Google DeepMind’s new AI uses large language models to crack real-world problems

Stay Ahead, Stay ONMINE

Google DeepMind’s new AI uses large language models to crack real-world problems

Google DeepMind has once again used large language models to discover new solutions to long-standing problems in math and computer science. This time the firm has shown that its approach can not only tackle unsolved theoretical puzzles, but improve a range of important real-world processes as well. Google DeepMind’s new tool, called AlphaEvolve, uses the Gemini 2.0 family of large language models (LLMs) to produce code for a wide range of different tasks. LLMs are known to be hit and miss at coding. The twist here is that AlphaEvolve scores each of Gemini’s suggestions, throwing out the bad and tweaking the good, in an iterative process, until it has produced the best algorithm it can. In many cases, the results are more efficient or more accurate than the best existing (human-written) solutions. “You can see it as a sort of super coding agent,” says Pushmeet Kohli, a vice president at Google DeepMind who leads its AI for Science teams. “It doesn’t just propose a piece of code or an edit, it actually produces a result that maybe nobody was aware of.” In particular, AlphaEvolve came up with a way to improve the software Google uses to allocate jobs to its many millions of servers around the world. Google DeepMind claims the company has been using this new software across all of its data centers for more than a year, freeing up 0.7% of Google’s total computing resources. That might not sound like much, but at Google’s scale it’s huge. Jakob Moosbauer, a mathematician at the University of Warwick in the UK, is impressed. He says the way AlphaEvolve searches for algorithms that produce specific solutions—rather than searching for the solutions themselves—makes it especially powerful. “It makes the approach applicable to such a wide range of problems,” he says. “AI is becoming a tool that will be essential in mathematics and computer science.” AlphaEvolve continues a line of work that Google DeepMind has been pursuing for years. Its vision is that AI can help to advance human knowledge across math and science. In 2022, it developed AlphaTensor, a model that found a faster way to solve matrix multiplications—a fundamental problem in computer science—beating a record that had stood for more than 50 years. In 2023, it revealed AlphaDev, which discovered faster ways to perform a number of basic calculations performed by computers trillions of times a day. AlphaTensor and AlphaDev both turn math problems into a kind of game, then search for a winning series of moves. FunSearch, which arrived in late 2023, swapped out game-playing AI and replaced it with LLMs that can generate code. Because LLMs can carry out a range of tasks, FunSearch can take on a wider variety of problems than its predecessors, which were trained to play just one type of game. The tool was used to crack a famous unsolved problem in pure mathematics. AlphaEvolve is the next generation of FunSearch. Instead of coming up with short snippets of code to solve a specific problem, as FunSearch did, it can produce programs that are hundreds of lines long. This makes it applicable to a much wider variety of problems. In theory, AlphaEvolve could be applied to any problem that can be described in code and that has solutions that can be evaluated by a computer. “Algorithms run the world around us, so the impact of that is huge,” says Matej Balog, a researcher at Google DeepMind who leads the algorithm discovery team. Survival of the fittest Here’s how it works: AlphaEvolve can be prompted like any LLM. Give it a description of the problem and any extra hints you want, such as previous solutions, and AlphaEvolve will get Gemini 2.0 Flash (the smallest, fastest version of Google DeepMind’s flagship LLM) to generate multiple blocks of code to solve the problem. It then takes these candidate solutions, runs them to see how accurate or efficient they are, and scores them according to a range of relevant metrics. Does this code produce the correct result? Does it run faster than previous solutions? And so on. AlphaEvolve then takes the best of the current batch of solutions and asks Gemini to improve them. Sometimes AlphaEvolve will throw a previous solution back into the mix to prevent Gemini from hitting a dead end. When it gets stuck, AlphaEvolve can also call on Gemini 2.0 Pro, the most powerful of Google DeepMind’s LLMs. The idea is to generate many solutions with the faster Flash but add solutions from the slower Pro when needed. These rounds of generation, scoring, and regeneration continue until Gemini fails to come up with anything better than what it already has. Number games The team tested AlphaEvolve on a range of different problems. For example, they looked at matrix multiplication again to see how a general-purpose tool like AlphaEvolve compared to the specialized AlphaTensor. Matrices are grids of numbers. Matrix multiplication is a basic computation that underpins many applications, from AI to computer graphics, yet nobody knows the fastest way to do it. “It’s kind of unbelievable that it’s still an open question,” says Balog. The team gave AlphaEvolve a description of the problem and an example of a standard algorithm for solving it. The tool not only produced new algorithms that could calculate 14 different sizes of matrix faster than any existing approach, it also improved on AlphaTensor’s record-beating result for multipying two four-by-four matrices. AlphaEvolve scored 16,000 candidates suggested by Gemini to find the winning solution, but that’s still more efficient than AlphaTensor, says Balog. AlphaTensor’s solution also only worked when a matrix was filled with 0s and 1s. AlphaEvolve solves the problem with other numbers too. “The result on matrix multiplication is very impressive,” says Moosbauer. “This new algorithm has the potential to speed up computations in practice.” Manuel Kauers, a mathematician at Johannes Kepler University in Linz, Austria, agrees: “The improvement for matrices is likely to have practical relevance.” By coincidence, Kauers and a colleague have just used a different computational technique to find some of the speedups AlphaEvolve came up with. The pair posted a paper online reporting their results last week. “It is great to see that we are moving forward with the understanding of matrix multiplication,” says Kauers. “Every technique that helps is a welcome contribution to this effort.” Real-world problems Matrix multiplication was just one breakthrough. In total, Google DeepMind tested AlphaEvolve on more than 50 different types of well-known math puzzles, including problems in Fourier analysis (the math behind data compression, essential to applications such as video streaming), the minimum overlap problem (an open problem in number theory proposed by mathematician Paul Erdős in 1955), and kissing numbers (a problem introduced by Isaac Newton that has applications in materials science, chemistry, and cryptography). AlphaEvolve matched the best existing solutions in 75% of cases and found better solutions in 20% of cases. Google DeepMind then applied AlphaEvolve to a handful of real-world problems. As well as coming up with a more efficient algorithm for managing computational resources across data centers, the tool found a way to reduce the power consumption of Google’s specialized tensor processing unit chips. AlphaEvolve even found a way to speed up the training of Gemini itself, by producing a more efficient algorithm for managing a certain type of computation used in the training process. Google DeepMind plans to continue exploring potential applications of its tool. One limitation is that AlphaEvolve can’t be used for problems with solutions that need to be scored by a person, such as lab experiments that are subject to interpretation. Moosbauer also points out that while AlphaEvolve may produce impressive new results across a wide range of problems, it gives little theoretical insight into how it arrived at those solutions. That’s a drawback when it comes to advancing human understanding. Even so, tools like AlphaEvolve are set to change the way researchers work. “I don’t think we are finished,” says Kohli. “There is much further that we can go in terms of how powerful this type of approach is.”

