In short: OpenAI has paused its Stargate UK data centre project, citing the high cost of industrial electricity in Britain and an unfavourable regulatory environment around AI copyright. The project, announced in September 2025 alongside Nvidia and Nscale, had planned to deploy 8,000 GPUs at sites in north-east England, scalable to 31,000 over time. OpenAI says it will move forward “when the right conditions” allow, though it has given no timeline. The pause is a significant setback for the UK government’s AI Growth Zones initiative and arrives as OpenAI prepares for a public listing.

What Stargate UK was supposed to be

Stargate UK was announced in September 2025 as a sovereign AI infrastructure project: a partnership between OpenAI, Nvidia, and British cloud provider Nscale to build data centre capacity in north-east England that would allow OpenAI’s models to run on local computing power. The sites earmarked were Cobalt Park near Newcastle and Blyth, both within the UK government’s designated AI Growth Zones, a framework the government had positioned as a centrepiece of its industrial strategy for artificial intelligence. The project was unveiled during US President Donald Trump’s state visit to Britain, giving it diplomatic as well as commercial significance. The initial phase involved off take of approximately 8,000 Nvidia AI processors, with an ambition to scale that to 31,000 GPUs over time,  capacity that would have enabled OpenAI to serve critical public services, regulated industries such as finance, and national security partnerships without routing data through US-based infrastructure. OpenAI never disclosed the total investment figure associated with the UK project. The broader US Stargate project remains on track with data centre construction under way across the United States, backed by a $40 billion bridge loan SoftBank secured to finance its participation, making the UK pause a geographic exception rather than a signal of retreat from AI infrastructure spending overall.

The energy cost problem

The most concrete obstacle OpenAI identified is the cost of electricity in Britain. UK industrial electricity prices are among the highest of any IEA member state,  more than four times those in the United States, Finland, Norway, and Sweden. For a data centre drawing 100 megawatts, that differential is not a line-item concern but a structural one: the economics of running large-scale AI inference workloads at a site where power costs four times as much as they do in Virginia or Texas are fundamentally different, and that gap compounds as capacity scales. The problem is not simply a matter of electricity tariffs. Grid connection requests in the UK surged from 41 gigawatts in November 2024 to 125 gigawatts by June 2025, with an estimated 75 gigawatts of that queue attributable to data centre projects. Buildings can be constructed in 18 to 24 months; grid connections take three to eight years. That mismatch means that even if a project clears the financial hurdle, it faces an infrastructure queue that the current regulatory and planning framework has not been designed to process at AI-infrastructure speeds. The UK government’s AI Growth Zones policy, published in November 2025, was intended in part to address exactly this bottleneck, but the zone designations do not resolve the underlying grid constraints, and OpenAI’s decision to pause suggests that the policy framework has not yet translated into the conditions that would make the investment viable.

The copyright sticking point

The regulatory concern OpenAI cited alongside energy costs points to a separate and more politically charged problem: the UK’s unresolved approach to AI copyright. UK lawmakers have been working to update the rules governing how AI models are trained on copyrighted material. The government’s preferred approach, a broad text and data mining exception with an opt-out mechanism for rights holders, was rejected by the majority of respondents to the government’s own consultation, with creative industries, publishers, and news organisations arguing that a broad exception would allow generative AI companies to train on their works without compensation or meaningful consent. The consultation produced no consensus, and the government has since delayed any legislative change. For OpenAI, which trains large language models on text scraped from the internet, the uncertainty about whether that training will be lawful, and on what terms, is a material business risk. A UK data centre is not simply a power facility,  it creates legal jurisdiction. If the UK eventually adopts a copyright framework that restricts training data use more tightly than the US, operating infrastructure in Britain could expose OpenAI to liability or compliance costs that do not apply to its US operations. The pause allows OpenAI to wait for that regulatory picture to clarify before committing capital.

