Google I/O showed how the path for AI-driven science is shifting

Stay Ahead, Stay ONMINE

Google I/O showed how the path for AI-driven science is shifting

EXECUTIVE SUMMARY During Tuesday’s Google I/O keynote, Demis Hassabis, the CEO of Google DeepMind, proclaimed that we are currently “standing in the foothills of the singularity.” It was a striking statement—the singularity is the theoretical future moment when AI rapidly exceeds human intelligence and dramatically transforms the world. But what struck me as I listened in the audience was the context in which he said those words. He was on stage to close out the session with a segment on scientific AI, the centerpiece of which was a video detailing how the company’s weather prediction software provided an advance alert about Hurricane Melissa’s catastrophic landfall in Jamaica last year—and potentially saved lives. If that software, called WeatherNext, helped anyone escape the storm or better fortify their home, that’s an enormous and meaningful achievement. But it’s hardly evidence of an impending singularity. The juxtaposition of Hassabis’ lofty rhetoric with the real-world results of WeatherNext highlighted the tension between two very different approaches to AI for science. The first focuses on AI tools, like WeatherNext, that are designed and trained to solve specific scientific problems. The second is agentic, LLM-based systems that could one day execute cutting-edge research projects without human involvement. This second vision powers a great deal of AI enthusiasm right now, including recent excitement around recursive self-improvement, or the idea that AI systems could eventually become the primary drivers of AI advancement—a process that would get faster and faster as the AI systems grow smarter. And agentic systems are now making real research contributions, sometimes with limited human guidance. Just this week, Pushmeet Kohli, Google Cloud’s chief scientist, published a piece in a special AI and science issue of the journal Daedalus, writing: “We are moving toward AI that doesn’t just facilitate science but begins to do science.” With autonomous AI scientists on the horizon, it’s harder to justify massive efforts to develop super-specialized tools—even one like AlphaFold, for which DeepMind scientists won a Nobel Prize, or a potentially life-saving system like WeatherNext. It also heralds a far stranger future for science, in which humans and AI systems collaborate as peers—or AI even makes scientific progress on its own. To be clear, Google does not appear to be abandoning its work on specialized AI for science tools. AlphaGenome and AlphaEarth Foundations, which are trained for genetics and Earth science applications respectively, were released last summer, and the newest version of WeatherNext came out in November. What’s more, such tools remain extremely popular among scientists. Last year, for instance, Google reported that protein structure predictions from AlphaFold have been used by over three million researchers worldwide. And Isomorphic Labs, a Google subsidiary that aims to use AlphaFold and related technologies to develop new drugs, just raised a $2 billion Series B funding round. But there are concrete signs of realignment, in both enthusiasm and resources. Last month, the Los Angeles Times reported that Google fellow John Jumper, who won the Nobel for AlphaFold, is now working on AI coding, not on science-specific AI tools. It’s not surprising that Google is assigning its best minds to the coding problem, as the company has recently taken a reputational hit because its coding tools don’t currently stand up to those offered by Anthropic and OpenAI. But it may also signal a prioritization of agentic science on Google’s part, as coding abilities are key to the success of some of those systems. Across the industry, agentic researcher systems are showing real potential. This week, OpenAI announced that one of their models had disproved an important mathematics conjecture—perhaps the most meaningful contribution that generative AI has made to mathematics so far, according to some mathematicians. Importantly, the model used by OpenAI is not specialized for solving mathematical problems, or even for research; according to the company, it’s a general-purpose reasoning model in the vein of GPT-5.5. If general agents can make independent contributions to mathematical research, they might soon be able to do the same in science (though the fact that ideas in science must be verified experimentally makes it a tougher domain for AI). Google is certainly devoting a lot of attention toward an agent-driven scientific future. The big scientific announcement at I/O was the new Gemini for Science package, which unites several of the company’s LLM-based scientific systems under one brand. This includes the hypothesis-generating AI Co-Scientist and algorithm-optimizing AlphaEvolve, which are still not publicly available—but as Google is now allowing any researcher to apply for access to Gemini for Science, they may soon see wider adoption in the scientific community. Scientists who were involved in early testing are enthusiastic about their potential: Gary Peltz, a Stanford geneticist, compared using the AI Co-Scientist to “consulting the oracle of Delphi” in a Nature Medicine article. Gemini for Science isn’t incompatible with specialized tools; to the contrary, agentic systems can be designed to call on such tools when they might be useful. And no agentic system can predict the structure that a protein will fold into without AlphaFold’s help (at least not yet). But the company seems to be shifting its public image—and at least some resources and personnel, such as Jumper—away from specifically developing those kinds of tools. Though it has only been five years since AlphaFold solved the protein-folding problem, both the technology and the discourse have quickly moved beyond that once-revolutionary achievement. Google has been careful to position this new set of scientific agents as an accelerant for human scientists, rather than a replacement for them—the choice of the name AI Co-Scientist as opposed to AI Scientist, for instance, appears quite deliberate. Hassabis uses that same human-centric framing when he talks about changes in the landscape of scientific AI. “For the next decade or so, we should think about AI as this amazing tool to help scientists,” Hassabis said in an interview published in the Daedalus issue. “Beyond that timeframe, it is hard to say with any certainty, but perhaps these systems will become more like collaborators.” But no one can be an effective scientific collaborator without also being a skilled scientist in their own right. And if Hassabis is anywhere near the mark when he talks about the “foothills of the singularity,” then AI scientists could eventually exceed the capabilities of their human counterparts. In a discussion with the journalist Mike Allen at I/O, Hassabis spoke of how he was initially inspired to pursue AI when he observed how progress in physics had stagnated since the 1970s; he wondered whether the human mind had reached its limits in that domain, and if AI could help to overcome that barrier. Superhuman agentic scientists would certainly fit that bill. We might not ever get anywhere near there, but Google seems to be aiming itself toward that summit.

