Power & Sites · United States

Nvidia's big bet on physical AI aims for safer robotaxis, humanoid robots

Credited to Ars Technica · arstechnica.com · 2h ago

Photograph · Ars Technica

Full-stack safety solution for physical AI is being used by robotics companies.

Investors bullish on AI data centers may be fueling Nvidia’s multi-trillion-dollar market capitalization, but the AI chipmaker has also bet billions of dollars on physical AI technologies such as robotics and self-driving cars. Part of that gambit has involved Nvidia developing a full-stack safety system that companies can build upon to reduce the risk of their machines harming nearby people.

Nvidia’s Halos system launched in 2025 with hardware and software tools to help developers implement guardrails in self-driving cars and other autonomous vehicles. Then the company expanded its safety architecture to more physical AI technologies by announcing Nvidia Halos for Robotics in June 2026—with the goal of enabling safe deployments of autonomous mobile robots in warehouses, humanoid robots walking around inside a factory, or even surgical robots.

“Now the AI models are getting capable, the robot hardware is getting capable, and a thing we thought is going to be the next bottleneck is safety,” Amit Goel, head of robotics ecosystem and edge computing at Nvidia, told Ars. “So that’s why we launched our Halos for Robotics to unlock the capability of these systems.”

The expanded safety offering for robotics comes as Nvidia CEO Jensen Huang described the physical AI business as already driving nearly $10 billion in annual revenue for the company during an appearance on the All-In Podcast in March 2026. Huang previously began highlighting physical AI as Nvidia’s second-most important growth category in 2025.

Expanding safety from AVs to robots

So what does Nvidia’s full-stack safety system involve? It starts with hardware, such as the Nvidia IGX Thor computing module for robotics and industrial applications that has an independent processor dedicated to safety-related workloads. “In the context of physical AI, you need to have your functional system and your safety system all running on the same silicon,” Goel said.

On the software side, the Halos operating system was designed to enable constant monitoring of “every hardware block” and “every software library” to swiftly spot any failures, Goel explained. The system also isolates safety-critical computing workloads to avoid any potential interference.

The system also includes the Nvidia Holoscan Sensor Bridge that connects sensor data with safety-related computer processing in a way that can easily identify corrupted data. This can be embedded in individual hardware components such as a microcontroller or Field Programmable Gate Array.

Last but not least, the Halos package also includes Nvidia’s simulations for testing robots in virtual environments, and an inspection lab program that allows robotics companies and other partners to get quick feedback on any safety “artifacts” that arise during robotic development.

However, adapting Nvidia Halos from autonomous vehicles to robotics required accommodating many different definitions of safety. The definition of functional safety for autonomous driving generally remains the same across different automotive companies and countries, Goel explained. But a robotic vacuum cleaning a hallway will have very different safety considerations compared to a robotic forklift handling heavy payloads in a warehouse loading docks.

“We had to reimagine and do a lot of foundational work to build a platform that is programmable, that gives all the hooks to the developers to actually define custom safety functions, but at the same time not break the underlying infrastructure that we built,” Goel told Ars. “Because giving flexibility can come at the cost of losing some control over the stack.”

Custom safety functions are also necessary because robots may operate in complex, diverse environments.

“If you are a robot coming down an aisle at six miles per hour, your senses do not see what are the blind spots and what is coming around the corner,” Gold said. “What the robots do is almost come to a halt before the turn, because they really don’t know what’s happening in all these factories, warehouses and hospitals—they’re unstructured environments and things can come from anywhere.”

Real-world robot deployments

The US startup Agility Robotics was first to incorporate Nvidia Halos into its latest Digit 5 humanoid robot that is designed to work safely near people without requiring isolated workstations or physical barriers. The Nvidia system helped Agility bring all the relevant safety sensors and hardware inside the robot, whereas previous Digit robot operations have also relied on external sensors placed around the robot’s work cell to ensure safety.

“Now the robot is literally unchained—the safety goes with the robot wherever it goes,” Goel told Ars. “Now they can operate everywhere in a factory or a warehouse without having to set up a brand new infrastructure every time they want to do a new task.”

