Big Tech's Nuclear Bet: Key Small Modular Reactors for Cloud Power

Major cloud providers and data center operators made high-profile commitments in 2024 to integrate nuclear SMRs into their future energy mix. These deals pair tech companies with advanced reactor developers or utilities, aiming to secure dedicated nuclear power for data center campuses. Below, I summarize the most significant SMR initiatives announced to date, focusing on who is involved, what the project entails and when it's expected to deliver power.

Title: Visualization - Description: Visualization

Other notable developments include Meta (Facebook) soliciting proposals for up to 4 GW of new nuclear generation to support its AI goals [1]. (Meta ultimately inked a deal in 2025 to source energy from an existing Illinois nuclear plant, indicating interest in nuclear now while SMRs mature). Additionally, smaller players are exploring SMRs. For example, crypto data center firm Standard Power announced plans to deploy 24 NuScale SMR modules at two sites for blockchain and HPC workloads [2]. These underscore a broad industry trend: Data center operators of all stripes are "going nuclear," either via next-gen SMRs or by tapping existing reactors, to meet skyrocketing power needs without carbon emissions.

Realistic deployment timelines

Despite the flurry of announcements, no SMR is yet operational in the U.S., and all these projects face multi-year development and approval processes. The earliest realistic SMR online dates for data center use are around 2028–2030 for first-of-a-kind units, with most deployments clustering in the early to mid-2030s. Below, we break down the expected timelines for each major initiative:

Google–Kairos Power: Target 2030

Google's agreement with Kairos aims for the first Kairos SMR by 2030 [3]. This aggressive timeline assumes Kairos' Hermes test reactor (Oak Ridge, TN) — which received its NRC construction permit in late 2023 — stays on track for operation by 2027 [4]. If all goes well, commercial Kairos reactors for Google would follow by the end of the decade. Google anticipates additional units coming online through 2035 (up to ~500 MW total) [5].

AWS–X-energy/Energy Northwest (Washington): Target ~2032–2035

The AWS-funded SMRs in Washington are expected "online beginning in the early 2030s" [6] [7]. This likely means first power around 2032 (assuming ~eight-year project timeline including licensing and construction), with potential expansion to 960 MW later in the decade. X-energy's Xe-100 reactor design (80 MW modules) is still under NRC review; its first unit (a DOE demonstration with Dow Chemical in Texas) won't be decided on until ~2028 [8] [9], suggesting the Washington deployment will follow thereafter. A 2024 industry analysis noted that essentially every advanced reactor project is targeting "2030" as a start date, often with DOE support, but few if any will beat that date by much [10].

AWS–Dominion (North Anna, Virginia): No firm date (likely mid-2030s)

This project is at an exploratory MoU stage, with Dominion only beginning an RFP for SMR technologies in mid-2024 [11]. No specific reactor design has been chosen yet. If a technology is selected by 2025–26, the earliest operation would be late 2030s, given the need for design certification and a combined license for a new unit at North Anna. (Notably, Dominion already holds an early site permit for a large reactor at North Anna, but using it for an SMR would still require new NRC approvals.)

Microsoft & Three Mile Island: Target 2028

While not an SMR, Microsoft's plan to revive the dormant TMI Unit 2 reactor highlights a pragmatic shortcut to get nuclear power sooner. Constellation Energy aims to restart the 837 MW reactor by 2028 exclusively to supply Microsoft [12]. This timeline is plausible because the plant (a conventional PWR) had operated until 2019, so the project involves re-licensing and refurbishment rather than building an entirely new reactor. Microsoft's willingness to pursue this underscores that new SMRs would not be ready in time for its near-term AI data center needs. By late 2025, Microsoft also signed 24/7 nuclear energy deals for some Azure regions (e.g., sourcing existing nuclear in Virginia for its Boydton data center) [13], essentially bridging the gap until SMRs come online in future years.

Oklo Micro-Reactors (Equinix, etc.): Target 2027–2028 for first unit

Oklo Inc. is pursuing a fast-track approach for its very small (15–50 MW) reactors. After an initial licensing setback in 2022, Oklo is preparing a new combined license application in 2025 [14] [15]. The company ambitiously aims to commission its first 50 MW "Aurora" reactor by 2027 at Idaho National Lab [16], which would make it the first advanced reactor of any kind in the U.S., if achieved. Its power purchase agreements with Equinix, Prometheus and others hinge on this success. However, this timeline is optimistic; Oklo must still obtain NRC approval, and even with a streamlined process, that's a tight schedule. More realistically, a late-2020s startup (2028–29) could occur if Oklo's approach succeeds. The quantity of reactors implied by its deals (e.g., 500 MW for Equinix means ~10–15 units) would then roll out through the 2030s. Oklo's CEO has suggested that subsequent units could be licensed in as little as six months each under new NRC rules, once an initial design is approved [17] — a claim reflecting hoped-for regulatory reforms (discussed below).