During Tuesday’s Google I/O keynote, Demis Hassabis, the CEO of Google DeepMind, proclaimed that we are currently “standing in the foothills of the singularity.” It was a striking statement—the singularity is the theoretical future moment when AI rapidly exceeds human intelligence and dramatically transforms the world. But what struck me as I listened in the audience was the context in which he said those words.

He was on stage to close out the session with a segment on scientific AI, the centerpiece of which was a video detailing how the company’s weather prediction software provided an advance alert about Hurricane Melissa’s catastrophic landfall in Jamaica last year—and potentially saved lives. If that software, called WeatherNext, helped anyone escape the storm or better fortify their home, that’s an enormous and meaningful achievement. But it’s hardly evidence of an impending singularity.

The juxtaposition of Hassabis’ lofty rhetoric with the real-world results of WeatherNext highlighted the tension between two very different approaches to AI for science. The first focuses on AI tools, like WeatherNext, that are designed and trained to solve specific scientific problems. The second is agentic, LLM-based systems that could one day execute cutting-edge research projects without human involvement.

This second vision powers a great deal of AI enthusiasm right now, including recent excitement around recursive self-improvement, or the idea that AI systems could eventually become the primary drivers of AI advancement—a process that would get faster and faster as the AI systems grow smarter. And agentic systems are now making real research contributions, sometimes with limited human guidance.

Just this week, Pushmeet Kohli, Google Cloud’s chief scientist, published a piece in a special AI and science issue of the journal Daedalus, writing: “We are moving toward AI that doesn’t just facilitate science but begins to do science.” With autonomous AI scientists on the horizon, it’s harder to justify massive efforts to develop super-specialized tools—even one like AlphaFold, for which DeepMind scientists won a Nobel Prize, or a potentially life-saving system like WeatherNext. It also heralds a far stranger future for science, in which humans and AI systems collaborate as peers—or AI even makes scientific progress on its own.

To be clear, Google does not appear to be abandoning its work on specialized AI for science tools. AlphaGenome and AlphaEarth Foundations, which are trained for genetics and Earth science applications respectively, were released last summer, and the newest version of WeatherNext came out in November.

What’s more, such tools remain extremely popular among scientists. Last year, for instance, Google reported that protein structure predictions from AlphaFold have been used by over three million researchers worldwide. And Isomorphic Labs, a Google subsidiary that aims to use AlphaFold and related technologies to develop new drugs, just raised a $2 billion Series B funding round.

But there are concrete signs of realignment, in both enthusiasm and resources. Last month, the Los Angeles Times reported that Google fellow John Jumper, who won the Nobel for AlphaFold, is now working on AI coding, not on science-specific AI tools. It’s not surprising that Google is assigning its best minds to the coding problem, as the company has recently taken a reputational hit because its coding tools don’t currently stand up to those offered by Anthropic and OpenAI. But it may also signal a prioritization of agentic science on Google’s part, as coding abilities are key to the success of some of those systems.

Across the industry, agentic researcher systems are showing real potential. This week, OpenAI announced that one of their models had disproved an important mathematics conjecture—perhaps the most meaningful contribution that generative AI has made to mathematics so far, according to some mathematicians.

Importantly, the model used by OpenAI is not specialized for solving mathematical problems, or even for research; according to the company, it’s a general-purpose reasoning model in the vein of GPT-5.5. If general agents can make independent contributions to mathematical research, they might soon be able to do the same in science (though the fact that ideas in science must be verified experimentally makes it a tougher domain for AI).