Google DeepMind’s new tool, called AlphaEvolve, uses the Gemini 2.0 family of large language models (LLMs) to produce code for a wide range of different tasks. LLMs are known to be hit and miss at coding. The twist here is that AlphaEvolve scores each of Gemini’s suggestions, throwing out the bad and tweaking the good, in an iterative process, until it has produced the best algorithm it can. In many cases, the results are more efficient or more accurate than the best existing (human-written) solutions.

“You can see it as a sort of super coding agent,” says Pushmeet Kohli, a vice president at Google DeepMind who leads its AI for Science teams. “It doesn’t just propose a piece of code or an edit, it actually produces a result that maybe nobody was aware of.”

In particular, AlphaEvolve came up with a way to improve the software Google uses to allocate jobs to its many millions of servers around the world. Google DeepMind claims the company has been using this new software across all of its data centers for more than a year, freeing up 0.7% of Google’s total computing resources. That might not sound like much, but at Google’s scale it’s huge.

Jakob Moosbauer, a mathematician at the University of Warwick in the UK, is impressed. He says the way AlphaEvolve searches for algorithms that produce specific solutions—rather than searching for the solutions themselves—makes it especially powerful. “It makes the approach applicable to such a wide range of problems,” he says. “AI is becoming a tool that will be essential in mathematics and computer science.”

