beyond the demo. To discuss an AI, cloud or data-driven project, email hello@massoda.me or contact me on WhatsApp.
Editorial note: This article is a technology and infrastructure analysis, not investment advice. Company milestones and financial data are identified as company-reported information.