Google is certainly devoting a lot of attention toward an agent-driven scientific future. The big scientific announcement at I/O was the new Gemini for Science package, which unites several of the company’s LLM-based scientific systems under one brand.

This includes the hypothesis-generating AI Co-Scientist and algorithm-optimizing AlphaEvolve, which are still not publicly available—but as Google is now allowing any researcher to apply for access to Gemini for Science, they may soon see wider adoption in the scientific community. Scientists who were involved in early testing are enthusiastic about their potential: Gary Peltz, a Stanford geneticist, compared using the AI Co-Scientist to “consulting the oracle of Delphi” in a Nature Medicine article.

Gemini for Science isn’t incompatible with specialized tools; to the contrary, agentic systems can be designed to call on such tools when they might be useful. And no agentic system can predict the structure that a protein will fold into without AlphaFold’s help (at least not yet). But the company seems to be shifting its public image—and at least some resources and personnel, such as Jumper—away from specifically developing those kinds of tools. Though it has only been five years since AlphaFold solved the protein-folding problem, both the technology and the discourse have quickly moved beyond that once-revolutionary achievement.

Google has been careful to position this new set of scientific agents as an accelerant for human scientists, rather than a replacement for them—the choice of the name AI Co-Scientist as opposed to AI Scientist, for instance, appears quite deliberate. Hassabis uses that same human-centric framing when he talks about changes in the landscape of scientific AI. “For the next decade or so, we should think about AI as this amazing tool to help scientists,” Hassabis said in an interview published in the Daedalus issue. “Beyond that timeframe, it is hard to say with any certainty, but perhaps these systems will become more like collaborators.”

But no one can be an effective scientific collaborator without also being a skilled scientist in their own right. And if Hassabis is anywhere near the mark when he talks about the “foothills of the singularity,” then AI scientists could eventually exceed the capabilities of their human counterparts.

In a discussion with the journalist Mike Allen at I/O, Hassabis spoke of how he was initially inspired to pursue AI when he observed how progress in physics had stagnated since the 1970s; he wondered whether the human mind had reached its limits in that domain, and if AI could help to overcome that barrier. Superhuman agentic scientists would certainly fit that bill. We might not ever get anywhere near there, but Google seems to be aiming itself toward that summit.

Stay Ahead

Explore More Insights

Stay ahead with more perspectives on cutting-edge power, infrastructure, energy, bitcoin and AI solutions. Explore these articles to uncover strategies and insights shaping the future of industries.

Midmarket AI gap: Small IT teams are being asked to do enterprise-sized jobs

A policy that can’t be enforced is a liability document, not a security control. It tells you what employees shouldn’t do, not what they’re doing. Visibility follows the same pattern. Eighty-seven percent are confident in their count of AI applications, yet only two-thirds can see the specific data being uploaded

Aker BP, Höegh Evi secure EU funding for Atlas CCS project

Aker BP and Höegh Evi have been awarded €223 million (about US$250 million) from the European Union Innovation Fund to support development of the Atlas carbon capture and storage (CCS) project in Norway. The Atlas project is designed as a large-scale CCS network linking industrial CO2 emitters in Europe with

Nvidia’s Berlin GTC: Faster AI factories, lower power use and more secure agents

Germany’s Agile Robots introduced Agile ONE SN, a humanoid reference platform built on Nvidia’s Isaac GR00T technology and powered by Jetson Thor. It has 128GB of unified memory and is scheduled to become available in early 2027. Telecom Operators Test AI-Powered Networks Nvidia and Nokia announced expanded work with telecom

Extreme wants to use idle network compute capacity for local AI

“We’ve spent decades building networks that reach every corner of the enterprise, and those networks hold a lot of idle compute. Exchange zones will put that compute to work, letting AI run locally on infrastructure customers already own. That delivers sovereignty, local inferencing, and better economics than sending everything to

