Get ONMINE

Stay Ahead, Stay ONMINE

How to build a better AI benchmark

It’s not easy being one of Silicon Valley’s favorite benchmarks.  SWE-Bench (pronounced “swee bench”) launched in November 2024 to evaluate an AI model’s coding skill, using more than 2,000 real-world programming problems pulled from the public GitHub repositories of 12 different Python-based projects.  In the months since then, it’s quickly become one of the most popular tests in AI. A SWE-Bench score has become a mainstay of major model releases from OpenAI, Anthropic, and Google—and outside of foundation models, the fine-tuners at AI firms are in constant competition to see who can rise above the pack. The top of the leaderboard is a pileup between three different fine tunings of Anthropic’s Claude Sonnet model and Amazon’s Q developer agent. Auto Code Rover—one of the Claude modifications—nabbed the number two spot in November, and was acquired just three months later. Despite all the fervor, this isn’t exactly a truthful assessment of which model is “better.” As the benchmark has gained prominence, “you start to see that people really want that top spot,” says John Yang, a researcher on the team that developed SWE-Bench at Princeton University. As a result, entrants have begun to game the system—which is pushing many others to wonder whether there’s a better way to actually measure AI achievement. Developers of these coding agents aren’t necessarily doing anything as straightforward cheating, but they’re crafting approaches that are too neatly tailored to the specifics of the benchmark. The initial SWE-Bench test set was limited to programs written in Python, which meant developers could gain an advantage by training their models exclusively on Python code. Soon, Yang noticed that high-scoring models would fail completely when tested on different programming languages—revealing an approach to the test that he describes as “gilded.” “It looks nice and shiny at first glance, but then you try to run it on a different language and the whole thing just kind of falls apart,” Yang says. “At that point, you’re not designing a software engineering agent. You’re designing to make a SWE-Bench agent, which is much less interesting.” The SWE-Bench issue is a symptom of a more sweeping—and complicated—problem in AI evaluation, and one that’s increasingly sparking heated debate: The benchmarks the industry uses to guide development are drifting further and further away from evaluating actual capabilities, calling their basic value into question. Making the situation worse, several benchmarks, most notably FrontierMath and Chatbot Arena, have recently come under heat for an alleged lack of transparency. Nevertheless, benchmarks still play a central role in model development, even if few experts are willing to take their results at face value. OpenAI cofounder Andrej Karpathy recently described the situation as “an evaluation crisis”: the industry has fewer trusted methods for measuring capabilities and no clear path to better ones.  “Historically, benchmarks were the way we evaluated AI systems,” says Vanessa Parli, director of research at Stanford University’s Institute for Human-Centered AI. “Is that the way we want to evaluate systems going forward? And if it’s not, what is the way?” A growing group of academics and AI researchers are making the case that the answer is to go smaller, trading sweeping ambition for an approach inspired by the social sciences. Specifically, they want to focus more on testing validity, which for quantitative social scientists refers to how well a given questionnaire measures what it’s claiming to measure—and, more fundamentally, whether what it is measuring has a coherent definition. That could cause trouble for benchmarks assessing hazily defined concepts like “reasoning” or “scientific knowledge”—and for developers aiming to reach the much-hyped goal of artificial general intelligence—but it would put the industry on firmer ground as it looks to prove the worth of individual models. “Taking validity seriously means asking folks in academia, industry, or wherever to show that their system does what they say it does,” says Abigail Jacobs, a University of Michigan professor who is a central figure in the new push for validity. “I think it points to a weakness in the AI world if they want to back off from showing that they can support their claim.” The limits of traditional testing If AI companies have been slow to respond to the growing failure of benchmarks, it’s partially because the test-scoring approach has been so effective for so long.  One of the biggest early successes of contemporary AI was the ImageNet challenge, a kind of antecedent to contemporary benchmarks. Released in 2010 as an open challenge to researchers, the database held more than 3 million images for AI systems to categorize into 1,000 different classes. Crucially, the test was completely agnostic to methods, and any successful algorithm quickly gained credibility regardless of how it worked. When an algorithm called AlexNet broke through in 2012, with a then unconventional form of GPU training, it became one of the foundational results of modern AI. Few would have guessed in advance that AlexNet’s convolutional neural nets would be the secret to unlocking image recognition—but after it scored well, no one dared dispute it. (One of AlexNet’s developers, Ilya Sutskever, would go on to cofound OpenAI.) A large part of what made this challenge so effective was that there was little practical difference between ImageNet’s object classification challenge and the actual process of asking a computer to recognize an image. Even if there were disputes about methods, no one doubted that the highest-scoring model would have an advantage when deployed in an actual image recognition system. But in the 12 years since, AI researchers have applied that same method-agnostic approach to increasingly general tasks. SWE-Bench is commonly used as a proxy for broader coding ability, while other exam-style benchmarks often stand in for reasoning ability. That broad scope makes it difficult to be rigorous about what a specific benchmark measures—which, in turn, makes it hard to use the findings responsibly.  Where things break down Anka Reuel, a PhD student who has been focusing on the benchmark problem as part of her research at Stanford, has become convinced the evaluation problem is the result of this push toward generality. “We’ve moved from task-specific models to general-purpose models,” Reuel says. “It’s not about a single task anymore but a whole bunch of tasks, so evaluation becomes harder.” Like the University of Michigan’s Jacobs, Reuel thinks “the main issue with benchmarks is validity, even more than the practical implementation,” noting: “That’s where a lot of things break down.” For a task as complicated as coding, for instance, it’s nearly impossible to incorporate every possible scenario into your problem set. As a result, it’s hard to gauge whether a model is scoring better because it’s more skilled at coding or because it has more effectively manipulated the problem set. And with so much pressure on developers to achieve record scores, shortcuts are hard to resist. For developers, the hope is that success on lots of specific benchmarks will add up to a generally capable model. But the techniques of agentic AI mean a single AI system can encompass a complex array of different models, making it hard to evaluate whether improvement on a specific task will lead to generalization. “There’s just many more knobs you can turn,” says Sayash Kapoor, a computer scientist at Princeton and a prominent critic of sloppy practices in the AI industry. “When it comes to agents, they have sort of given up on the best practices for evaluation.” In a paper from last July, Kapoor called out specific issues in how AI models were approaching the WebArena benchmark, designed by Carnegie Mellon University researchers in 2024 as a test of an AI agent’s ability to traverse the web. The benchmark consists of more than 800 tasks to be performed on a