The Massachusetts-based robotics company Boston Dynamics has also been an early partner in the Nvidia Halos safety accreditation program, with a focus on building a safety platform that encompasses their Spot robot dogs, wheeled Stretch robots with a large robotic arm, and Atlas humanoid robots that could start deploying in the automotive factories of parent company Hyundai by 2028.

Other companies working to incorporate Nvidia Halos include the German company KION Group that deploys self-driving forklifts and other autonomous mobile robots, along with the South Korean company LG that is developing its own humanoid robot based on Nvidia’s IAI foundation model for humanoid robots called Isaac GROOT.

It remains to be seen if humanoid robots can become autonomous, general-purpose robot workers capable of handling practically any human task. But the latest push by the United States, China and Europe to deploy more robots extends well beyond humanoids—and is putting robots to the test in many new environments.

“Earlier robots were mostly confined in one station doing one thing,” Goel said. “But in the new age of robotics with a generalized body as well as generalized skills, it is essential that the robots can do multiple tasks in multiple environments—and the only way you can do that in production is if you improve safety.”

Power & SitesUnited States

Lawmakers Punt On Stripping Approvals From 15 Projects In Data Center Alley

Credited to Bisnow · bisnow.com14h ago

Useful

A proposal to remove existing approvals from 15 data center projects in the industry’s most important hub stalled Tuesday as Loudoun County lawmakers tabled the measure due to insufficient support.

In a sector already grappling with local opposition that is making projects harder to finance, the prospect of officials canceling existing approvals could be a sign of additional risks to come — particularly in a county that has long served as a canary in the coal mine for the industry's biggest challenges.

Loudoun County, Virginia, is the world’s largest data center hub, with more than 250 such facilities in operation and more than 120 in the development pipeline. Much of the world’s internet traffic passes through the county’s eastern edge, where a dense cluster known as Data Center Alley established the county as the industry's global capital.

This status emerged in part because Loudoun was among the first areas to intentionally court data centers when the sector was still relatively niche, building a political and regulatory environment considered uniquely friendly to the industry’s needs.

That landscape has shifted sharply. In March 2025, the county Board of Supervisors effectively ended by right zoning for new data centers, requiring even projects in industrial areas to undergo legislative review and receive a special exemption from the board. For many projects, the chances of being derailed by local opposition are far higher.

Still, some projects already seeking approval were exempted from review under the old rules. Today, 15 projects totaling roughly 10.5M SF of data center development across 30 buildings are advancing without having to go through the new special exemption process.

But last month, two Loudoun supervisors introduced legislation to revoke the exemption provision for legacy projects. If passed, the bill would effectively strip those 15 projects of their by right zoning approval.

The measure stalled Tuesday after supervisors tabled the discussion without a vote amid little support beyond the bill’s sponsors. The board also tabled another measure introduced by the same lawmakers, framed as a compromise, that would have required an audit of the legacy projects.

Board Vice Chair Mike Turner was among the measure’s opponents, despite voting against the initial legacy resolution in 2025. He said canceling it now amounted to pulling the rug out from under developers in bad faith.

“I don't support changing grandfathering. To me, it's just unethical,” he said at Tuesday's meeting. “We set the rules. Businesses made decisions based on those rules. We live by the rules.”

The 15 projects range from just over 100K SF to more than 4M SF and are spread across Loudoun County. The largest is Digital Realty ’s Digital Dulles Phase 3, a seven-building expansion near Dulles International Airport that would bring the existing campus to 7.8M SF. Other prominent projects include BlackChamber Group ’s Tech Park at Dulles and three proposed data centers adjacent to the Washington Commanders ’ practice facility.

Proponents of ending the legacy protections say they are acting on behalf of residents who want to further rein in data center growth. While 15 projects may seem insignificant compared to the scale of development already built in the county, these projects would amount to a major development wave in almost any other U.S. community.