Ovintiv adds to Montney position with $890-million deal for Logan Energy

The deal also includes two new oil batteries in Simonette—the first of which came online in this year’s third quarter and the second expected online by the end of first-quarter 2027—with combined capacity of 24,000 b/d, a 50% interest in a 90-MMcfd Simonette natural gas processing plant, and a 65% interest in a 57-MMcfd Pouce Coupe gas plant. Ovintiv said the infrastructure is expected to enhance margins and support future oil growth. “The Logan assets have demonstrated leading oil productivity performance and will immediately compete for capital alongside the rest of our portfolio,” said Brendan McCracken, president and chief executive officer (CEO) of Ovintiv. The company estimates the acquisition will generate about $60 million/year in cost synergies, primarily through reduced capital, operating, and overhead costs. Ovintiv said the transaction is expected to be immediately accretive to key operational and financial metrics, including about 5% accretion to cash flow per share in 2027. Ovintiv’s acquisition of Logan assets is expected to close in first-quarter 2027, subject to shareholder, court, and regulatory approvals, after which, the operator expects its core Montney position to increase to about 720,000 net acres. Also included in the transaction is about 60,000 net acres with exposure to the Duvernay shale. Logan Oil spin-out As part of the transaction, Logan will spin out Logan Oil Corp. to existing shareholders ahead of closing. The company will include the Sawn Lake steam-assisted gravity drainage (SAGD) project in Alberta’s Peace River region, to be acquired through a previously announced deal to acquire Andora Energy Corp. from CanAsia Energy Corp.; Clearwater assets in the Nipisi area to be acquired from Woodcote Energy Inc.; and certain undeveloped lands in northwest Alberta, including Charlie Lake lands. The assets will be excluded from the Ovintiv transaction and retained by Logan shareholders through the new company. Logan

EQT plans to buy Range and create gas giant with 9+ bcfed of production

That’ll be the case particularly in Appalachia, where many EQT and Range operations sit side by side and where new power generation facilities are being built to serve demand from data centers and other users. Rice said EQT today is exposed to 3 bcfd of demand from local markets. “That’s before we think about adding the sustainable growth,” Rice added. “We’ve got a significant amount of supply that we’re going to be able to support with gas supply contracts. This long-duration inventory that we have is going to allow us to continue to extend that opportunity set.” EQT CEO Jeremy Knop added that Range’s acreage along and near parts of the Ohio River is “almost like beachfront real estate” as economic development leaders in the region look to capitalize on long-term tailwinds from the data center, energy and advanced manufacturing sectors. Those factors, Rice said, helped kick EQT-Range talks into gear from a timing perspective. He told analysts that the long-term prospect of Range assets growing their capacity to 5 bcfd from today’s roughly 2.5 bcfd and EQT growing its productive capacity to more than 12 bcfd will allow the combined organization to be more aggressive in the market. “Seeing how we can put these companies together and take a significant piece of this opportunity, I think to me was the compelling catalyst for us to act now,” Rice said. Another contributor to making the deal work is Range’s natural gas liquids production, which accounts for about 30% of the company’s total output. Andrew Dittmar, principal analyst at Enverus Intelligence Research, said working more liquids into its profile has been a key strategic priority. “It helps lower the cost of supply relative to gas prices and improves cash flow when gas struggles, as it has through 2026,” Dittmar wrote. “That

Equinor, Aker BP to develop tight NCS reservoirs

Equinor ASA and Aker BP ASA have entered into a strategic collaboration on tight reservoirs on the Norwegian Continental Shelf (NCS). Through technology development, standardization, and greater scale, the companies aim to enable development of more discoveries and increase oil and gas recovery. The collaboration covers prospects, discoveries, and projects across the NCS, with the companies agreeing to contribute relevant opportunities from their respective portfolios and jointly identify where technology development, standardization, and exchange of experience can create the greatest value. The collaboration also includes resource maturation and well stimulation solutions. The ambition is to establish a sufficient and predictable activity level to support a competitive value chain, with the expertise and supplier capacity required to develop more tight reservoir discoveries. The collaboration will be governed through a joint steering structure and multidisciplinary working groups comprising representatives from both companies. Regional opportunities Large volumes of oil and gas remain in tight reservoirs on the NCS, with the Norwegian Offshore Directorate estimating a recoverable resource potential of about 2.8 billion boe in mapped tight reservoirs, comparable in size to the expected recoverable resources of Johan Sverdrup field. Many of the nearly 90 discoveries that have yet to receive a development decision are located in such reservoirs. Tight reservoirs have low permeability, meaning that oil and gas flow very slowly through the rock. Using conventional methods, only a limited share of the resources can be recovered, and many discoveries are too costly to develop on a standalone basis. The Norwegian Offshore Directorate has therefore encouraged companies to work together.