set of cloned websites mimicking Reddit, Wikipedia, and others. Kapoor and his team identified an apparent hack in the winning model, called STeP. STeP included specific instructions about how Reddit structures URLs, allowing STeP models to jump directly to a given user’s profile page (a frequent element of WebArena tasks). This shortcut wasn’t exactly cheating, but Kapoor sees it as “a serious misrepresentation of how well the agent would work had it seen the tasks in WebArena for the first time.” Because the technique was successful, though, a similar policy has since been adopted by OpenAI’s web agent Operator. (“Our evaluation setting is designed to assess how well an agent can solve tasks given some instruction about website structures and task execution,” an OpenAI representative said when reached for comment. “This approach is consistent with how others have used and reported results with WebArena.” STeP did not respond to a request for comment.) Further highlighting the problem with AI benchmarks, late last month Kapoor and a team of researchers wrote a paper that revealed significant problems in Chatbot Arena, the popular crowdsourced evaluation system. According to the paper, the leaderboard was being manipulated; many top foundation models were conducting undisclosed private testing and releasing their scores selectively. Today, even ImageNet itself, the mother of all benchmarks, has started to fall victim to validity problems. A 2023 study from researchers at the University of Washington and Google Research found that when ImageNet-winning algorithms were pitted against six real-world data sets, the architecture improvement “resulted in little to no progress,” suggesting that the external validity of the test had reached its limit. Going smaller For those who believe the main problem is validity, the best fix is reconnecting benchmarks to specific tasks. As Reuel puts it, AI developers “have to resort to these high-level benchmarks that are almost meaningless for downstream consumers, because the benchmark developers can’t anticipate the downstream task anymore.” So what if there was a way to help the downstream consumers identify this gap? In November 2024, Reuel launched a public ranking project called BetterBench, which rates benchmarks on dozens of different criteria, such as whether the code has been publicly documented. But validity is a central theme, with particular criteria challenging designers to spell out what capability their benchmark is testing and how it relates to the tasks that make up the benchmark. “You need to have a structural breakdown of the capabilities,” Reuel says. “What are the actual skills you care about, and how do you operationalize them into something we can measure?” The results are surprising. One of the highest-scoring benchmarks is also the oldest: the Arcade Learning Environment (ALE), established in 2013 as a way to test models’ ability to learn how to play a library of Atari 2600 games. One of the lowest-scoring is the Massive Multitask Language Understanding (MMLU) benchmark, a widely used test for general language skills; by the standards of BetterBench, the connection between the questions and the underlying skill was too poorly defined. BetterBench hasn’t meant much for the reputations of specific benchmarks, at least not yet; MMLU is still widely used, and ALE is still marginal. But the project has succeeded in pushing validity into the broader conversation about how to fix benchmarks. In April, Reuel quietly joined a new research group hosted by Hugging Face, the University of Edinburgh, and EleutherAI, where she’ll develop her ideas on validity and AI model evaluation with other figures in the field. (An official announcement is expected later this month.)  Irene Solaiman, Hugging Face’s head of global policy, says the group will focus on building valid benchmarks that go beyond measuring straightforward capabilities. “There’s just so much hunger for a good benchmark off the shelf that already works,” Solaiman says. “A lot of evaluations are trying to do too much.” Increasingly, the rest of the industry seems to agree. In a paper in March, researchers from Google, Microsoft, Anthropic, and others laid out a new framework for improving evaluations—with validity as the first step.  “AI evaluation science must,” the researchers argue, “move beyond coarse grained claims of ‘general intelligence’ towards more task-specific and real-world relevant measures of progress.”  Measuring the “squishy” things To help make this shift, some researchers are looking to the tools of social science. A February position paper argued that “evaluating GenAI systems is a social science measurement challenge,” specifically unpacking how the validity systems used in social measurements can be applied to AI benchmarking.  The authors, largely employed by Microsoft’s research branch but joined by academics from Stanford and the University of Michigan, point to the standards that social scientists use to measure contested concepts like ideology, democracy, and media bias. Applied to AI benchmarks, those same procedures could offer a way to measure concepts like “reasoning” and “math proficiency” without slipping into hazy generalizations. In the social science literature, it’s particularly important that metrics begin with a rigorous definition of the concept measured by the test. For instance, if the test is to measure how democratic a society is, it first needs to establish a definition for a “democratic society” and then establish questions that are relevant to that definition.  To apply this to a benchmark like SWE-Bench, designers would need to set aside the classic machine learning approach, which is to collect programming problems from GitHub and create a scheme to validate answers as true or false. Instead, they’d first need to define what the benchmark aims to measure (“ability to resolve flagged issues in software,” for instance), break that into subskills (different types of problems or types of program that the AI model can successfully process), and then finally assemble questions that accurately cover the different subskills. It’s a profound change from how AI researchers typically approach benchmarking—but for researchers like Jacobs, a coauthor on the February paper, that’s the whole point. “There’s a mismatch between what’s happening in the tech industry and these tools from social science,” she says. “We have decades and decades of thinking about how we want to measure these squishy things about humans.” Even though the idea has made a real impact in the research world, it’s been slow to influence the way AI companies are actually using benchmarks.  The last two months have seen new model releases from OpenAI, Anthropic, Google, and Meta, and all of them lean heavily on multiple-choice knowledge benchmarks like MMLU—the exact approach that validity researchers are trying to move past. After all, model releases are, for the most part, still about showing increases in general intelligence, and broad benchmarks continue to be used to back up those claims.  For some observers, that’s good enough. Benchmarks, Wharton professor Ethan Mollick says, are “bad measures of things, but also they’re what we’ve got.” He adds: “At the same time, the models are getting better. A lot of sins are forgiven by fast progress.” For now, the industry’s long-standing focus on artificial general intelligence seems to be crowding out a more focused validity-based approach. As long as AI models can keep growing in general intelligence, then specific applications don’t seem as compelling—even if that leaves practitioners relying on tools they no longer fully trust.  “This is the tightrope we’re walking,” says Hugging Face’s Solaiman. “It’s too easy to throw the system out, but evaluations are really helpful in understanding our models, even with these limitations.” Russell Brandom is a freelance writer covering artificial intelligence. He lives in Brooklyn with his wife and two cats. This story was supported by a grant from the Tarbell Center for AI Journalism.