A pause, not a cancellation, and the IPO context

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

OpenAI’s statement was calibrated to leave the door open. “We continue to explore Stargate U.K. and will move forward when the right conditions such as regulation and the cost of energy enable long-term infrastructure investment,” the company said, framing the decision as contingent rather than final. The timing, however, is notable. OpenAI closed a $122 billion funding round at an $852 billion valuation in late March 2026, extending participation to retail investors for the first time in a move widely interpreted as groundwork for a public offering analysts expect as early as the fourth quarter of 2026. Companies approaching an IPO typically tighten their capital allocation discipline, avoid open-ended international commitments that could weigh on reported cash burn, and reduce exposure to projects with uncertain timelines. Pausing a data centre project that faces both energy cost headwinds and an unresolved copyright regime fits that pattern. The UK government, which had promoted Stargate UK as a signal of international investor confidence in Britain’s AI ambitions, described the decision as disappointing and said it remained in dialogue with OpenAI. OpenAI’s international Stargate expansion has not been without complications elsewhere either, its Abu Dhabi data centre plans drew an explicit threat from Iranian authorities amid escalating regional tensions, suggesting that sovereign AI infrastructure projects carry geopolitical risk profiles that are becoming a distinct factor in OpenAI’s site selection calculus. Meanwhile, Oracle appointed a new CFO this week to manage its $50 billion data centre construction programme as the central operating partner in the US Stargate project,  a contrast that illustrates where AI infrastructure spending remains active and where it is being reconsidered. The year 2025 established infrastructure access and energy as the primary competitive variables in AI,  and for the UK, OpenAI’s pause is a signal that it has not yet solved either.

Gemini is getting a new feature in the form of notebooks, further integrating with NotebookLM, Google announced on Wednesday.

NotebookLM is easily one of Google’s best AI tools available. It allows you to add sources to a notebook and ask questions or generate a series of outputs, based only on the material you’ve given it, making it self-contained. 

Advertisement

It’s different from the Gemini chatbot on its own, which will search the entire internet for an answer. NotebookLM’s answers are grounded only in your sources. It’s great for studying for school, getting prepped for work and even creative inspiration. There’s no wrong way to use it, and its flexibility lends itself to be explored, so it comes as no surprise that it’s getting further integrated into Google’s own chatbot.

Late last year, Google allowed you to add a NotebookLM notebook into Gemini so it would have all the context you added. Google says to look at notebooks as personal knowledge bases. You can add files, past chats and documents on a particular subject, and you can always jump back into that conversation with Gemini and pick up where you left off. Having a dedicated space to keep things in one place is great on its own for Gemini, but you can use those notebooks in NotebookLM, too. 

Notebooks make it easier to keep all the information on a particular subject in one place and give Gemini everything it needs without having to manually add details all over again. Now, that integration goes even further, and you can create notebooks directly in Gemini. 

Notebooks can be synchronized between Gemini and NotebookLM, making the data in one tool instantly available in the other. This means you can add artifacts to the notebook in Gemini and immediately have the option to create a Video Overview, Infographic or other outputs within NotebookLM. Having the same synced notebook in both tools allows you to use each tool in its own way with the same database.

When the feature is available, you’ll see a new notebooks section in the side menu bar of Gemini, allowing you to quickly create one or access others you’ve made in the past. 

Google is rolling out Notebooks in Gemini to AI Ultra, Pro and Plus plans on the web, and will expand access to mobile, more countries across Europe and free users in the coming weeks. 

We’ve teamed up with T10 BESPOKE to raffle off a one-of-a-kind T10 BESPOKE, eCoustics Special Edition, True Luxury Wireless Hi-fidelity In-ear Computers ($4,000 USD value) custom made by HiFi Bear himself!

How to Enter

Buy a raffle ticket online for $50 each, to be entered into a random drawing after all tickets are sold.

Odds of Winning

We’re only selling 80 tickets for this raffle so each ticket buys a 1/80 chance.