Supervisors Juli E. Briskman and Laura A. TeKrony, who filed the rollback legislation, said in a joint statement that data centers are creating sprawl and unsightly infrastructure that have brought the county to “a breaking point.”

“We owe it to the community,” TeKrony said Tuesday. “It will assure residents that we not only care about data center development but we also care about their health, safety, quality of life and property values.”

But the prospect of rescinding legacy approvals has raised concerns about legal and financial exposure for the county. In a staff report released ahead of Tuesday’s meeting, officials estimated that ending the protections could trigger developer lawsuits and expose the county to as much as $60M in potential liability, a figure county administrators cautioned may be too low.

Several lawmakers pointed to that risk as their main reason for opposing the exemption rollback measure, warning that a wave of litigation could threaten the county’s finances and bond rating.

Some developers have already signaled they would sue if the rules change. Nick Albu, a land use attorney representing Chuck Kuhn ’s JK Land Holdings, wrote to the board Monday that its estimate “grossly understates the damage claims the county will face.”

“In the event of unlawful action by the board, the owner of JK Data Center Applications will pursue rights and remedies to the full extent of the law,” the letter says.

The staff report also highlights the risk that rolling back legacy exemptions poses to the county’s reputation in the business community, not just data center developers.

Companies make long-term investment decisions based on predictable rules and government commitments, the report says, and “reversing an established process undermines confidence in our word.”

That loss of trust could ultimately weaken Loudoun’s competitiveness and complicate its efforts to diversify its economy beyond data centers, the report warns.

“Continued success and diversification of the business sector is dependant on preserving that trust and confidence to attract new businesses,” it says.

Yet even with no changes enacted Tuesday, Loudoun County's reputation in the data center sector as friendly territory for developers has eroded over the past year as lawmakers have pursued increasingly aggressive limits on the industry’s growth.

Just last month, the Board of Supervisors voted to prepare a resolution that would halt consideration of data center applications for up to 12 months. The move could establish a road map for other localities to work around a legal doctrine in Virginia that effectively doesn’t allow for moratoriums.

While Loudoun County carries outsized importance for the data center sector due to the scale of the industry’s footprint there, the area has also been a bellwether for challenges facing the sector amid its unprecedented building boom.

Loudoun County was the first area to experience power constraints that delayed projects, previewing the grid bottlenecks that now shape the geography of the industry’s growth across the U.S. Meanwhile, neighboring Prince William County is considered ground zero for the wave of backlash derailing billions of dollars worth of data center projects nationwide, while regulatory measures pursued in the Virginia Capitol are now increasingly appearing in other states.

Should the county ultimately prove willing to walk back its legacy exemption provisions in any way, it could set off alarm bells far from Data Center Alley.

Continue reading this story with a free account

View the articleBisnow

Front Method

Power & Sites · United States

Google Deal Backs 890 MW of Constellation Nuclear Uprates at 11 PJM Reactors

Credited to Power Magazine · powermag.com · 21h ago

Photograph · Power Magazine

Hyperscaler Google and Constellation Energy, the largest nuclear operator in the U.S., have signed a 20-year power purchase agreement (PPA) designed to support $4.3 billion of upgrades at 11 nuclear units across the PJM Interconnection, which is slated to add 890 MW of generating capacity from Constellation’s existing reactors by the end of 2032.

The companies announced the agreement Oct. 6 as part of a broader strategic energy collaboration that also includes a separate 15-year supply agreement covering 2,700 MW from Constellation’s existing PJM fleet. Constellation said the 890 MW of incremental capacity will come from equipment and technology upgrades at 11 nuclear units across six sites in Illinois, Pennsylvania, and New Jersey. The first uprate is expected online by 2028. In a blog post, Google said all 890 MW would be online before the end of 2032.

The scale makes the agreement one of the largest corporate-backed nuclear uprate programs announced to date and adds another major tranche of existing-reactor expansion to PJM. It arrives one week after Constellation signed a separate nuclear agreement with Amazon, announcing a 20-year agreement that covers 690 MW from the 1,790-MW Calvert Cliffs nuclear plant in Maryland, including an approximately 190-MW uprate expected online between 2030 and 2032. Constellation said its agreement with Amazon will support more than $3 billion of investment at Calvert Cliffs and help support relicensing the two-unit plant for another 20 years.