It’s not easy being one of Silicon Valley’s favorite benchmarks. 

SWE-Bench (pronounced “swee bench”) launched in November 2024 to evaluate an AI model’s coding skill, using more than 2,000 real-world programming problems pulled from the public GitHub repositories of 12 different Python-based projects. 

In the months since then, it’s quickly become one of the most popular tests in AI. A SWE-Bench score has become a mainstay of major model releases from OpenAI, Anthropic, and Google—and outside of foundation models, the fine-tuners at AI firms are in constant competition to see who can rise above the pack. The top of the leaderboard is a pileup between three different fine tunings of Anthropic’s Claude Sonnet model and Amazon’s Q developer agent. Auto Code Rover—one of the Claude modifications—nabbed the number two spot in November, and was acquired just three months later.

Despite all the fervor, this isn’t exactly a truthful assessment of which model is “better.” As the benchmark has gained prominence, “you start to see that people really want that top spot,” says John Yang, a researcher on the team that developed SWE-Bench at Princeton University. As a result, entrants have begun to game the system—which is pushing many others to wonder whether there’s a better way to actually measure AI achievement.

Developers of these coding agents aren’t necessarily doing anything as straightforward cheating, but they’re crafting approaches that are too neatly tailored to the specifics of the benchmark. The initial SWE-Bench test set was limited to programs written in Python, which meant developers could gain an advantage by training their models exclusively on Python code. Soon, Yang noticed that high-scoring models would fail completely when tested on different programming languages—revealing an approach to the test that he describes as “gilded.”

“It looks nice and shiny at first glance, but then you try to run it on a different language and the whole thing just kind of falls apart,” Yang says. “At that point, you’re not designing a software engineering agent. You’re designing to make a SWE-Bench agent, which is much less interesting.”

The SWE-Bench issue is a symptom of a more sweeping—and complicated—problem in AI evaluation, and one that’s increasingly sparking heated debate: The benchmarks the industry uses to guide development are drifting further and further away from evaluating actual capabilities, calling their basic value into question. Making the situation worse, several benchmarks, most notably FrontierMath and Chatbot Arena, have recently come under heat for an alleged lack of transparency. Nevertheless, benchmarks still play a central role in model development, even if few experts are willing to take their results at face value. OpenAI cofounder Andrej Karpathy recently described the situation as “an evaluation crisis”: the industry has fewer trusted methods for measuring capabilities and no clear path to better ones. 

“Historically, benchmarks were the way we evaluated AI systems,” says Vanessa Parli, director of research at Stanford University’s Institute for Human-Centered AI. “Is that the way we want to evaluate systems going forward? And if it’s not, what is the way?”

A growing group of academics and AI researchers are making the case that the answer is to go smaller, trading sweeping ambition for an approach inspired by the social sciences. Specifically, they want to focus more on testing validity, which for quantitative social scientists refers to how well a given questionnaire measures what it’s claiming to measure—and, more fundamentally, whether what it is measuring has a coherent definition. That could cause trouble for benchmarks assessing hazily defined concepts like “reasoning” or “scientific knowledge”—and for developers aiming to reach the much–hyped goal of artificial general intelligence—but it would put the industry on firmer ground as it looks to prove the worth of individual models.

“Taking validity seriously means asking folks in academia, industry, or wherever to show that their system does what they say it does,” says Abigail Jacobs, a University of Michigan professor who is a central figure in the new push for validity. “I think it points to a weakness in the AI world if they want to back off from showing that they can support their claim.”

The limits of traditional testing

If AI companies have been slow to respond to the growing failure of benchmarks, it’s partially because the test-scoring approach has been so effective for so long. 

One of the biggest early successes of contemporary AI was the ImageNet challenge, a kind of antecedent to contemporary benchmarks. Released in 2010 as an open challenge to researchers, the database held more than 3 million images for AI systems to categorize into 1,000 different classes.