AlphaEvolve continues a line of work that Google DeepMind has been pursuing for years. Its vision is that AI can help to advance human knowledge across math and science. In 2022, it developed AlphaTensor, a model that found a faster way to solve matrix multiplications—a fundamental problem in computer science—beating a record that had stood for more than 50 years. In 2023, it revealed AlphaDev, which discovered faster ways to perform a number of basic calculations performed by computers trillions of times a day. AlphaTensor and AlphaDev both turn math problems into a kind of game, then search for a winning series of moves.

FunSearch, which arrived in late 2023, swapped out game-playing AI and replaced it with LLMs that can generate code. Because LLMs can carry out a range of tasks, FunSearch can take on a wider variety of problems than its predecessors, which were trained to play just one type of game. The tool was used to crack a famous unsolved problem in pure mathematics.

AlphaEvolve is the next generation of FunSearch. Instead of coming up with short snippets of code to solve a specific problem, as FunSearch did, it can produce programs that are hundreds of lines long. This makes it applicable to a much wider variety of problems.

In theory, AlphaEvolve could be applied to any problem that can be described in code and that has solutions that can be evaluated by a computer. “Algorithms run the world around us, so the impact of that is huge,” says Matej Balog, a researcher at Google DeepMind who leads the algorithm discovery team.

Survival of the fittest

Here’s how it works: AlphaEvolve can be prompted like any LLM. Give it a description of the problem and any extra hints you want, such as previous solutions, and AlphaEvolve will get Gemini 2.0 Flash (the smallest, fastest version of Google DeepMind’s flagship LLM) to generate multiple blocks of code to solve the problem.

It then takes these candidate solutions, runs them to see how accurate or efficient they are, and scores them according to a range of relevant metrics. Does this code produce the correct result? Does it run faster than previous solutions? And so on.

AlphaEvolve then takes the best of the current batch of solutions and asks Gemini to improve them. Sometimes AlphaEvolve will throw a previous solution back into the mix to prevent Gemini from hitting a dead end.

When it gets stuck, AlphaEvolve can also call on Gemini 2.0 Pro, the most powerful of Google DeepMind’s LLMs. The idea is to generate many solutions with the faster Flash but add solutions from the slower Pro when needed.

These rounds of generation, scoring, and regeneration continue until Gemini fails to come up with anything better than what it already has.

Number games

The team tested AlphaEvolve on a range of different problems. For example, they looked at matrix multiplication again to see how a general-purpose tool like AlphaEvolve compared to the specialized AlphaTensor. Matrices are grids of numbers. Matrix multiplication is a basic computation that underpins many applications, from AI to computer graphics, yet nobody knows the fastest way to do it. “It’s kind of unbelievable that it’s still an open question,” says Balog.

The team gave AlphaEvolve a description of the problem and an example of a standard algorithm for solving it. The tool not only produced new algorithms that could calculate 14 different sizes of matrix faster than any existing approach, it also improved on AlphaTensor’s record-beating result for multipying two four-by-four matrices.

AlphaEvolve scored 16,000 candidates suggested by Gemini to find the winning solution, but that’s still more efficient than AlphaTensor, says Balog. AlphaTensor’s solution also only worked when a matrix was filled with 0s and 1s. AlphaEvolve solves the problem with other numbers too.

“The result on matrix multiplication is very impressive,” says Moosbauer. “This new algorithm has the potential to speed up computations in practice.”

Manuel Kauers, a mathematician at Johannes Kepler University in Linz, Austria, agrees: “The improvement for matrices is likely to have practical relevance.”

By coincidence, Kauers and a colleague have just used a different computational technique to find some of the speedups AlphaEvolve came up with. The pair posted a paper online reporting their results last week.

“It is great to see that we are moving forward with the understanding of matrix multiplication,” says Kauers. “Every technique that helps is a welcome contribution to this effort.”