Combined, the Google and Amazon agreements will bolster about 1,080 MW of planned Constellation nuclear uprates through long-term contracts with technology companies.

On Tuesday, Constellation also selected Google Cloud and Gemini Enterprise under an expanded five-year technology agreement. The companies separately agreed to a strategic framework to evaluate new clean generation, storage, and demand response opportunities “across the U.S. to support Google’s growth.” While Constellation said the framework “is intended to identify potential projects, commercial structures, and policy pathways that could support additional clean technologies in the years ahead,” it said it will prioritize near-term efforts in PJM.

A Nuclear Fleet at the Center of a Much Larger Power Business

Constellation enters the new Google agreement following its $21.8 billion acquisition of Calpine in January. The Baltimore-based company now controls about 55 GW of generating capacity spanning nuclear, natural gas, geothermal, hydro, wind, solar, oil, and battery storage. Constellation also operates a large competitive retail business serving roughly 2.5 million customer accounts nationwide. While the Calpine acquisition added about 23 GW of predominantly natural gas-fired, geothermal, battery, and solar generation after required asset divestitures, Constellation has described the combination as a way to pair its nuclear fleet with dispatchable gas generation and expand its ability to serve growing power demand across multiple markets.

Nuclear, for now, remains central to the company’s generation portfolio and commercial strategy. Constellation’s nuclear fleet produced 88.8 TWh during the first six months of 2026 and recorded a 92.7% capacity factor across the plants it operates. In its second-quarter filing, Constellation said it had already signed an additional 920 MW of long-term nuclear PPAs with investment-grade customers, generally lasting 15 to 20 years and beginning between 2029 and 2032.

Among them is a 176-MW PPA with Walmart signed in June for Constellation’s Dresden Clean Energy Center in Illinois. That agreement includes 30 MW of expanded generating capacity from planned uprates and two 15-year contract terms beginning in 2029 and 2030. It follows Constellation’s December 2025 license renewal for Dresden, whose two units are now licensed to operate through 2049 and 2051.

The Google deal announced on Tuesday will likely include Braidwood, Byron, LaSalle, and Quad Cities in Illinois; Limerick in Pennsylvania; and Salem in New Jersey, according to reporting from N uclear Newswire. The six sites host 12 reactors. Constellation has not revealed which unit is excluded, and it is unclear how the 890 MW will be allocated among the units. POWER has asked Constellation to confirm the sites and provide details.

The six sites are concentrated in Illinois, Pennsylvania, and New Jersey. Constellation wholly owns and operates Braidwood near Joliet, Byron southwest of Rockford, LaSalle southeast of Ottawa, and Limerick northwest of Philadelphia. It also operates Quad Cities near Cordova, Illinois, where it holds a 75% ownership stake. Salem, in southern New Jersey, is operated by PSEG, which owns 57% of the plant; Constellation holds the remaining 43%.

The fleet spans several generations of U.S. light-water reactor technology. Braidwood, Byron, and Salem each have two Westinghouse four-loop pressurized water reactors (PWRs). LaSalle operates two GE Type 5 boiling water reactors (BWRs), Limerick has two GE Type 4 BWRs, and Quad Cities operates two older GE Type 3 BWRs. Licensed thermal power ranges from 2,957 MWt per unit at Quad Cities to 3,645 MWt at Braidwood and Byron. Current site electrical capacity ranges from about 1,870 MW at Quad Cities to 2,386 MW at Braidwood.

All 12 reactors are already operating under renewed licenses that extended their original 40-year terms to 60 years, but their remaining operating horizons vary considerably. Quad Cities is slated to reach its current license limit in December 2032, and Constellation plans to seek subsequent license renewal (SLR) in early 2027. Salem’s licenses run through 2036 and 2040, and PSEG plans an SLR filing in 2027. LaSalle is licensed through 2042 and 2043, Byron through 2044 and 2046, Braidwood through 2046 and 2047, and Limerick through 2044 and 2049. If approved, subsequent license renewal could extend operation to 80 years.