Crucially, the test was completely agnostic to methods, and any successful algorithm quickly gained credibility regardless of how it worked. When an algorithm called AlexNet broke through in 2012, with a then unconventional form of GPU training, it became one of the foundational results of modern AI. Few would have guessed in advance that AlexNet’s convolutional neural nets would be the secret to unlocking image recognition—but after it scored well, no one dared dispute it. (One of AlexNet’s developers, Ilya Sutskever, would go on to cofound OpenAI.)

A large part of what made this challenge so effective was that there was little practical difference between ImageNet’s object classification challenge and the actual process of asking a computer to recognize an image. Even if there were disputes about methods, no one doubted that the highest-scoring model would have an advantage when deployed in an actual image recognition system.

But in the 12 years since, AI researchers have applied that same method-agnostic approach to increasingly general tasks. SWE-Bench is commonly used as a proxy for broader coding ability, while other exam-style benchmarks often stand in for reasoning ability. That broad scope makes it difficult to be rigorous about what a specific benchmark measures—which, in turn, makes it hard to use the findings responsibly. 

Where things break down

Anka Reuel, a PhD student who has been focusing on the benchmark problem as part of her research at Stanford, has become convinced the evaluation problem is the result of this push toward generality. “We’ve moved from task-specific models to general-purpose models,” Reuel says. “It’s not about a single task anymore but a whole bunch of tasks, so evaluation becomes harder.”

Like the University of Michigan’s Jacobs, Reuel thinks “the main issue with benchmarks is validity, even more than the practical implementation,” noting: “That’s where a lot of things break down.” For a task as complicated as coding, for instance, it’s nearly impossible to incorporate every possible scenario into your problem set. As a result, it’s hard to gauge whether a model is scoring better because it’s more skilled at coding or because it has more effectively manipulated the problem set. And with so much pressure on developers to achieve record scores, shortcuts are hard to resist.

For developers, the hope is that success on lots of specific benchmarks will add up to a generally capable model. But the techniques of agentic AI mean a single AI system can encompass a complex array of different models, making it hard to evaluate whether improvement on a specific task will lead to generalization. “There’s just many more knobs you can turn,” says Sayash Kapoor, a computer scientist at Princeton and a prominent critic of sloppy practices in the AI industry. “When it comes to agents, they have sort of given up on the best practices for evaluation.”

In a paper from last July, Kapoor called out specific issues in how AI models were approaching the WebArena benchmark, designed by Carnegie Mellon University researchers in 2024 as a test of an AI agent’s ability to traverse the web. The benchmark consists of more than 800 tasks to be performed on a set of cloned websites mimicking Reddit, Wikipedia, and others. Kapoor and his team identified an apparent hack in the winning model, called STeP. STeP included specific instructions about how Reddit structures URLs, allowing STeP models to jump directly to a given user’s profile page (a frequent element of WebArena tasks).

This shortcut wasn’t exactly cheating, but Kapoor sees it as “a serious misrepresentation of how well the agent would work had it seen the tasks in WebArena for the first time.” Because the technique was successful, though, a similar policy has since been adopted by OpenAI’s web agent Operator. (“Our evaluation setting is designed to assess how well an agent can solve tasks given some instruction about website structures and task execution,” an OpenAI representative said when reached for comment. “This approach is consistent with how others have used and reported results with WebArena.” STeP did not respond to a request for comment.)

Further highlighting the problem with AI benchmarks, late last month Kapoor and a team of researchers wrote a paper that revealed significant problems in Chatbot Arena, the popular crowdsourced evaluation system. According to the paper, the leaderboard was being manipulated; many top foundation models were conducting undisclosed private testing and releasing their scores selectively.

Today, even ImageNet itself, the mother of all benchmarks, has started to fall victim to validity problems. A 2023 study from researchers at the University of Washington and Google Research found that when ImageNet-winning algorithms were pitted against six real-world data sets, the architecture improvement “resulted in little to no progress,” suggesting that the external validity of the test had reached its limit.

Going smaller

For those who believe the main problem is validity, the best fix is reconnecting benchmarks to specific tasks. As Reuel puts it, AI developers “have to resort to these high-level benchmarks that are almost meaningless for downstream consumers, because the benchmark developers can’t anticipate the downstream task anymore.” So what if there was a way to help the downstream consumers identify this gap?

In November 2024, Reuel launched a public ranking project called BetterBench, which rates benchmarks on dozens of different criteria, such as whether the code has been publicly documented. But validity is a central theme, with particular criteria challenging designers to spell out what capability their benchmark is testing and how it relates to the tasks that make up the benchmark.

“You need to have a structural breakdown of the capabilities,” Reuel says. “What are the actual skills you care about, and how do you operationalize them into something we can measure?”

The results are surprising. One of the highest-scoring benchmarks is also the oldest: the Arcade Learning Environment (ALE), established in 2013 as a way to test models’ ability to learn how to play a library of Atari 2600 games. One of the lowest-scoring is the Massive Multitask Language Understanding (MMLU) benchmark, a widely used test for general language skills; by the standards of BetterBench, the connection between the questions and the underlying skill was too poorly defined.

BetterBench hasn’t meant much for the reputations of specific benchmarks, at least not yet; MMLU is still widely used, and ALE is still marginal. But the project has succeeded in pushing validity into the broader conversation about how to fix benchmarks. In April, Reuel quietly joined a new research group hosted by Hugging Face, the University of Edinburgh, and EleutherAI, where she’ll develop her ideas on validity and AI model evaluation with other figures in the field. (An official announcement is expected later this month.) 