Real-world problems

Matrix multiplication was just one breakthrough. In total, Google DeepMind tested AlphaEvolve on more than 50 different types of well-known math puzzles, including problems in Fourier analysis (the math behind data compression, essential to applications such as video streaming), the minimum overlap problem (an open problem in number theory proposed by mathematician Paul Erdős in 1955), and kissing numbers (a problem introduced by Isaac Newton that has applications in materials science, chemistry, and cryptography). AlphaEvolve matched the best existing solutions in 75% of cases and found better solutions in 20% of cases.

Google DeepMind then applied AlphaEvolve to a handful of real-world problems. As well as coming up with a more efficient algorithm for managing computational resources across data centers, the tool found a way to reduce the power consumption of Google’s specialized tensor processing unit chips.

AlphaEvolve even found a way to speed up the training of Gemini itself, by producing a more efficient algorithm for managing a certain type of computation used in the training process.

Google DeepMind plans to continue exploring potential applications of its tool. One limitation is that AlphaEvolve can’t be used for problems with solutions that need to be scored by a person, such as lab experiments that are subject to interpretation.

Moosbauer also points out that while AlphaEvolve may produce impressive new results across a wide range of problems, it gives little theoretical insight into how it arrived at those solutions. That’s a drawback when it comes to advancing human understanding.

Even so, tools like AlphaEvolve are set to change the way researchers work. “I don’t think we are finished,” says Kohli. “There is much further that we can go in terms of how powerful this type of approach is.”

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.

Broadcom resets private cloud strategy with VMware Cloud Foundation 9.0

“Private cloud is not a location, but an operating model for our customers. We’re resetting the bar on what a good private cloud platform needs to look like in the industry. We’re catering to two key personas: the cloud admins who build and operate infrastructure, and the developers who need

Essential commands for Linux server management

Any Linux systems administrator needs to be proficient with a wide range of commands for user management, file handling, system monitoring, networking, security and more. This article covers a range of commands that are essential for managing a Linux server. Keep in mind that some commands will depend on the

Multicloud explained: Why it pays to diversify your cloud strategy

Flexibility. While most cloud vendors pitch themselves as a total cloud solution, the truth is that each major offering has strengths and weaknesses, and companies may not want to commit to one vendor if they have multiple cloud use cases. For instance, an organization might use Microsoft’s Azure cloud for its

Understanding how data fabric enhances data security and governance

“If security is already inconsistent across hybrid or multi-cloud setups, teams will subsequently struggle to get their data fabric architecture as secure as it needs to be,” Inamdar said. How data fabric enhances security While there are some challenges, the reason why so many organizations choose to deploy data fabric

DOE Approves Fourth Loan Disbursement to Restart the Palisades Nuclear Plant

WASHINGTON— U.S. Department of Energy (DOE) Secretary Chris Wright today announced the release of the fourth loan disbursement to Holtec to help fund the restart of the Palisades Nuclear Plant. Today’s action disburses $100,451,904 of the up to $1.52 billion loan guarantee to Holtec for the Palisades Nuclear Plant, which will be America’s first restart of a commercial nuclear reactor that ceased operations, subject to U.S. Nuclear Regulatory Commission (NRC) approvals. “Under President Trump’s leadership, the Department of Energy is taking a leading role in unleashing the American nuclear renaissance,” said Secretary Chris Wright. “The Palisades Nuclear Plant will help to reinvigorate our nuclear industrial base and will reestablish the United States as the world’s nuclear energy leader.” This disbursement marks Holtec’s fourth disbursement of funds from the Loan Programs Office (LPO) since the September 2024 announcement of the loan’s financial close. To date, $251,878,038 of DOE guaranteed loan funds have been disbursed to Holtec to help fund the reopening as it continues to make important milestones toward plant restart, including the NRC’s issuance of the final environmental assessment and finding of no significant impact. Today’s announcement highlights the Energy Department’s leading role in advancing President Trump’s Executive Order 14302, Reinvigorating the Nuclear Industrial Base, through funding the restart of nuclear plants. DOE remains committed to fulfill this mission in order to maximize the speed and scale of nuclear capacity in the United States, ensuring American’s have access to reliable, abundant, and affordable energy. ###