As Google noted, “A nuclear uprate modernizes turbines, steam generators, and digital control systems to boost the power output of an existing reactor.” Steam-generator work would apply to the PWRs at Braidwood, Byron, and Salem, which use separate steam generators to transfer heat from the reactor coolant system to the secondary side. The BWRs at LaSalle, Limerick, and Quad Cities instead produce steam directly in the reactor vessel and route it to the turbine, so their uprates would involve a different combination of reactor, steam-cycle, turbine-generator, feedwater, control, and other balance-of-plant modifications.

Constellation said the 20-year PPA will support “new equipment and technology” across the 11 units and increase “thermal and electrical efficiency” to unlock additional firm power. The company emphasized that the resulting 890 MW would be “entirely new,” increasing the amount of generation available across PJM rather than reallocating existing output. It characterized uprates as a near-term means of adding firm capacity, saying modernization of turbines, steam generators, and digital control systems at operating plants can produce substantial capacity gains without the “multi-year timelines” and “interconnection bottlenecks” associated with greenfield construction.

The Nuclear Regulatory Commission (NRC) recognizes three categories of power uprate, as POWER reported in March. A measurement uncertainty recapture (MUR) uprate typically recovers less than 2% of licensed power by installing more precise feedwater-flow instruments. A stretch uprate, typically up to 7%, adjusts instrumentation setpoints and operating procedures without major hardware changes. An extended power uprate (EPU) can add up to 20% but is the most capital-intensive, often requiring significant modifications to turbines, condensate pumps, main generators, and transformers. The NRC targets reviews of six, nine, and 12 months, respectively.

According to the NRC’s expected-applications list (last updated in early September), Constellation plans 12 uprate filings between the fourth quarter of 2026 and the end of 2030, including five EPUs and seven MURs. The same list shows PSEG planning stretch uprate applications for Salem Units 1 and 2 in the second quarter of 2027. As Salem’s operator and NRC licensee, PSEG would submit those applications.

Affiliated With DOE’s Uprate Push as Part of UPRISE

The Google agreement, notably, arrives a day after the Department of Energy’s (DOE’s) Office of Energy Dominance Financing (EDF) announced a conditional loan commitment of up to $4.2 billion to finance uprates and modernization at Vistra’s Beaver Valley plant in Pennsylvania and its Davis-Besse and Perry plants in Ohio. As POWER reported, DOE said the projects could add 433 MW of new capacity, all of it under contract to Meta, and preserve nearly 4 GW of existing baseload output. While the commitment is not yet a loan, DOE said Vistra must satisfy “certain technical, legal, environmental, and financial conditions” before the agency signs definitive financing documents and funds it. All three nuclear plants operate in PJM.

It also follows a separate Google-backed uprate program in the Southeast. On Sept. 21, Georgia Power and Google announced an agreement under which Google will support uprates on Georgia Power’s owned portion of the nuclear units at Plants Vogtle and Hatch, seeking to add about 96 MW of new capacity.

That deal, however, is structured differently from Constellation’s PPA. Google plans to subscribe to a new Nuclear Uprate (NU-1) tariff and receive the zero-emission credits associated with the uprated output. Georgia Power said the structure “also helps protect non-participating customers from incremental costs related to the uprate work” and projects about $900 million in customer benefits over the life of the units. The agreement is subject to approval by the Georgia Public Service Commission (PSC). Georgia Power, notably, has also requested approval of a new EPU for Hatch Units 1 and 2 ( Docket 44280, Docket 56002 ). The PSC approved an EPU for Vogtle Units 1 and 2 in Georgia Power’s 2025 Integrated Resource Plan.

