The multi-year deal reportedly dissolved not over pricing or capacity constraints, but because of rigid federal regulatory hurdles.

The breakdown underscores a stark new reality in the artificial intelligence (AI) boom: even the world’s most capitalized technology titans are facing severe shortages of computing power, forcing fierce rivals into complex infrastructure-sharing partnerships.

According to sources familiar with the matter, the multibillion-dollar transaction foundered on a single government security standard, the Federal Risk and Authorization Management Program (FedRAMP). Microsoft reportedly required the leased server capacity to meet these stringent metrics to accommodate highly lucrative, government-adjacent workloads.

However, Oracle’s public cloud architecture lacked specific FedRAMP certifications required for the deal. Sources indicate Oracle executives balked at the immediate implementation, citing the massive engineering overhead and compressed timeline required to retrofit its public cloud framework. Without certification, the strategic value of the lease evaporated for Microsoft, prompting the company to walk away from the negotiating table entirely.

Oracle has aggressively pushed back against the characterization of a failed deal, though it stopped short of providing specific corrections.

“The details mentioned in the article are inaccurate,” an Oracle spokesperson said in a statement. “Microsoft is both an OCI (Oracle Cloud Infrastructure) partner and a customer. We have a tremendously collaborative and fruitful partnership, where we often talk about ways we can expand upon our ongoing work together.”

Microsoft declined to comment on the matter but is reportedly still evaluating alternative infrastructure leasing options. Following Tuesday’s reports, shares of Microsoft fell 2%, while Oracle stock dropped 2.3%.

The appetite for raw computing power has forced tech giants into unprecedented spending sprees. Microsoft has recently committed $25 billion (AUD) to infrastructure in Australia and $3.2 billion to Sweden, yet its internal build-outs still struggle to keep pace with generative AI demands.

Concurrently, Oracle spent $55.7 billion on data centers in its last fiscal year — anchored largely by an expansive partnership with OpenAI. Analysts note that a $3 billion lease to Microsoft would have provided crucial diversification for Oracle, mitigating the financial risk of over-relying on a limited pool of AI clients.

The conflicting accounts emerge amid broader Wall Street anxieties.

Analysts at Goldman Sachs recently warned that the AI infrastructure build-out has entered a “moment of vulnerability,” where massive forward capital expenditures risk outpacing near-term commercial demand. Whether the collapse represents a localized regulatory dispute or a broader willingness by tech giants to reject unfavorable terms, it serves as a potent reminder that the unglamorous machinery of compliance can still halt the momentum of the AI gold rush.

The post Compliance Clashes Stall $3 Billion Cloud Accord Between Microsoft, Oracle: Report appeared first on Techstrong IT.

GENCI plans to install the machine at the Très Grand Centre de Calcul, a supercomputing center near Paris, and make it available for research in 2027. The agency will connect the system to its Joliot-Curie supercomputer so researchers can develop workflows dividing work between traditional HPC and quantum processors. The system will later connect to Alice Recoque, the European exascale supercomputer planned for the same site, GENCI said.

Alice & Bob built its technology to address quantum computing’s persistent error problem using cat qubits. Today’s quantum computers remain highly susceptible to environmental noise that can corrupt calculations. Developers generally address that problem by combining many error-prone physical qubits into a more reliable logical qubit, but the required hardware remains one of the field’s largest obstacles.

Alice & Bob’s cat qubits are designed to reduce the number of physical qubits needed for error correction by suppressing bit-flip errors at the qubit level. Named for the Schrödinger’s cat thought experiment, a cat qubit encodes quantum information in two opposite coherent states of a microwave field inside a superconducting circuit, and the company uses an engineered two-photon stabilization process to maintain those states. As the average number of photons increases, bit-flip errors become exponentially less likely.

That protection does not eliminate every source of error. Cat qubits remain vulnerable to phase flips, which disrupt how the states in a quantum superposition relate to each other and still require error correction. But by suppressing bit flips at the hardware level, the architecture could reduce the number of physical qubits needed to build reliable logical qubits. So far, that potential hardware advantage has been supported by theoretical work and small experimental systems but has not yet been demonstrated at a practical scale.