See Them at AXPONA

They’ll be on display at AXPONA at the Renaissance Convention Hotel in Schaumburg, IL, from April 10-12, 2026 (2nd level lobby Left Hand Living Room just past the AXPONA help desk and directly adjacent to “the Gather Bar”).

The Prize

One T10 BESPOKE, eCoustics Special Edition, True Luxury Wireless Hi-fidelity In-ear Computers includes one pair of wireless earbuds and one pendant charging case in an exclusive, eCoustics Special Edition design ($4,000 USD value) as shown in the photos.

Product Details

Solid mirror-polished brass and snow-white ceramic zirconium, highlighted with blood-red natural reticulated Python skin, and custom eCoustics logo livery. The units feature high res DAC, Twin Cadence Tensilica Hi-Fi DSP’s, Class D and Class AB selectable amplification, quad Knowles Quartz series analog microphones, and custom Sonion balanced armatures (absolutely stunning hi-fi sound simply beyond anything you’ve ever experienced. Dual ARM M4 CPU and M0 co-processors enable on-board voice recognition and control, while 6-axis accelerometer/gyroscopes convert subtle head-motions into gestures to control music, telephony, and more. All this, in what is at the same time the world’s smallest and lightest wireless in-ear monitor; less than half the size and weight of an AirPod. All T10 Bespoke are built with a novel chassis, mechanically assembled with fine-pitch custom stainless steel watch screws which enable the units to be easily serviced, repaired, and upgraded with new hardware technology, forever, preventing technological obsolescence and enabling the lifetime warranty offered by EAR Micro.

Learn more about T10 Bespoke.

Florida Attorney General James Uthmeier announced on Thursday his office will investigate OpenAI for its alleged harm to minors, potential to threaten national security, and its possible link to a shooting that took place at Florida State University last year.

“ChatGPT may likely have been used to assist the murderer in the recent mass school shooting at Florida State University that tragically took two lives,” Attorney General Uthmeier said in a video posted to social media.

On the day of the FSU shooting last April, the suspect allegedly asked ChatGPT how the country would react to a shooting at FSU, and what time it would be busiest at the FSU student union. These messages could potentially be used as evidence against the suspect in an October trial about the shooting.

The attorney general cited further concerns about ChatGPT’s encouragement of suicide in certain instances, which have been documented in multiple lawsuits brought by families against OpenAI. He also mentioned his concern that the Chinese Communist Party could use OpenAI’s technology against the United States.

“As big tech rolls out these technologies, they should not — they cannot — put our safety and security at risk,” he said. “We support innovation. But that doesn’t give any company the right to endanger our children, facilitate criminal activity, empower America’s enemies, or threaten our national security.”

OpenAI is throwing its support behind an Illinois state bill that would shield AI labs from liability in cases where AI models are used to cause serious societal harms, such as death or serious injury of 100 or more people or at least $1 billion in property damage.

The effort seems to mark a shift in OpenAI’s legislative strategy. Until now, OpenAI has largely played defense, opposing bills that could have made AI labs liable for their technology’s harms. Several AI policy experts tell WIRED that SB 3444—which could set a new standard for the industry—is a more extreme measure than bills OpenAI has supported in the past.

The bill would shield frontier AI developers from liability for “critical harms” caused by their frontier models as long as they did not intentionally or recklessly cause such an incident, and have published safety, security, and transparency reports on their website. It defines a frontier model as any AI model trained using more than $100 million in computational costs, which likely could apply to America’s largest AI labs, like OpenAI, Google, xAI, Anthropic, and Meta.

“We support approaches like this because they focus on what matters most: Reducing the risk of serious harm from the most advanced AI systems while still allowing this technology to get into the hands of the people and businesses—small and big—of Illinois,” said OpenAI spokesperson Jamie Radice in an emailed statement. “They also help avoid a patchwork of state-by-state rules and move toward clearer, more consistent national standards.”