Irene Solaiman, Hugging Face’s head of global policy, says the group will focus on building valid benchmarks that go beyond measuring straightforward capabilities. “There’s just so much hunger for a good benchmark off the shelf that already works,” Solaiman says. “A lot of evaluations are trying to do too much.”

Increasingly, the rest of the industry seems to agree. In a paper in March, researchers from Google, Microsoft, Anthropic, and others laid out a new framework for improving evaluations—with validity as the first step. 

“AI evaluation science must,” the researchers argue, “move beyond coarse grained claims of ‘general intelligence’ towards more task-specific and real-world relevant measures of progress.” 

Measuring the “squishy” things

To help make this shift, some researchers are looking to the tools of social science. A February position paper argued that “evaluating GenAI systems is a social science measurement challenge,” specifically unpacking how the validity systems used in social measurements can be applied to AI benchmarking. 

The authors, largely employed by Microsoft’s research branch but joined by academics from Stanford and the University of Michigan, point to the standards that social scientists use to measure contested concepts like ideology, democracy, and media bias. Applied to AI benchmarks, those same procedures could offer a way to measure concepts like “reasoning” and “math proficiency” without slipping into hazy generalizations.

In the social science literature, it’s particularly important that metrics begin with a rigorous definition of the concept measured by the test. For instance, if the test is to measure how democratic a society is, it first needs to establish a definition for a “democratic society” and then establish questions that are relevant to that definition. 

To apply this to a benchmark like SWE-Bench, designers would need to set aside the classic machine learning approach, which is to collect programming problems from GitHub and create a scheme to validate answers as true or false. Instead, they’d first need to define what the benchmark aims to measure (“ability to resolve flagged issues in software,” for instance), break that into subskills (different types of problems or types of program that the AI model can successfully process), and then finally assemble questions that accurately cover the different subskills.

It’s a profound change from how AI researchers typically approach benchmarking—but for researchers like Jacobs, a coauthor on the February paper, that’s the whole point. “There’s a mismatch between what’s happening in the tech industry and these tools from social science,” she says. “We have decades and decades of thinking about how we want to measure these squishy things about humans.”

Even though the idea has made a real impact in the research world, it’s been slow to influence the way AI companies are actually using benchmarks. 

The last two months have seen new model releases from OpenAI, Anthropic, Google, and Meta, and all of them lean heavily on multiple-choice knowledge benchmarks like MMLU—the exact approach that validity researchers are trying to move past. After all, model releases are, for the most part, still about showing increases in general intelligence, and broad benchmarks continue to be used to back up those claims. 

For some observers, that’s good enough. Benchmarks, Wharton professor Ethan Mollick says, are “bad measures of things, but also they’re what we’ve got.” He adds: “At the same time, the models are getting better. A lot of sins are forgiven by fast progress.”

For now, the industry’s long-standing focus on artificial general intelligence seems to be crowding out a more focused validity-based approach. As long as AI models can keep growing in general intelligence, then specific applications don’t seem as compelling—even if that leaves practitioners relying on tools they no longer fully trust. 

“This is the tightrope we’re walking,” says Hugging Face’s Solaiman. “It’s too easy to throw the system out, but evaluations are really helpful in understanding our models, even with these limitations.”

Russell Brandom is a freelance writer covering artificial intelligence. He lives in Brooklyn with his wife and two cats.

This story was supported by a grant from the Tarbell Center for AI Journalism.

Shape
Shape
Stay Ahead

Explore More Insights

More Articles

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.

Shape

Ovintiv raises 2026 guidance on productivity gains

Speaking to analysts and investors on July 24 after Ovintiv reported its second-quarter results, McCracken and his team said the efficiency gains stem from a cocktail of innovations around well designs, development patterns, and the usage of proppants and surfactants, among other things. “It starts with the culture, that relentless

Read More »

Atos launches sovereign cloud service to power comeback

Atos has launched a new sovereign cloud platform aimed squarely at European public sector bodies, healthcare providers and defense organizations. It’s the latest effort by European companies in their fight back against US dominance. Atos Sovereign Cloud offers a range of controls for data management, providing customers with resilience and

Read More »

Magnolia expands Giddings position with $4-billion WildFire Energy acquisition

In the filing, Magnolia said WildFire’s second-quarter 2025 production is expected to average 53,000 boe/d, about 70% oil, primarily from the Eagle Ford, Austin Chalk, and Woodbine formations. Magnolia said the acquisition would strengthen its position in the Eagle Ford/Austin Chalk trend by expanding its inventory of high-return drilling locations, adding development flexibility and longer laterals, and leveraging its technical expertise to improve well performance and lower costs. “WildFire has a large, low-decline oily PDP base with historic development centered on the Eagle Ford. While there are significant future Eagle Ford development opportunities, our technical teams see extensive future potential in the Austin Chalk with further upside in the Woodbine as well as other appraisal opportunities that should expand on our success in Giddings since 2018,” said Chris Stavros, Magnolia’s chairman, president, and chief executive officer. The deal is expected to result in a pro forma position in Giddings of more than 1.25 million net acres, add more than 500 miles of gas-gathering pipelines, and offer various cost savings, the company said. “Magnolia is guiding to $100 million in run rate synergies by the end of 2027, with savings coming from the chance to deploy long laterals, shared facilities and infrastructure and additional sand sourcing for operations from WildFire’s in-basin mine. As always, successful execution will be key for the longer-term success of the deal,” Enverus’ Dittmar said. Total consideration consists of $2.65 billion in cash, 32.2 million shares of Magnolia Class A common stock, and the assumption of $600 million of outstanding debt.