Oracle's SMR Campus: Timeline TBD (likely 2030s)

Larry Ellison's announcement of a 1 GW data center campus powered by three SMRs [18] did not include dates. Given the early stage, it is likely to envision reactors that become available in the mid-2030s. Industry observers note this proclamation as symbolic, indicating shifting attitudes, rather than a project with concrete timelines yet [19]. Oracle would still need to partner with an SMR vendor (be it NuScale, GE Hitachi's BWRX-300, TerraPower, or others) and navigate licensing, so any Oracle SMR is at least a decade out.

In summary, 2028–2030 is the earliest window for seeing initial SMR units operational for data center use (with Google's Kairos unit and possibly Oklo's microreactor as contenders for 2030 or just before). Most of the big-power SMR projects (100+ MW) — such as those involving X-energy, Kairos and others — will land in the 2030s, assuming schedules hold. In fact, several industry experts caution that even 2030 might be optimistic for first-of-a-kind projects, and "mid-2030s" is more likely for broad commercial availability of SMR-powered data centers [20].

For example, a principal analyst at Omdia noted that current nuclear projects are essentially pilot/demonstration reactors on a long regulatory road, and he pegs the mid-2030s as the earliest for nuclear energy to meaningfully power cloud infrastructure [21]. Similarly, data center research leads at CBRE and others predict a decade or more before nuclear (SMRs or otherwise) is a common solution for hyperscalers [22].

Regulatory and logistical hurdles

Why the long timelines? Building new nuclear in the U.S. remains a slow, complex endeavor, and SMRs (though smaller) are no exception. Several key hurdles must be overcome before these reactor projects can deliver power:

NRC licensing and approval

Nuclear regulatory approval is a multi-year process. For large conventional reactors, obtaining a Combined License from the Nuclear Regulatory Commission can take 5–7+ years. SMR developers are trying new approaches to accelerate this (e.g., Oklo's plan for a combined construction/operation license and the NRC's upcoming Part 53 framework for advanced reactors), but progress is incremental. Only one SMR design (NuScale's 77 MWe module) has full NRC design certification so far (approved January 2023). All other designs mentioned (Kairos, X-energy, Oklo, etc.) are in pre-licensing or early application stages. For instance, X-energy's Xe-100 is proceeding via the DOE demonstration program; the NRC just agreed to target an 18-month review for its construction permit (potentially by late 2026) [23] — a faster-than-usual timeline. Even so, X-energy's first unit (at Dow's Texas site) won't get a final go/no-go until around 2028, as Dow has said it won't commit investment until then [24] [25]. Kairos Power, meanwhile, secured a construction permit for its non-power test reactor (Hermes) in December 2023 — notably, the first such permit for a Gen-IV advanced reactor in the U.S. — but it still must apply for operating licenses and then for its full-scale commercial units. In summary, none of the SMRs destined for data centers has yet completed the licensing gauntlet, and regulatory review itself adds years of uncertainty to each project.

First-of-a-kind engineering

Every SMR project that will power data centers is essentially a first-of-a-kind (FOAK) deployment of a new reactor design. Whether it's Kairos's fluoride salt reactor, X-energy's helium-cooled pebble bed, or Oklo's fast microreactor, these technologies have never delivered power to the grid before. Demonstration plants and prototypes (often heavily funded by DOE) must prove the designs work as expected. Any delays or issues in prototype construction can push back commercial timelines further. For example, TerraPower's Natrium (another advanced reactor, backed by Bill Gates) had aimed for 2028 in Wyoming, but fuel supply issues (HALEU fuel scarcity) have pushed its schedule toward 2030+ [26] — illustrating how even well-resourced projects encounter snags. High-assay low-enriched uranium (HALEU) fuel is required by many advanced reactors (X-energy, Oklo, Kairos all need it), yet currently the U.S. has no large-scale HALEU production. The DOE is scrambling to supply small batches and encouraging domestic enrichment capacity [27] [28], but until that supply chain is established (likely late 2020s), fuel availability could bottleneck SMR deployment.

Permitting and construction challenges

Beyond reactor design approval, there are site-specific permits, environmental assessments, and construction to contend with. Building a nuclear facility, even a smaller SMR, often requires local and state permitting, grid interconnection studies, and sometimes approval from the public utility commission if a utility is involved. The construction itself for nuclear projects is typically a 5+ year endeavor from first concrete to commissioning, even for SMRs. Data center companies are learning that they must plan far ahead: "Begin securing permits, sites, and operational expertise now to prepare for SMRs by the 2030s," advises one data center industry COO [29]. In some cases, companies mitigate risk by choosing existing nuclear sites (e.g. Energy Northwest's project is at the Columbia Generating Station site, and Dominion's would be at North Anna) to leverage existing infrastructure and a community accustomed to nuclear power. This can shave some time off (no need to find a brand new site) and improve public acceptance, but it doesn't eliminate the rigorous build timeline.