Under its definition of critical harms, the bill lists a few common areas of concern for the AI industry, such as a bad actor using AI to create a chemical, biological, radiological, or nuclear weapon. If an AI model engages in conduct on its own that, if committed by a human, would constitute a criminal offense and leads to those extreme outcomes, that would also be a critical harm. If an AI model were to commit any of these actions under SB 3444, the AI lab behind the model may not be held liable, so long as it wasn’t intentional and they published their reports.

Federal and state legislatures in the US have yet to pass any laws specifically determining whether AI model developers, like OpenAI, could be liable for these types of harm caused by their technology. But as AI labs continue to release more powerful AI models that raise novel safety and cybersecurity challenges, such as Anthropic’s Claude Mythos, these questions feel increasingly prescient.

In her testimony supporting SB 3444, a member of OpenAI’s Global Affairs team, Caitlin Niedermeyer, also argued in favor of a federal framework for AI regulation. Niedermeyer struck a message that’s consistent with the Trump administration’s crackdown on state AI safety laws, claiming it’s important to avoid “a patchwork of inconsistent state requirements that could create friction without meaningfully improving safety.” This is also consistent with the broader view of Silicon Valley in recent years, which has generally argued that it’s paramount for AI legislation to not hamper America’s position in the global AI race. While SB 3444 is itself a state-level safety law, Niedermeyer argued that those can be effective if they “reinforce a path toward harmonization with federal systems.”

“At OpenAI, we believe the North Star for frontier regulation should be the safe deployment of the most advanced models in a way that also preserves US leadership in innovation,” Niedermeyer said.

Scott Wisor, policy director for the Secure AI project, tells WIRED he believes this bill has a slim chance of passing, given Illinois’ reputation for aggressively regulating technology. “We polled people in Illinois, asking whether they think AI companies should be exempt from liability, and 90 percent of people oppose it. There’s no reason existing AI companies should be facing reduced liability,” Wisor says.

Threat actors using a previously undocumented phishing-as-a-service (PhaaS) platform called “VENOM” are targeting credentials of C-suite executives across multiple industries.

The operation has been active since at least last November and appears to target specific individuals who serve as CEOs, CFOs, or VPs at their companies.

VENOM also seems to be closed access, as it has not been promoted on public channels and underground forums, thus reducing its exposure to researchers.

The VENOM attack chain

The phishing emails, observed by researchers at cybersecurity company Abnormal, impersonated Microsoft SharePoint document-sharing notifications as part of internal communication.

The messages are highly personalized and include random HTML noise such as fake CSS classes and comments. The attacker also injects fake email threads tailored to the target, increasing credibility.

A QR code rendered in Unicode is provided for the victim to scan for access. The trick is designed to bypass scanning tools and shift the attack to mobile devices.

“The target’s email address is double Base64-encoded in the URL fragment—the portion after the # character,” Abnormal researchers explain.

“Fragments are never transmitted in HTTP requests, making the target’s email invisible to server-side logs and URL reputation feeds.”

When the victim scans the QR code, they are taken to a landing page that serves as a filter for security researchers and sandboxed environments, ensuring that only real targets are redirected to the phishing platform. Users outside the threat actor’s interest are redirected to legitimate websites to reduce suspicion.

Those who pass the tests are taken to a credential-harvesting page that proxies a Microsoft login flow in real time, relaying credentials and multi-factor authentication (MFA) codes to Microsoft APIs and capturing the session token.

Apart from the adversary-in-the-middle (AiTM) method, Abnormal has also observed a device-code phishing tactic in which the victim is tricked into approving access  to their Microsoft account for a rogue device.

This method has become very popular over the past year due to its effectiveness and resistance to password resets, with at least 11 phishing kits currently offering it as an option.

In both methods, VENOM quickly establishes persistent access during the authentication process. In the AiTM flow, it registers a new device on the victim’s account. In the device code flow, it obtains a token that also provides access to the account.