Read More »

Vår Energi inks deal to acquire BlueNord

Vår Energi ASA has agreed to buy BlueNord ASA as part of a proposed merger that, if completed, will expand Vår Energi’s presence beyond the Norwegian Continental Shelf (NCS), positioning the operator as Europe’s largest independent oil and gas producer. Acqusition of BlueNord would add producing assets on the Danish Continental Shelf (DCS) to Vår Energi’s current holdings, with the combined post-merger portfolio anticipated to lift long-term production to about 450,000 boe/d, with about 2.4 billion boe of reserves and resources and an estimated reserve and resource life of about 15 years. BlueNord’s portfolio includes interests in the Tyra, Halfdan, Dan, and Gorm hub areas, which are part of the Danish Underground Consortium operated by TotalEnergies SE. The assets are expected to contribute about 45,000 boe/d of net production beginning in 2026 and include about 195 million boe of net 2P reserves and 2C contingent resources, extending production beyond 2040. “The transaction marks a significant milestone in Vår Energi’s growth journey, creating the largest independent producer of oil and gas in Europe with a long-term production target of [about 450,000 b/d] and reinforcing our role as a reliable and secure supplier of energy to Europe,” said Nick Walker, Vår Energi’s chief executive officer. Vår Energi said the DCS assets complement its existing North Sea operations because of their geological, operational, and fiscal similarities to the NCS. The combination also expands the company’s exposure to European natural gas markets through access to the Nybro and Den Helder gas delivery points. The combined portfolio would maintain a production mix of about 65% oil and 35% natural gas, with operating costs projected to remain at $10-11/boe. The proposed merger remains subject to approval by BlueNord shareholders, regulatory and governmental approvals, license and partner consent, and other customary conditions. If approved, the companies said

Read More »

Bahrain’s GPIC enlists Fluor for new unit at Sitra complex

Gulf Petrochemical Industries Co. (GPIC) has awarded Fluor Corp. a contract to execute front-end engineering and design (FEED) for a proposed aromatics plant to be built at GPIC’s petrochemicals complex located across 60 hectares of reclaimed land in Sitra, Bahrain. As part of the contract, Fluor will deliver a FEED study based on commercially proven process technologies for the plant’s targeted production of 1.2 million tonnes/year (tpy) of paraxylene and 500,000 tpy of benzene, the service provider said on July 21. Critical building blocks for plastics, polyester fibers, and packaging materials, paraxylene and benzene production from the plant would help meet global demand for high‑performance consumer and industrial products, as well as expand capabilities of GPIC’s current operations at Sitra, Fluor said. GPIC’s existing complex currently uses a feedstock of natural gas domestically produced in Bahrain to produce about 1.2 million tonnes/day of ammonia, 1.2 million tonnes/day of methanol, and 1.7 million tonnes/day of urea. Neither Fluor nor GPIC revealed details regarding a timeline for completion of the proposed aromatics plant. GPIC is a joint venture of Bahrain Petroleum Co. (33.3%), SABIC Agri-Nutrients Investment Co. (33.3%), and Kuwait’s Petrochemical Industries Co. (PIC; 33.3%).

Read More »

Oil prices surge as Hormuz, Bab el-Mandeb risks escalate amid renewed US–Iran tensions

Oil prices jumped on Wednesday, July 22, with escalating geopolitical tensions and mounting risks to key maritime chokepoints driving the rally. International Brent crude rose nearly 5% to above $95/bbl, its highest level in almost 6 weeks, while US crude climbed more than 4% to above $88/bbl. The gains extend a strong upward trend, with prices up about 30% since the start of the month and more than 55% year to date, reversing declines seen after a mid-June memorandum of understanding (MOU) between the US and Iran. Stay updated on oil price volatility, shipping disruptions, LNG market analysis, and production output through OGJ’s Iran war content hub. The earlier agreement, aimed at de-escalating conflict and reopening the Strait of Hormuz, was declared “over” on July 8 by President Donald Trump. Since then, hostilities have intensified, with US forces carrying out an 11th consecutive night of strikes on Iran. Comments from US Secretary of State Marco Rubio further dampened expectations for near-term diplomacy, noting that while Washington remains open to talks, Iran does not appear to be engaging seriously. At the same time, security risks to global shipping have increased. The UK Maritime Trade Operations (UKMTO)  has reported multiple recent attacks on vessels in the region, including incidents that forced crews to abandon ships. As a result, traffic through the Strait of Hormuz has fallen sharply, with just 13 vessels transiting Monday and 9 on Tuesday, according to MarineTraffic data. Concerns are also growing at the Bab el-Mandeb Strait, another critical oil transit route linking the Red Sea to the Gulf of Aden. Iranian-backed Houthi forces in Yemen have threatened a maritime blockade targeting Saudi Arabia, raising fears of broader supply disruptions. While vessel traffic through Bab el-Mandeb remains relatively steady—73 ships transited Tuesday—it has edged lower and signs of hesitation among

Read More »