Ghana’s PHDC advancing Phase 2 of multibillion downstream petroleum hub

Ghana’s Petroleum Hub Development Corp. (PHDC) has entered a memorandum of understanding (MOU) with an international consortium for a proposed $28-billion second phase of the country’s multiphased petroleum hub to be built in the Jomoro municipal district of the country’s Western Region. Signed with a consortium led by UAE-based Global Sphere FZE on behalf of investors from South Korea, Mali, and the UAE, the MOU covers development of Phase 2’s planned 300,000-b/sd refinery, two petrochemical plants, 4 million cu m of storage, and a multi-berth jetty under a build-own-operate model that—if approved—would lead to start of construction within 6 months of the agreement’s Oct. 19 signing, PHDC said in a release. Phase 2 will be the second tranche of the three-phased project that collectively aims to add three grassroots refineries totaling 900,000 b/sd, five new petrochemical plants, fresh storage capacity of 10 million cu m, and at least two jetties supporting imports and exports. Dr. Toni Aubynn, PHDC’s chief executive officer, described the MOU as an opportunity to develop Ghana’s industrial capacity and create jobs. Investor representative Boubacar Belco Bah said the consortium would work with PHDC, AMS Technologies, the project development consortium, and technical and financial institutions on project preparation and risk allocation, including plans for participation by Ghanaian contractors, professionals, and service providers. While its confirmation of this latest agreement sets out Phase 2’s proposed installations and construction window, PHDC did not specify financing commitments, detailed partner responsibilities, or a commissioning schedule. The precise identity of AMS Technologies was neither revealed nor could be confirmed, but PHDC previously announced an investment consortium dubbed “AMS/Double Tree Capital Bank” visited PHDC offices in early December 2025 with an expression of interest in exploring investment opportunities for Phase 2 of the project. Advancing the project Previously estimated at a combined cost

Oil eases on diplomacy hopes; European gas rises on winter supply concerns

Oil prices declined Oct. 26 as renewed diplomatic efforts raised hopes of easing US-Iran tensions, while European natural gas prices climbed as colder weather and disrupted LNG supplies heightened concerns ahead of winter. Brent crude futures traded around $115/bbl, down about 1.6%, while US West Texas Intermediate (WTI) declined about 2% to $101/bbl in early trading. Both remained near 6-month highs following last week’s rally. On Oct. 26, senior officials from Saudi Arabia, Pakistan, and Turkey were expected to hold security talks with Iranian leaders in Tehran. This comes amid concerns that the US and Israel might launch fresh strikes against Iran, while attacks by the Houthi movement against Saudi Arabia continue to escalate. Iranian officials have indicated that the discussions will cover regional conflicts—including the situation in Lebanon—as well as efforts to establish a security framework led by Middle Eastern nations. Nevertheless, the persistent Houthi attacks on Saudi targets, combined with Iranian actions against oil tankers in the Strait of Hormuz, continue to fuel deep concerns regarding the security of oil infrastructure and maritime shipping. Meantime, low European gas inventories and colder weather heightened concerns over supplies for the winter heating season. EU storage stood at 74.6% of capacity as of Oct. 25, well below historical averages for late October, while disruptions to LNG shipments from Qatar and the UAE constrained supplies available to replenish inventories. Competition with Asian buyers for available LNG cargoes added to the pressure. Front-month Dutch TTF futures touched €97.50/MWh on Oct. 26, their highest level since December 2022.

When does a BI value stop being the right number?

For energy companies, the assumptions behind a BI value can change materially during the policy period. Margins and throughput can shift, while repair timelines can extend. External events can also change the economics of an operation quickly. Yet the declared BI value and the policy provisions governing recovery may remain unchanged. This raises a practical question: what should trigger a reassessment, and how will the policy respond if the exposure has changed materially? A BI value has a shelf life A BI calculation draws on assumptions about revenue, margins, production, throughput, fixed costs and recovery time. Those inputs can change at different rates and for different reasons. For a downstream operator, a significant movement in crack spreads can change the earnings exposure associated with an interruption. For a midstream operation, a change in throughput or a prolonged constraint on a critical pipeline can alter the financial impact. Across the energy value chain, an extended repair or recovery period can materially increase the duration of a loss. Organizations should establish materiality thresholds for reassessment, supported by a consistent methodology for revisiting the valuation when conditions change. Where a material change increases exposure, organizations may also need to consider whether to redeclare their values. Aon’s approach to business interruption valuation provides a methodology for reassessing the valuation when material changes occur. The percentage on a volatility clause tells only part of the story A stated volatility allowance can overstate the protection available if the exposure has increased materially or operations resume during the recovery period. The Lloyd’s Market Association introduced an updated Onshore Energy BI Volatility Clause in 2024 after carriers identified the previous version as unsuitable for partial losses. The revised clause was intended to address the policy’s indemnity exposure in those circumstances. Organizations should model the provision against realistic loss