Germany Holds Early Talks on How to Exit SEFE

Germany’s economy ministry is studying options for how to exit nationalized energy company Securing Energy for Europe GmbH, people with knowledge of the matter said. Some officials have been holding early-stage deliberations as they evaluate a range of possible ways to exit SEFE, which could include a sale or breakup of the business, or a potential merger with fellow nationalized energy company Uniper SE, according to the people. The economy ministry department overseeing SEFE is tasked with drawing up a plan by mid-2025. SEFE, a former Gazprom unit, has itself been speaking with Boston Consulting Group as it seeks to examine the economic rationale for various options including a potential merger with Uniper, the people said. It previously studied that possibility during the energy crisis in 2022. A spokesperson for Germany’s economy ministry, which manages the SEFE holding, said it “cannot confirm” whether it is studying exit options, adding that it’s not currently considering a merger with Uniper. A SEFE spokesperson said it works with diverse consultancies including BCG, declining to comment on the nature of the work. The finance ministry, which oversees the Uniper stake, declined to comment. Uniper and BCG also declined to comment. Combining the two nationalized energy companies would only be considered as a potential fallback option in case the government’s sale of Uniper fails to gain traction, some of the people said. While at least three suitors – including Equinor ASA, Czech billionaire Daniel Kretinsky’s EPH and Brookfield Asset Management Ltd. – are considering bids for Uniper, talks to sell the company in one piece are complicated by the utility’s diverse portfolio. Some involved parties are skeptical of the merits of a potential merger between Uniper and SEFE, since any tie-up would have to pass rigorous antitrust hurdles and could delay the government’s exit, according to the

European Nations Lead Push for De-Escalation in Israel-Iran War

Talks aimed at de-escalating the week-long war between Israel and Iran got under way in Geneva on Friday after US President Donald Trump signaled he would give diplomacy a chance before deciding whether to intervene militarily. Iranian Foreign Minister Abbas Araghchi is meeting counterparts from the UK, France and Germany to discuss what he called “nuclear and regional issues” around the ongoing conflict. French President Emmanuel Macron is among those leaders urging Iran to return to negotiations over its nuclear program. Oil prices fell following a report from Reuters that Iran is ready to discuss limitations on uranium enrichment, but won’t consider stopping entirely while it’s under military attacks. Before negotiations with the US were suspended, Tehran had signaled its willingness to accept some restrictions on its enrichment activities, while Israel and US have said the Islamic Republic shouldn’t be allowed to enrich uranium at all. Araghchi on Friday accused Israel of derailing the diplomacy with its strikes, telling the United Nations Human Rights Council that Iranian officials were scheduled to hold a next round of indirect talks with their US counterparts to “craft a promising agreement” that would make progress in resolving the nuclear issue. Israel launched its surprise attack on Iran last week, saying the threat of its sworn enemy acquiring nuclear weapons had to be neutralized. Iran responded with waves of missiles and drones of its own, and there have been heavy casualties on both sides. Trump, who is scheduled to attend a national security meeting in the Oval Office on Friday, has publicly mused for days about the US joining the fray, but appears to have taken a step back after a run of tough rhetoric, including demands for Tehran residents to relocate and threats toward Iran’s Supreme Leader Ayatollah Ali Khamenei. “Based on the fact

CenterPoint sends mobile generators to San Antonio, to support Texas grid

CenterPoint Energy will deploy 15 large mobile generators to the San Antonio area to help reduce the risk of energy shortfalls this summer, the Texas utility said Monday. Installation of the first five units has begun and the remaining 10 will be installed in the coming weeks. The move “will immediately lower monthly bills for our Houston-area customers,” Jason Ryan, CenterPoint’s executive vice president of regulatory services and government affairs, said in a statement. The deployment is expected to reduce bills for Houston-area customers by approximately $2/month by 2027, the utility said. The generators range from 27 MW to 32 MW and were acquired following Winter Storm Uri, which devastated the Texas grid in 2021. CenterPoint subsequently came under fire when the mobile units were not deployed after Hurricane Beryl last summer. The utility opted to forego some profits associated with the mobile generation lease, amid the controversy. “CenterPoint will receive no revenue or profit from the 15 large units based on the agreement” with the Electric Reliability Council of Texas, the utility said Monday. According to the June agreement, ERCOT determined that the proposed retirement of CPS Energy’s V.H. Braunig Units 1, 2, and 3 “poses a significant risk” to its system because those units “help to mitigate risk of cascading outages associated with the post-contingency overload of certain transmission lines importing power into the greater San Antonio area.” ERCOT recommended Unit 3 continue operating under a reliability must-run agreement, but said CenterPoint’s mobile fleet “could help further mitigate the identified reliability risks more cost-effectively than committing V.H. Braunig Units 1 and 2 through RMR agreements.” According to CenterPoint, it’s generators will deliver approximately $200 million of value to the state’s grid. The agreement with ERCOT indicates the generators could run until March 2027. In the aftermath of Hurricane Beryl,