Global LNG trade hits record in 2025 as 2026 tests market resilience

Global LNG trade reached a record 437 million tonnes in 2025, up 6.3% year on year (y-o-y) and marking the fastest growth since 2022, according to the International Gas Union’s (IGU) World LNG Report 2026. The increase of roughly 25 million tonnes was driven primarily by rising US supply, alongside higher exports from Qatar, Malaysia, Angola, and Nigeria. Canada and the Mauritania–Senegal project also shipped their first LNG cargoes, expanding the pool of exporting countries. Investment kept pace with market growth. Developers sanctioned 68.4 million tonnes/year (tpy) of new liquefaction capacity in 2025—the highest annual total since 2019—bringing approvals over the 2021–25 period to about 206 million tpy, roughly double the volume sanctioned in the previous 5-year cycle. Much of the new capacity was concentrated in US Gulf Coast projects. The outlook for 2026, however, is more uncertain. The Middle East conflict has knocked Qatar and the UAE—together about 16% of global liquefaction capacity—off the market for periods this year, and missile strikes on Qatar’s Ras Laffan complex are expected to keep roughly 12.8 million tpy of capacity offline for 3-5 years. Shell PLC’s separately published LNG Outlook 2026 is blunter about the near-term picture: Depending on how quickly the Strait of Hormuz reopens, 2026 could see global LNG trade contract year-on-year—something that’s never happened before in the past decade of rapid growth Shell has tracked. The Asia Pacific has absorbed most of the supply shock so far, responding through storage draws, fuel switching, demand curtailment and increased spot buying, while a wave of US cargoes has been rerouted from Europe toward Asia to fill the gap. Despite near-term volatility, both reports highlight a strong long-term trajectory. IGU expects global LNG supply capacity, including existing and under-construction projects, to exceed 700 million tonnes by 2030, a roughly 40% increase from

Read More »

INA discovers gas in northern Adriatic

Croatia’s Industrija Nafte DD (INA) has discovered more gas as part of a five-well drilling campaign in the northern Adriatic Sea offshore Croatia.  The first well in the campaign, Ana-4 DIR in the existing North Adriatic field, has been completed after reaching a total depth of 1,282 m. Drilling operations were carried out by the Labin drilling rig, operated by the crew of INA’s service company CROSCO. Initial testing across three reservoirs delivered a total gas flow rate of about 160,000 cu m/d. Preparations are under way to tie the well into the surface production system to carry out an extended well test aimed at reservoir clean-up, detailed characterization of production potential, and collection of key data to support the preparation of the reserves report. The successful completion of the first well confirmed the remaining gas potential of offshore fields that INA will continue to develop in the coming years, the company said. The Labin rig is now being moved to the location of IKA JZ-6 DIR, the next well in the campaign. Construction investment in the five wells is expected to total about EUR 65 million.

Read More »

When Buildability Breaks: What Prince William and New York Signal for Data Center Development

For several years, the Prince William Digital Gateway represented data center ambition at its largest scale: a proposed 2,100-acre technology corridor near Gainesville, Virginia, capable of accommodating tens of millions of square feet of digital infrastructure. Its location also made it uniquely contentious. The corridor bordered Manassas National Battlefield Park and other historic, environmental and residential resources, drawing the data center development debate beyond its usual industry and land-use constituencies. Opposition increasingly centered not only on the project’s scale, but on whether development of that magnitude belonged alongside one of the country’s most significant Civil War landscapes. In July 2026, that vision effectively ended. QTS Data Centers terminated its participation in the Digital Gateway and withdrew its remaining petitions before the Supreme Court of Virginia. The decision followed Compass Datacenters’ withdrawal in April, leaving neither of the project’s original developers pursuing the corridor. QTS said it reached the decision after “careful consideration,” while emphasizing that Virginia remains an important market for the company. From Proposed Capacity to Executable Capacity The collapse of the Digital Gateway is more than the cancellation of one unusually large development. It comes as the data center industry confronts a widening gap between announced capacity and executable capacity. Power remains the most visible constraint. But permitting discipline, environmental review, community acceptance and the durability of political support are increasingly determining whether a project can progress from land control and conceptual capacity to construction and operation. A separate development in New York underscored that shift less than two weeks after QTS withdrew. On July 14, Gov. Kathy Hochul issued Executive Order 62, establishing what the state describes as the nation’s first statewide moratorium on new hyperscale data centers. The order temporarily holds in abeyance certain incomplete state environmental permit applications for data centers capable of drawing at

Read More »

Q&A: Google’s AI and computing chief talks about its shapeshifting data centers

Mark Lohmeyer: We’ve seen the rise of agents and agentic use cases. Years ago, it was the chat phase: Ask a question, get an answer. Now we’re in the agentic era, where you express your intent, agents spin off multiple sub-agents, working in parallel, preserving state. This is a radical shift in what infrastructure needs to do; make them fast, cost effective, secure, reliable. We’re delivering infrastructure optimized for the age of agents. NW: What’s the goal of the infrastructure buildout, and what should customers expect regarding costs? ML: Ultimately, it’s about enabling customers with leading-edge capabilities and models at scale cost-effectively. With agents, inference transactions increase by 50x, 100x versus non-agentic workloads. We’re driving the cost per transaction down exponentially. In our latest platforms, we reduce the cost by almost 2x for the same work. Customers serve twice the number of users at the same cost, directly driving profitability.