The researchers note that MFA is no longer sufficient as a defense. C-suite executives should use FIDO2 authentication, disable the device code flow when not needed, and block token abuse by implementing stricter conditional access policies.

Automated pentesting proves the path exists. BAS proves whether your controls stop it. Most teams run one without the other.

This whitepaper maps six validation surfaces, shows where coverage ends, and provides practitioners with three diagnostic questions for any tool evaluation.

Advertisement

Get Your Copy Now

Gerard “Gus” Gaynor, a long-serving IEEE volunteer and former engineering director at 3M, died on 9 March. The IEEE Life Fellow was 104.

Readers of The Institute might remember Gus from his 2022 profile: “From Fixing Farm Equipment to Becoming a Director at 3M.” Just last year, he and I coauthored twoarticles. One discusses how to leverage relationships to boost your career growth. The other weighs the pros and cons of pursuing a technical or managerial career path. He was 103 years old then. How many IEEE members can claim a centenarian coauthor?

I first met Gus in 2009 at the IEEE Technical Activities Board (TAB) meeting in San Juan, Puerto Rico. We sat together in the airplane on our way back to Minneapolis, our hometown. At home I told many of my friends about the remarkable person—who was 87 years young at the time—with whom I chatted during our six-hour flight.

A decade later, he and I met for lunch in Minneapolis. He drove himself to the restaurant, just asking for a hand to navigate the snowy sidewalk.

A dedicated IEEE volunteer

Gus’s involvement with IEEE predates the organization. He joined the Institute of Radio Engineers, a predecessor society, as a student member in 1942. Twenty years later he became an active IEEE volunteer.

He served on the TAB’s finance committee and the Publications Services and Products Board. He was president of the IEEE Engineering Management Society (now the Technology and Engineering Management Society ), and he was the Technology Management Council’s first president. He was the founding editor of IEEE-USA’s online magazine Today’s Engineer, which reported on government legislation and issues affecting U.S. members’ careers. The magazine is now available as the e-newsletter IEEE-USA InSight.

He authored several books on technology management, published by IEEE-USA.

IEEE Life Fellow Gerard “Gus” Gaynor died on 9 March.The Gaynor Family

Most recently, after the formation of TEMS in 2015, he became an active member of its executive committee. He served two terms as vice president of publications.

At 100 years old, he led the launch of a new publication, TEMS Leadership Briefs, a novel short-format open-access publication aimed at technology leaders.

Gus, who is a former member of The Institute’s editorial advisory board, also worked with Kathy Pretz, The Institute’s editor in chief, to start an ongoing series of TEMS-sponsored career-interest articles. He coauthored several of them.

Throughout his 64 years as an IEEE volunteer, he received several honors. They include IEEE EMS’s Engineering Manager of the Year Award, the IEEE TEMS Career Achievement Award, and the IEEE-USA McClure Citation of Honor. In 2014 he was inducted into the IEEE Technical Activities Board Hall of Honor.

A 25-year career at 3M

Gus received a degree in electrical engineering in 1950 from the University of Michigan in Ann Arbor. He worked for several companies including Automatic Electric (now part of Nokia) and Johnson Farebox (now part of Genfare), before joining 3M in 1962.

During his successful 25-year career at 3M, he served as chief engineer for a division in Italy, established the innovation department, and led the design and installation of the company’s first computerized manufacturing facilities. He retired as director of engineering in 1987.

Last year, IEEE Life Fellow Michael Condry, a former TEMS president, organized a Zoom call with Gus and other leaders of the society to celebrate Gus’s 104th birthday. Gus looked well and was his usual upbeat self, telling everyone: “I’m good. Everything’s well. I can’t complain.”

Gus was married to Shirley Margaret Karrels Gaynor, who passed away in 2018. He lives on in the hearts and minds of his seven children, seven grandchildren, two great-grandchildren, and innumerable friends and IEEE colleagues.