NVIDIA’s DSX POD Pushes AI Factory Construction Toward the Assembly Line

Crusoe Brings an Established Modular Model to DSX POD For Crusoe, NVIDIA’s announcement builds on a modular computing strategy already in operation. In its October 21 announcement, Crusoe confirmed that its Spark modular AI data center platform will support DSX POD, extending a manufacturing and deployment approach the company says it has pursued since 2020. Crusoe has expanded its manufacturing capabilities through the acquisition of electrical equipment specialist Easter-Owens Electric and the development of its 352,000-square-foot Crusoe Spark Factory in Brighton, Colorado. The company also introduced Crusoe Spark Liquid, a sixth-generation modular system engineered for ultra-high-density infrastructure, including NVIDIA Vera Rubin NVL72. Next-generation liquid-cooled units are planned for 2027. Crusoe’s vertically integrated model connects infrastructure design and manufacturing with deployment and cloud operations through Crusoe Cloud. The company says it can deliver productive AI capacity in as little as three months, although that timeline is distinct from NVIDIA’s narrower 14-day prepared-site deployment claim. “Modular isn’t a new strategy or pivot for us. It’s been in our DNA since our founding,” said Bryan Schramm, Crusoe’s head of Spark. Crusoe’s involvement demonstrates that the manufacturing model predates DSX POD. NVIDIA is not inventing modular data center infrastructure; it is attempting to establish a common architecture around capabilities that suppliers such as Crusoe have already spent years developing. That distinction matters as the industry evaluates whether standardized AI factory designs can deliver manufacturing efficiencies at a scale beyond individual suppliers’ existing operations. Vertiv and Polar Put the Deployment Model to Work in Norway The European dimension of NVIDIA’s announcement is reflected in a deployment involving Vertiv and Polar Data Centers at Herøya, Norway. According to Vertiv’s October 21 announcement, Polar is applying a factory-built infrastructure approach to the initial 41 MW phase of a multiphase AI data center campus designed to scale to

Hyperscalers allow more scrutiny of controversial data center plans

Hyperscalers are getting decidedly twitchy about citizens’ reactions to their secretive data center construction plans, and are looking for ways to de-escalate some of the conflicts. Earlier this month, Amazon announced its Data Center Commitment which, among other things, would disclose more information about energy and water use and consider a moratorium asking government officials to sign non-disclosure agreements covering their planning applications. Amazon is not alone in this. Earlier this week, Meta said that it was reviewing its use of NDAs after the company was widely criticized for its secrecy in proposing Hyperion, a Louisiana data center in July.

Google will pay people to use less electricity so its data centers can use more

Google has an answer to some of the electricity supply problems that data center construction is causing: It has signed an agreement with energy supplier Voltus to build what it’s calling a 1-gigawatt virtual power plant, expanding on an earlier 100 megawatt deal the two signed earlier this year. This power plant is virtual in the sense that it doesn’t generate any electricity itself, but aggregates so-called flexible grid capacity into one central resource that can be deployed at times of peak demand. This flexible capacity can be in the form of batteries or of load-shedding: asking others to temporarily stop consuming electricity by shutting down industrial processes or adjusting tens of thousands of smart thermostats at once, effectively moving their electricity usage to a time when there is less demand. Voltus operates by offering incentives to companies to reduce their power consumption temporarily, and also offers inducements to electricity resellers that they can pass on to their customers.

Big Tech’s Next Nuclear Bet: More Power From Reactors Already Running

The next wave of nuclear power for AI data centers may come not from new reactors, but from generating stations that have been operating for decades. In a pair of agreements announced just a week apart, Google and Amazon committed to long-term electricity purchases supporting approximately 1,080 MW of additional nuclear generating capacity from Constellation Energy’s existing fleet. The transactions follow earlier commitments by Meta to support reactor upgrades at operating nuclear facilities in Ohio, Pennsylvania and Illinois, establishing a broader commercial model in which hyperscale technology companies help finance increases in nuclear output. Under its October 6 agreement with Google, Constellation plans to invest more than $4.3 billion in equipment and technology upgrades at 11 operating reactor units across Illinois, Pennsylvania and New Jersey. The 20-year power purchase agreement supports 890 MW of new generating capacity, with the first upgrades expected to begin delivering additional electricity in 2028. A separate 15-year agreement covers 2,700 MW of existing nuclear generation, underscoring the distinction between securing today’s electricity supply and financing additional capacity. Amazon’s September 30 agreement similarly supports 190 MW of new generation at Maryland’s Calvert Cliffs nuclear station, alongside purchases of existing output. Together, the two arrangements underpin more than $7 billion in planned infrastructure investment. Unlike proposed advanced reactors, these facilities already have operating licenses, grid connections, established workforces and decades of operating experience. Increasing their output avoids some of the cost, permitting and construction challenges associated with building entirely new nuclear plants, although significant upgrades still require extensive engineering, equipment modifications and regulatory approval. The potential scale is substantial. According to the Nuclear Regulatory Commission’s current application schedule, the agency anticipates 31 applications for reactor power uprates through 2032, representing approximately 2,421 MW of potential additional electrical generating capacity. That is roughly equivalent to the output of