Read More »

Google transforms its data center architecture for agent era

Google adjusted the Google Kubernetes Engine into an agent-native environment, where agents could be quickly spun up in sandboxes and containers. “From an infrastructure perspective, you need to spin up a bunch of TPUs or GPUs very rapidly. Then you need to be able to run them and spin them back down,” Lohmeyer said. Google also made drastic improvements to its silicon to support its middleware changes. It recently introduced new AI chips, with the TPU-8t for training, and TPU-8i for inference. The 8t chip has three times more computing power than the previous-generation Ironwood chip. The 8i chip has 384 megabytes of SRAM and 288GB of HBM3e memory, which is 50% more than the previous-generation chip. The platform is optimized for KV cache (key-value cache), which stores important contextual information needed by agents to make decisions, which reduces the round trips to other memory and storage systems. “Being able to store more of the KV cache directly on the chip allows you to respond much more rapidly and cost-effectively,” Lohmeyer said.

Read More »

10 Reasons You Cannot Afford to Miss DCF Trends Summit 2026

The data center industry has no shortage of AI infrastructure ambition. What it lacks is certainty. Power is harder to secure. Designs are advancing faster than facilities can be built. Supply chains remain vulnerable. Liquid cooling is adding operational demands. Projects that look viable on paper can still stall on permitting, commissioning or community opposition. The question in 2026 is no longer how large the AI opportunity may become. It is what can actually be delivered, and who has learned how to deliver it. That question defines the 2026 Data Center Frontier Trends Summit, August 4–6 at the Hyatt Regency Reston. Across three days, the people building, powering, financing and operating next-generation infrastructure will examine what is working, where execution is failing and how the market is responding. This is not a conference about whether AI will create demand. It is about who will be able to meet it. The advantage will belong to those who join the conversation before its conclusions become market consensus. Here are 10 reasons to be in the room. 1. The industry has entered the execution era For several years, the market has been defined by projected demand, capacity, density and investment. The next phase will be defined by execution. AI data center announcements remain abundant. Energized, commissioned and operational capacity is harder to find. DCFTS begins with a live editorial calibration, followed by “The New Geography of AI,” featuring EdgeCore CEO Lee Kestler, Data Center Frontier founder Rich Miller and DCF Editor in Chief Matt Vincent. The focus: how power, entitled land, utility partnerships and execution speed are determining where AI capacity can be built—and who can deliver it. Demand creates opportunity. Execution determines who captures it. 2. Power will be treated as the foundation of AI strategy Power is no longer one workstream

Read More »

Time to Power: Sage Geosystems CEO Cindy Taff on Geothermal’s AI Infrastructure Moment

Three years ago, the data center industry’s energy conversation was largely framed around emissions. Hyperscale operators were setting carbon-free energy targets, signing renewable power agreements, and aligning their expanding infrastructure portfolios with corporate sustainability commitments. The arrival of generative AI has not eliminated those priorities. But it has reordered them. “Three years ago, data center energy, they were really focused on low emissions, no emissions,” said Cindy Taff, CEO of Sage Geosystems. “Now the primary challenge is just enough energy.” Speaking on the Data Center Frontier Show podcast, Taff described an energy market being reshaped by the speed and physical scale of AI infrastructure development. After decades of relatively flat U.S. electricity demand, AI has introduced a new class of concentrated, rapidly arriving industrial load. The result is a shift away from thinking only about how much generating capacity exists in aggregate and toward a harder question: Can usable power be delivered at a specific site, on a predictable schedule, in the quantities an AI campus requires? For hyperscalers, neocloud providers, data center developers, utilities, and energy companies, that distinction is becoming central to project execution. “I think time to power is the most precious metric right now versus cost or total capacity,” Taff said. Capacity on Paper Is Not Power at the Site Announcements of new generation can create the appearance of an energy system capable of meeting rising data center demand. But a megawatt located far from a planned campus, trapped behind a transmission constraint, or unavailable until the next decade has limited value to a developer trying to energize an AI facility within several years. “Aggregate capacity is not going to solve the problem if the power really isn’t where and when you need it,” Taff said. Data centers are large physical facilities tied to specific parcels,

Read More »

Tech Explainer: Data Center Cooling – Air, Evaporative, Liquid, and Hybrid Approaches

Data Center Cooling Glossary The following definitions reflect common terminology used in Department of Energy guidance, ASHRAE TC 9.9 materials, Berkeley Lab resources and Green Grid efficiency metrics. Adiabatic Cooling — A cooling process that uses water evaporation to lower the temperature of air before it reaches a heat exchanger or cooling coil. It can reduce compressor demand but consumes water when evaporative assistance is active. Air-Cooled Data Center — A facility in which heat is removed from IT equipment primarily by moving conditioned air through servers, even if that heat is later transferred to water or refrigerant elsewhere in the cooling system. Air Handler — Equipment that moves, filters and conditions air before delivering it to a data hall or other controlled space. Air-Side Economizer — A system that uses suitable outdoor air, either directly or mixed with return air, to reduce or avoid compressor-based refrigeration. Airflow Management — The practice of delivering conditioned air where it is needed while preventing hot exhaust air from recirculating into server inlets. Approach Temperature — The temperature difference between the two fluids leaving a heat exchanger at their closest thermal point. In a cooling tower, it commonly refers to the difference between leaving-water temperature and entering-air wet-bulb temperature. A smaller approach generally indicates more effective heat transfer. ASHRAE TC 9.9 — The ASHRAE technical committee focused on mission-critical facilities, data centers, technology spaces and electronic equipment. It is a major source of environmental and thermal guidance for data center operators and equipment manufacturers. Blanking Panel — A panel installed in unused rack spaces to prevent hot exhaust air from recirculating to server intakes. British Thermal Unit, or BTU — A unit of heat energy commonly used to express the heating or cooling capacity of equipment. Cabinet — An enclosure, also commonly called

Read More »

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.

Read More »

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

Read More »

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

Read More »

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

Read More »
More Articles