How the potential end of Energy Star could affect apartment operators

In early May, sources from within the U.S. Environmental Protection Agency reported that the federal government intended to end its long-running Energy Star program for energy-efficient appliances, according to the New York Times — spelling the possible end of a widely used voluntary tool in the multifamily industry. The sunset of the Energy Star program would be part of a larger agency initiative to eliminate divisions that oversee efforts related to climate change and energy efficiency, as reported by the New York Times. As of now, the program is still active, and no further reports or official announcements have been made on its status. Energy Star has saved consumers and businesses $500 billion in energy costs since its founding in 1992, according to the program’s 2023 report. On a yearly basis, it creates roughly $40 billion in energy savings at a cost of $32 million to taxpayers, according to the Institute for Market Transformation, a Washington, D.C.-based non-profit focused on high-quality buildings. The program counts thousands of private organizations as its partners, including nearly 40% of Fortune 500 companies, according to the report. Its Sustained Excellence award winners, recognized for years or decades of support, include a number of multifamily companies: Houston-based Hines, Dallas-based CBRE, New York City-based Nuveen Real Estate and New York City-based Tishman Speyer, among others. Since media reports surfaced last month about the program’s demise, more than 1,200 organizations have signed letters appealing to the EPA to continue the Energy Star program, according to the IMT. “The real estate industry is really aligned … on the idea that it is essential to maintain the Energy Star program within the federal government,” Alex Dews, CEO of the IMT, told Multifamily Dive. “It is a public good that cannot be replicated in the same way outside of government.”

From backup to backbone: Why utility-led DERs must drive MISO’s resource adequacy plans

Jigar Shah is managing partner at Multiplier, an advisory firm, and former director of the U.S. Department of Energy Loan Programs Office. The United States is facing an unprecedented surge in electricity demand, projected to grow by more than 150 GW by 2030, rivaling energy expansions seen only during World War II. Our power grid must evolve faster than ever. The recent Federal Energy Regulatory Commission ruling and the submission of a revised fast-track interconnection process by the Midcontinent Independent System Operator underscore the critical need for innovative, scalable solutions that enhance resource adequacy. One path forward lies in utility-led distributed energy resources (DERs), a model exemplified by Xcel Energy’s Distributed Capacity Procurement (DCP) program included in its 2024 integrated resource plan and mirrored by utility battery storage proposals from Exelon’s Maryland utilities in PJM. These programs highlight how utilities can approach deploying DERs like solar and storage up to ten times faster than traditional virtual power plant approaches, addressing load growth with precision, speed and scale. Distributed energy resources have often been underused due to lack of predictability of use and the random siting of assets, issues that utility-led programs directly solve by integrating DERs into system planning. By strategically siting, deploying and dispatching distributed assets, utilities can provide flexible capacity that smooths peak demand, defers or replaces costly combustion turbines and peaker plants, and minimizes expensive transmission and distribution upgrades as demand grows rapidly. This flexible, modular approach also narrows the uncertainty band around new generation buildout, enabling capacity to scale alongside real-time load growth while supporting economic expansion and community development. As DERs move from being mere “backup” resources to becoming the “backbone” of the power system, they provide an energy-dense, resilient and cost-stabilizing solution. The U.S. Department of Energy’s 2025 “Pathways to Commercial Liftoff: Virtual Power

Can Intel cut its way to profit with factory layoffs?