AMD Helion Puts CPU Density at the Center of Agentic AI Infrastructure

AMD expects the growth of agentic applications to substantially increase demand for server CPUs alongside GPU infrastructure. During her October 12 OCP keynote, CEO Lisa Su said the company now projects the server CPU market will exceed $200 billion by 2030, a significant upward revision from its earlier $120 billion estimate. While that remains AMD’s market forecast rather than an independent measure of future demand, it underscores the company’s conviction that general-purpose computing will play a much larger role in AI infrastructure. Two Rack Designs for Different Markets Helion’s two principal configurations reflect different approaches to deploying and managing agentic workloads. For hyperscale operators, the architecture emphasizes high-density CPU infrastructure dedicated to agent execution, with model inference handled by separate GPU-accelerated systems. This separation allows operators to expand CPU and GPU capacity independently, potentially improving resource utilization as the balance between agent activity and inference demand changes. The approach also introduces infrastructure considerations. Agents may communicate repeatedly with models hosted elsewhere in a data center or across distributed environments, making network performance, latency and workload orchestration important factors in overall system efficiency. The ability to scale computing resources independently is an architectural advantage, but realizing that benefit will depend on how effectively those resources are connected and managed. For enterprise deployments, Helion takes a more integrated approach, combining EPYC CPU blades, Instinct MI350P GPU blades and general-purpose computing resources within the same rack. The reference design accommodates up to 16 MI350P accelerators in two independent eight-GPU servers, each occupying five rack units. These GPU nodes operate separately rather than through a common rack-level scale-up fabric, reflecting the architecture’s emphasis on enterprise inference rather than the tightly coupled accelerator environments used for frontier-model training. The enterprise configuration also accommodates dedicated resources for databases, storage and technical computing, supporting the broader range

AI’s Fragmented Regulatory Landscape Becomes a Data Center Design Challenge

Designing for Verifiable Operations For an operator serving tenants across jurisdictions, a claim about where a workload runs must be one that the responsible service provider can substantiate. That means identifying who can administer the workload, how data moves between regions, where backups are maintained, and which records document those arrangements. A tenant promising to run a workload within a particular jurisdiction needs assurance that its data, backups, and access arrangements satisfy the applicable requirements. Workload placement alone cannot establish where a backup resides or from where an administrator can access a system. Those questions must be addressed across compute, storage, network paths, security controls, and third-party access, with responsibilities established during procurement and verified before service begins. Not all of those controls necessarily reside with the data center operator. A colocation provider may be responsible for physical access, facility security, and designated infrastructure services, while a tenant or cloud provider controls the applications, storage configuration, backup locations, and logical access to workloads. The operating agreement must distinguish between what the facility provider can verify directly and what remains within the tenant’s systems. The EU AI Act makes that division of responsibility particularly important. Its requirements for covered high-risk AI systems include logging, documentation, human oversight, robustness, and cybersecurity, subject to the law’s applicable implementation dates. These obligations attach to the relevant providers and deployers of AI systems, rather than automatically to every company hosting them. A data center operator may nevertheless be required by contract to provide physical-access records, security evidence, or information about infrastructure location so that a tenant can substantiate its own compliance. A further distinction concerns data residency. The AI Act is not itself a general requirement to keep all AI workloads or associated data within the European Union. Geographic restrictions may instead arise from data-protection

Microsoft will invest $80B in AI data centers in fiscal 2025

And Microsoft isn’t the only one that is ramping up its investments into AI-enabled data centers. Rival cloud service providers are all investing in either upgrading or opening new data centers to capture a larger chunk of business from developers and users of large language models (LLMs). In a report published in October 2024, Bloomberg Intelligence estimated that demand for generative AI would push Microsoft, AWS, Google, Oracle, Meta, and Apple would between them devote $200 billion to capex in 2025, up from $110 billion in 2023. Microsoft is one of the biggest spenders, followed closely by Google and AWS, Bloomberg Intelligence said. Its estimate of Microsoft’s capital spending on AI, at $62.4 billion for calendar 2025, is lower than Smith’s claim that the company will invest $80 billion in the fiscal year to June 30, 2025. Both figures, though, are way higher than Microsoft’s 2020 capital expenditure of “just” $17.6 billion. The majority of the increased spending is tied to cloud services and the expansion of AI infrastructure needed to provide compute capacity for OpenAI workloads. Separately, last October Amazon CEO Andy Jassy said his company planned total capex spend of $75 billion in 2024 and even more in 2025, with much of it going to AWS, its cloud computing division.