The three deals, meanwhile, echo DOE’s March-launched Utility Power Reactor Incremental Scaling Effort (UPRISE), which is targeting 2.5 GW of additional nuclear capacity by 2027 and 5 GW by 2029. While UPRISE does not involve a dedicated appropriation, DOE has pledged technical support to plant owners and the NRC, “matchmaking” workshops pairing plants with large end-users, and expanded use of federal loan authority. EDF has more than $289 billion in available authority and can finance up to 80% of eligible uprate costs. DOE, notably, has sorted the program’s targets into three buckets: balance-of-plant and operational improvements, physical upgrades that require license amendments, and restarts.

In its blog post, Google said the Constellation agreement brings the new nuclear capacity it has enabled from uprates and restarts, including the Georgia uprates, to more than 1.5 GW. The company said that by the end of the decade, its agreements are expected to enable “nearly 25%” of DOE’s 5-GW goal. However, because the Constellation uprates are slated to arrive in stages through 2032, only part of the 890 MW may count toward that 2030 target.

A Response to ‘Bring Your Own Power’ in PJM

On Tuesday, the companies notably framed the agreement as “a direct response to PJM’s ‘Bring Your Own Power’ proposal, providing a practical model for how major power consumers can expand new supply alongside growing operational requirements, without state or federal mandates.” They said the approach aligns with the White House Ratepayer Protection Pledge, which Google and six other technology companies signed March 4, and allows Google to support its growth “without passing on costs to residential customers.”

The PJM mechanism is formally called Bring Your Own New Capacity (BYONC). PJM asked the Federal Energy Regulatory Commission (FERC) to approve it in an Aug. 13 filing (Docket No. ER26-3515-000), as POWER reported. BYONC is the way a new large load (one with a peak of at least 50 MW at a single site) could reduce or eliminate its exposure to Interim Resource Adequacy Service (IRAS). IRAS is PJM’s proposed framework, formerly called “Connect and Manage,” that would make uncovered new large loads available for reduction during grid emergencies starting June 1, 2027. PJM said its BYONC eligibility rules are designed to ensure that capacity is “new and not repackaged” and that its benefits are “real and lasting.” Qualifying resources generally must offer that capacity into PJM’s capacity auctions as price takers for 10 consecutive delivery years. Beginning with the 2029/2030 delivery year, PJM also proposes to exclude uncovered new large-load demand from the demand curve used in those auctions. In a Sept. 23 answer filed in the docket, PJM again asked FERC to make the rules effective no later than Oct. 12 so they could apply to the 2029/2030 capacity auction scheduled to begin in December. As of Oct. 7, PJM’s large-load page listed the docket as pending FERC approval.

PJM’s Aug. 13 filing explicitly allows uprates to count as BYONC, under a condition that bears directly on how the Constellation work is classified. Eligible resources include “the incremental capacity created by an uprate of Existing Generation Capacity Resources, but only to the extent such incremental capacity is created through an increase in such Generation Capacity Resource’s installed capacity and increase in Capacity Interconnection Rights associated with that Generation Capacity Resource.” Under that test, efficiency gains would qualify only if they raise a unit’s rated capacity and its interconnection rights to the grid. PJM was equally direct about existing output. “Allowing existing Capacity Resources to contract with New Large Loads would not solve the problem that the BYONC scheme is designed to address,” it told FERC in its Sept. 23 answer.

Neither company has yet said whether the 890 MW of uprate capacity would be registered as BYONC, or whether the uprates would increase the units’ Capacity Interconnection Rights. The separate 15-year agreement covering 2,700 MW, meanwhile, adds no new capacity. Under that contract, Google will buy output from Constellation plants already operating in PJM, a “financial structure” Constellation said “ensures that Constellation’s operating generation assets continue to deliver energy and capacity to the PJM market.” Under the eligibility rules PJM proposed, capacity that has already cleared a PJM auction generally cannot qualify as BYONC.