Matt Kimball, principal analyst at Moor Insights & Strategy, said, “While I’m sure tariffs have some impact on Intel’s layoffs, this is actually pretty simple — these layoffs are largely due to the financial challenges Intel is facing in terms of declining revenues.” The move, he said, “aligns with what the company had announced some time back, to bring expenses in line with revenues. While it is painful, I am confident that Intel will be able to meet these demands, as being able to produce quality chips in a timely fashion is critical to their comeback in the market.” Intel, said Kimball, “started its turnaround a few years back when ex-CEO Pat Gelsinger announced its five nodes in four years plan. While this was an impressive vision to articulate, its purpose was to rebuild trust with customers, and to rebuild an execution discipline. I think the company has largely succeeded, but of course the results trail a bit.” Asked if a combination of layoffs and the moving around of jobs will affect the cost of importing chips, Kimball predicted it will likely not have an impact: “Intel (like any responsible company) is extremely focused on cost and supply chain management. They have this down to a science and it is so critical to margins. Also, while I don’t have insights, I would expect Intel is employing AI and/or analytics to help drive supply chain and manufacturing optimization.” The company’s number one job, he said, “is to deliver the highest quality chips to its customers — from the client to the data center. I have every confidence it will not put this mandate at risk as it considers where/how to make the appropriate resourcing decisions. I think everybody who has been through corporate restructuring (I’ve been through too many to count)

Intel appears stuck between ‘a rock and a hard place’

Intel, said Kimball, “started its turnaround a few years back when ex-CEO Pat Gelsinger announced its five nodes in four years plan. While this was an impressive vision to articulate, its purpose was to rebuild trust with customers, and to rebuild an execution discipline. I think the company has largely succeeded, but of course the results trail a bit.” Asked if a combination of layoffs and the moving around of jobs will affect the cost of importing chips, Kimball predicted it will likely not have an impact: “Intel (like any responsible company) is extremely focused on cost and supply chain management. They have this down to a science and it is so critical to margins. Also, while I don’t have insights, I would expect Intel is employing AI and/or analytics to help drive supply chain and manufacturing optimization.” The company’s number one job, he said, “is to deliver the highest quality chips to its customers — from the client to the data center. I have every confidence it will not put this mandate at risk as it considers where/how to make the appropriate resourcing decisions. I think everybody who has been through corporate restructuring (I’ve been through too many to count) realizes that, when planning for these, ensuring the resilience of these mission critical functions is priority one.” Added Bickley, “trimming the workforce, delaying construction of the US fab plants, and flattening the decision structure of the organization are prudent moves meant to buy time in the hopes that their new chip designs and foundry processes attract new business.”

Next-gen AI chips will draw 15,000W each, redefining power, cooling, and data center design

“Dublin imposed a 2023 moratorium on new data centers, Frankfurt has no new capacity expected before 2030, and Singapore has just 7.2 MW available,” said Kasthuri Jagadeesan, Research Director at Everest Group, highlighting the dire situation. Electricity: the new bottleneck in AI RoI As AI modules push infrastructure to its limits, electricity is becoming a critical driver of return on investment. “Electricity has shifted from a line item in operational overhead to the defining factor in AI project feasibility,” Gogia noted. “Electricity costs now constitute between 40–60% of total Opex in modern AI infrastructure, both cloud and on-prem.” Enterprises are now forced to rethink deployment strategies—balancing control, compliance, and location-specific power rates. Cloud hyperscalers may gain further advantage due to better PUE, renewable access, and energy procurement models. “A single 15,000-watt module running continuously can cost up to $20,000 annually in electricity alone, excluding cooling,” said Manish Rawat, analyst at TechInsights. “That cost structure forces enterprises to evaluate location, usage models, and platform efficiency like never before.” The silicon arms race meets the power ceiling AI chip innovation is hitting new milestones, but the cost of that performance is no longer just measured in dollars or FLOPS — it’s in kilowatts. The KAIST TeraLab roadmap demonstrates that power and heat are becoming dominant factors in compute system design. The geography of AI, as several experts warn, is shifting. Power-abundant regions such as the Nordics, the Midwest US, and the Gulf states are becoming magnets for data center investments. Regions with limited grid capacity face a growing risk of becoming “AI deserts.”