Regulatory reform pace

There is recognition at the federal level that licensing needs to speed up. The NRC is developing a new Part 53 rule for advanced reactors to enable more performance-based, risk-informed licensing (the recent ADVANCE Act of 2023 also pushes for this). However, experts note that NRC process changes are not happening as fast as the funding and enthusiasm for SMRs [30]. In practice, each first unit still undergoes close scrutiny. The good news is that the NRC has shown some flexibility – for example, cutting X-energy's construction permit review from 36 to 18 months due to thorough pre-application work [31]. If more such efficiencies are realized and if subsequent units of a certified design can be approved in months instead of years (as Oklo claims [32]), the late 2030s could see a much faster replication of SMRs. But reaching that point depends on the first few projects navigating the current regulatory maze successfully.

Financing and market factors

Building nuclear plants is a capital-intensive endeavor. While our focus is on the timeline, it's worth noting that financing can influence schedules. Some projects may wait on final investment decisions (e.g., Dow won't commit on X-energy until it sees acceptable economics by 2028 [33] [34]). Likewise, if a reactor's cost ends up uncompetitive, deployments could be slowed or scaled back. Tech firms like Amazon and Google are investing directly (e.g., Amazon's ~$500 million into X-energy [35]) helps de-risk these projects. Government incentives also matter: the Inflation Reduction Act's nuclear production tax credits and investment credits are available, and extensions of those (currently, reactors starting construction by 2032–33 can qualify [36]) may need to be extended if the SMRs slip schedule.

Public and regulatory confidence

Ultimately, public perception and regulatory oversight will continue to be a key factor. High-profile support from tech leaders (and the government's embrace of nuclear for data centers [37]) has started to erode the old taboos. Still, any safety incident or licensing mishap could cause delays. The worldwide track record for SMRs is limited — only a few are running (e.g., Russia's floating reactor, a Chinese demo SMR), so every U.S. project is breaking new ground. Bodies like the IAEA and industry consortia are working on safety standards and best practices [38], which should help streamline future approvals. However, until the first few U.S. SMRs are built and operating safely, expect regulators to be cautious and thorough, resulting in longer timelines compared to non-nuclear projects.

Key SMR projects for cloud/data center power (summary table)

The table below compiles the key SMR-linked projects and plans involving commercial tech companies and data center operators. It highlights who is involved (company and partners), what/where the project is, and the status, timeline and regulatory stage for each:

**Table legend:** PPA = Power Purchase Agreement; NRC = Nuclear Regulatory Commission; COL = Combined License; RFP = Request for Proposals.

As shown above, several high-profile SMR projects are in motion, but most are in early phases (feasibility studies, MOUs or the licensing and design stage). The only project with a concrete near-term date is Microsoft's use of an existing reactor by 2028, since that leverages proven technology. For true SMRs, 2030 is a pivotal target year, as both Google and multiple Oklo customers are eyeing that date for inaugural reactors. The bulk of deployments will follow in 2031–2035 once those first units break ground, prove themselves, and regulatory processes (and supply chains) become more streamlined.

SMRs are coming, just not overnight

Tech companies and data center operators have placed significant bets on SMRs to address their long-term energy and sustainability challenges. In 2024, we saw a wave of commitments: Google, Amazon, Microsoft, Oracle, Meta, Equinix and others all made moves to secure nuclear power capacity, with a particular interest in small modular reactors that can eventually be sited near data center hubs. This represents a remarkable shift in the industry mindset — nuclear energy is now on the table as a viable (even necessary) option for powering the cloud.

The timeline for SMRs to deliver power to these companies is measured in years to a decade+, not months. Realistically, the first U.S. SMRs serving commercial technology loads are expected to be online around 2030, and these will be pilot deployments. Widespread use of SMR-powered data centers is unlikely before the mid-2030s [64], given the time needed to license, build and scale these new reactors. In the interim, hyperscalers are employing stopgaps (e.g., tapping output from existing large nuclear plants (as Microsoft and Meta are doing) and investing in renewable projects) to meet their rapid growth in power demand.

The regulatory and engineering hurdles are substantial, but not insurmountable. In the next two to three years, watch for critical milestones that will signal whether the SMR timeline is accelerating or slipping: e.g., NRC approvals of designs like X-energy's and Oklo's; successful startup of first-of-kind reactors (Kairos's test reactor by 2027, Oklo's demo by 2027–28); and the groundbreaking of flagship projects, like the Energy Northwest/X-energy plant in Washington. Government support and private capital are lining up to push this forward, given the dual imperatives of greening the grid and powering the digital economy.

The SMR revolution for data centers is in motion, but patience is required. If all goes as planned, by the early 2030s, we will see the first small reactors generating power for cloud infrastructure — a historic convergence of IT and nuclear tech. By the mid-to-late 2030s, SMRs could be a more routine part of new data center energy plans, helping fuel the AI era with carbon-free, always-on power. As one industry expert put it, everyone hopes these timelines can speed up "and that we will see nuclear sooner" [65], but prudence dictates planning for a longer horizon. For now, 2030 is the magic number on the calendar, with a long checklist to be completed between now and then to turn today's SMR promises into a powered-up reality.