The stakes are set by PJM’s pressing supply gap. In a separate July 31 filing, PJM proposed a one-time Reliability Backstop Procurement to close a 6,831.3-MW shortfall from its 2028/2029 capacity auction. That procurement seeks resources under contracts of up to 15 years, while proposing a weighted-average cost cap of $555/MW-day. On Sept. 29, FERC accepted the proposal but suspended it for five months, to an effective date of Feb. 28, 2027, subject to refund and the outcome of further proceedings. PJM said FERC had identified “three discrete issues requiring further consideration” and that the procurement, originally set to open Sept. 30, would not begin on schedule. “Actual timing is to be determined,” PJM added.

The agreement’s demand flexibility provision, which incorporates “operational load-shaping and demand-response capabilities to curtail non-critical power consumption during high-stress grid events,” addresses similar emergency conditions for which IRAS is designed. Both companies, notably, have been pressing for flexibility to count in grid planning. Google and Constellation are among several industry members of the AI Energy Management Alliance (AEMA), a coalition launched Sept. 16 by Emerald AI, Google, and NVIDIA. AEMA is advocating faster or larger grid connections for data centers that accept enforceable curtailment obligations, according to its launch release. Google said in March that it had integrated 1 GW of demand-response capacity into long-term contracts with multiple U.S. utilities.

AI for Uprates and Plant Operations

Tuesday’s agreement notably also expands Constellation’s use of Google’s cloud and AI tools. Under a five-year technology agreement, Constellation said it will build Gemini Enterprise “agentic workflows directly into its core operations.” Google’s blog post suggests the end goal is to accelerate uprate planning, streamline construction, and optimize plant output.

Gemini Enterprise is Google Cloud’s platform for building and deploying AI agents, as POWER reported in September. The platform essentially provides software tools that can query a company’s internal systems in plain language, combine data that is normally kept in separate departments, and recommend or carry out decisions across them. At a Google Cloud media roundtable on Sept. 2, for example, NextEra Energy said Grid Composer, a dispatch and outage-scheduling tool it built on the Gemini Enterprise Agent Platform and rolled out across Florida Power and Light’s generating fleet, had saved customers more than $20 million so far in 2026. NextEra said the tool draws on about half a trillion data points a day from real-time telemetry, load data, and generation profiles, compares manual dispatch decisions against an AI-optimized alternative, and was built in less than 12 weeks.

Constellation and Google plan to home in on three initial focus areas. The first, which Constellation calls “speed to power,” applies AI to site selection, power flow modeling, and permitting and interconnection planning. The company said this would shorten “the engineering and regulatory work on uprates, so new capacity reaches the grid sooner and at lower cost.” The second, generation optimization, covers asset health monitoring, outage and turnaround management, and scenario optimization intended to speed construction. The third will use Google Cloud platforms that Constellation described as “defense-grade” to secure its operational technology (OT) networks, many of which, the company said, “are air-gapped and connect to the bulk electric system.”

“Together, Google and Constellation are setting a new standard for how the technology and energy sectors can collaborate to responsibly expand the nation’s energy supply,” the companies said. “By pairing commercial investments in nuclear uprates with advanced Google technologies, this agreement supports the rapid growth of artificial intelligence and digital infrastructure that directly strengthens the resilience, affordability, and sustainability of the power grid for decades to come.”

— Sonal Patel is senior editor at POWER magazine ( @sonalcpatel, @POWERmagazine ).

Creating Value from Data: Maximizing Plant Performance Through Real-Time Analytics

Modern process industries are experiencing fluctuating market conditions and tight operational margins, leading chemical engineers to rely on real-time data to boost efficiency and reduce costs. Yet, many organizations are at different stages in their digital transformation journey. Some are just starting, while others are looking to optimize existing solutions. This webinar explores practical ways […]

Smart MRO: Using Technology and Assessment to Enhance Power Plant Performance

How One Electric Utility Is Using AI and Extreme Weather Modeling to Make Critical Grid Decisions

archives + podcasts + webinars + e-letters + video + exclusive content + events + much more!

For over 142 years, POWER has been the unwavering pillar of the power and energy industry, offering unparalleled insights on the latest trends, best practices, and news through our robust channels: print, digital media, and in-person events.