Edge reality check: What we’ve learned about scaling secure, smart infrastructure

Enterprises are pushing cloud resources back to the edge after years of centralization. Even as major incumbents such as Google, Microsoft, and AWS pull more enterprise workloads into massive, centralized hyperscalers, use cases at the edge increasingly require nearby infrastructure—not a long hop to a centralized data center—to take advantage of the torrents of real-time data generated by IoT devices, sensor networks, smart vehicles, and a panoply of newly connected hardware. Not long ago, the enterprise edge was a physical one. The central data center was typically located in or very near the organization’s headquarters. When organizations sought to expand their reach, they wanted to establish secure, speedy connections to other office locations, such as branches, providing them with fast and reliable access to centralized computing resources. Vendors initially sold MPLS, WAN optimization, and SD-WAN as “branch office solutions,” after all. Lesson one: Understand your legacy before locking in your future The networking model that connects centralized cloud resources to the edge via some combination of SD-WAN, MPLS, or 4G reflects a legacy HQ-branch design. However, for use cases such as facial recognition, gaming, or video streaming, old problems are new again. Latency, middle-mile congestion, and the high cost of bandwidth all undermine these real-time edge use cases.

Cisco capitalizes on Isovalent buy, unveils new load balancer

The customer deploys the Isovalent Load Balancer control plane via automation and configures the desired number of virtual load-balancer appliances, Graf said. “The control plane automatically deploys virtual load-balancing appliances via the virtualization or Kubernetes platform. The load-balancing layer is self-healing and supports auto-scaling, which means that I can replace unhealthy instances and scale out as needed. The load balancer supports powerful L3-L7 load balancing with enterprise capabilities,” he said. Depending on the infrastructure the load balancer is deployed into, the operator will deploy the load balancer using familiar deployment methods. In a data center, this will be done using a standard virtualization automation installation such as Terraform or Ansible. In the public cloud, the load balancer is deployed as a public cloud service. In Kubernetes and OpenShift, the load balancer is deployed as a Kubernetes Deployment/Operator, Graf said. “In the future, the Isovalent Load Balancer will also be able to run on top of Cisco Nexus smart switches,” Graf said. “This means that the Isovalent Load Balancer can run in any environment, from data center, public cloud, to Kubernetes while providing a consistent load-balancing layer with a frictionless cloud-native developer experience.” Cisco has announced a variety of smart switches over the past couple of months on the vendor’s 4.8T capacity Silicon One chip. But the N9300, where Isovalent would run, includes a built-in programmable data processing unit (DPU) from AMD to offload complex data processing work and free up the switches for AI and large workload processing. For customers, the Isovalent Load Balancer provides consistent load balancing across infrastructure while being aligned with Kubernetes as the future for infrastructure. “A single load-balancing solution that can run in the data center, in public cloud, and modern Kubernetes environments. This removes operational complexity, lowers cost, while modernizing the load-balancing infrastructure in preparation

Oracle’s struggle with capacity meant they made the difficult but responsible decisions

IDC President Crawford Del Prete agreed, and said that Oracle senior management made the right move, despite how difficult the situation is today. “Oracle is being incredibly responsible here. They don’t want to have a lot of idle capacity. That capacity does have a shelf life,” Del Prete said. CEO Katz “is trying to be extremely precise about how much capacity she puts on.” Del Prete said that, for the moment, Oracle’s capacity situation is unique to the company, and has not been a factor with key rivals AWS, Microsoft, and Google. During the investor call, Katz said that her team “made engineering decisions that were much different from the other hyperscalers and that were better suited to the needs of enterprise customers, resulting in lower costs to them and giving them deployment flexibility.” Oracle management certainly anticipated a flurry of orders, but Katz said that she chose to not pay for expanded capacity until she saw finalized “contracted noncancelable bookings.” She pointed to a huge capex line of $9.1 billion and said, “the vast majority of our capex investments are for revenue generating equipment that is going into data centers and not for land or buildings.”

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