John Deere unveils more autonomous farm machines to address skill labor shortage

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Self-driving tractors might be the path to self-driving cars. John Deere has revealed a new line of autonomous machines and tech across agriculture, construction and commercial landscaping. The Moline, Illinois-based John Deere has been in business for 187 years, yet it’s been a regular as a non-tech company showing off technology at the big tech trade show in Las Vegas and is back at CES 2025 with more autonomous tractors and other vehicles. This is not something we usually cover, but John Deere has a lot of data that is interesting in the big picture of tech. The message from the company is that there aren’t enough skilled farm laborers to do the work that its customers need. It’s been a challenge for most of the last two decades, said Jahmy Hindman, CTO at John Deere, in a briefing. Much of the tech will come this fall and after that. He noted that the average farmer in the U.S. is over 58 and works 12 to 18 hours a day to grow food for us. And he said the American Farm Bureau Federation estimates there are roughly 2.4 million farm jobs that need to be filled annually; and the agricultural work force continues to shrink. (This is my hint to the anti-immigration crowd). John Deere’s autonomous 9RX Tractor. Farmers can oversee it using an app. While each of these industries experiences their own set of challenges, a commonality across all is skilled labor availability. In construction, about 80% percent of contractors struggle to find skilled labor. And in commercial landscaping, 86% of landscaping business owners can’t find labor to fill open positions, he said. “They have to figure out how to do

2025 playbook for enterprise AI success, from agents to evals

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More 2025 is poised to be a pivotal year for enterprise AI. The past year has seen rapid innovation, and this year will see the same. This has made it more critical than ever to revisit your AI strategy to stay competitive and create value for your customers. From scaling AI agents to optimizing costs, here are the five critical areas enterprises should prioritize for their AI strategy this year. 1. Agents: the next generation of automation AI agents are no longer theoretical. In 2025, they’re indispensable tools for enterprises looking to streamline operations and enhance customer interactions. Unlike traditional software, agents powered by large language models (LLMs) can make nuanced decisions, navigate complex multi-step tasks, and integrate seamlessly with tools and APIs. At the start of 2024, agents were not ready for prime time, making frustrating mistakes like hallucinating URLs. They started getting better as frontier large language models themselves improved. “Let me put it this way,” said Sam Witteveen, cofounder of Red Dragon, a company that develops agents for companies, and that recently reviewed the 48 agents it built last year. “Interestingly, the ones that we built at the start of the year, a lot of those worked way better at the end of the year just because the models got better.” Witteveen shared this in the video podcast we filmed to discuss these five big trends in detail. Models are getting better and hallucinating less, and they’re also being trained to do agentic tasks. Another feature that the model providers are researching is a way to use the LLM as a judge, and as models get cheaper (something we’ll cover below), companies can use three or more models to

OpenAI’s red teaming innovations define new essentials for security leaders in the AI era

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More OpenAI has taken a more aggressive approach to red teaming than its AI competitors, demonstrating its security teams’ advanced capabilities in two areas: multi-step reinforcement and external red teaming. OpenAI recently released two papers that set a new competitive standard for improving the quality, reliability and safety of AI models in these two techniques and more. The first paper, “OpenAI’s Approach to External Red Teaming for AI Models and Systems,” reports that specialized teams outside the company have proven effective in uncovering vulnerabilities that might otherwise have made it into a released model because in-house testing techniques may have missed them. In the second paper, “Diverse and Effective Red Teaming with Auto-Generated Rewards and Multi-Step Reinforcement Learning,” OpenAI introduces an automated framework that relies on iterative reinforcement learning to generate a broad spectrum of novel, wide-ranging attacks. Going all-in on red teaming pays practical, competitive dividends It’s encouraging to see competitive intensity in red teaming growing among AI companies. When Anthropic released its AI red team guidelines in June of last year, it joined AI providers including Google, Microsoft, Nvidia, OpenAI, and even the U.S.’s National Institute of Standards and Technology (NIST), which all had released red teaming frameworks. Investing heavily in red teaming yields tangible benefits for security leaders in any organization. OpenAI’s paper on external red teaming provides a detailed analysis of how the company strives to create specialized external teams that include cybersecurity and subject matter experts. The goal is to see if knowledgeable external teams can defeat models’ security perimeters and find gaps in their security, biases and controls that prompt-based testing couldn’t find. What makes OpenAI’s recent papers noteworthy is how well they define using human-in-the-middle

Stay Ahead, Stay ONMINE