GENCI intends to use the system as a research platform for quantum error correction and hybrid computing, giving users experience with techniques designed for future quantum computers. The agency describes the purchase as the world’s first acquisition of a cat qubit quantum computer and the first early fault-tolerant quantum computer (eFTQC) planned for a European supercomputing center. The Alice & Bob machine will join Ruby, a neutral atom system from Pasqal, and Lucy, a photonic system from Quandela, giving researchers access to three different quantum computing architectures at one site.

Once operational, the machine will be available at no charge to academic and industrial researchers conducting open research through GENCI’s eDARI platform. The purchase is fully funded through the France Hybrid HPC Quantum Initiative, a €72.3 million program under France 2030.

The acquisition is also part of France’s long-term quantum program. GENCI said the system will help users prepare for two larger universal quantum prototypes that PROQCIMA, a program led by the French Ministry of the Armed Forces, aims to deliver by 2032. In May, President Emmanuel Macron announced another €1 billion for the strategy from 2026 through 2030.

The post France to Integrate Alice & Bob Cat-Qubit System With National HPC appeared first on Techstrong IT.

The easy conclusion is that teams are inefficient.

That’s usually wrong.

The Work Isn’t the Problem

A significant portion of time in IT isn’t spent on “the work.” It’s spent on everything around it. Downloading repositories, updating libraries, gathering requirements, aligning on dependencies. None of that is waste. It’s preparation.

It’s the modern version of the quote often attributed to Abraham Lincoln: if you had an hour to chop down a tree, you’d spend most of it sharpening the axe.

That work matters. But it’s not what’s slowing teams down.

The Real Cost Is Context Switching

What actually gets in the way is everything layered on top of it: meetings to organize the meeting, status updates with little or no substance, the classic “this meeting could have been an email.” And while those are easy to joke about, the real problem isn’t just that they take time. It’s that they break focus.

Every time someone shifts from deep work into a meeting, they’re not just losing those 30 minutes. They’re switching contexts. They’re moving from one mental space into another, often juggling priorities, personalities, and politics, and then trying to get back to where they were.

That transition has a cost.

Studies estimate it takes more than 20 minutes to fully return to the original task after an interruption. To make matters worse, research shows that interrupted work can take twice as many errors compared to uninterrupted work.

So when we talk about team productivity, we’re often measuring the wrong thing. The issue isn’t that people aren’t working hard or that they don’t know what to do. It’s that they’re constantly being pulled away from the conditions that allow them to do their best work in the first place.

This is a Leadership Problem

Here’s the uncomfortable part: Most of that interruption is organizational, not individual. Meetings get scheduled. Status updates are required. Check-ins have become the cultural default. None of it is malicious, but it accumulates and the team pays the cost.

The leaders who move the needle aren’t necessarily the best coders or the sharpest architects. They’re the ones who do something simpler and harder: they ask their teams what’s getting in the way, and then they actually do something about it.

If the team says they work better without meetings on Mondays, leaders should believe them and protect that time. If staff says a particular reporting requirement is pure overhead, leaders need to get specific about the details they need, and solicit better ways to collect that information. If they say they need fewer interruptions to ship quality work, that’s not a preference. That’s a performance variable.

Leadership needs to recognize that this isn’t a one-time fix, either. The first attempt might not be right, but keep in mind that it’s iteration, not failure. Ignoring the feedback entirely isn’t a neutral choice. It’s a decision to keep the problem.

Fix the System, Not the People

Most IT teams aren’t failing because they can’t do the work. They’re failing because the environment makes the work nearly impossible.

Fixing inefficiency in IT isn’t about squeezing more output out of people. It’s about creating the conditions where meaningful work can actually happen.

The real question isn’t “why is my team slow?” It’s “what have we done to make speed possible?”

The post The Real Reason Your IT Team Isn’t Getting Anything Done appeared first on Techstrong IT.

A coalition of tech companies is committing an additional $915 million to carbon removal projects, supporting a sector that many climate experts view as essential for achieving long-term emissions targets.

The post Anthropic Joins Frontier as Carbon Removal Initiative Commits Another $915 Million appeared first on Techstrong IT.

HPE is stepping up its effort to attract organizations evaluating alternatives to VMware, unveiling a migration program that provides one year of free HPE Morpheus VM Essentials licensing and access to HPE Zerto for a nominal $1 fee during the transition period.

The post HPE Expands Offensive Against VMware With Migration Incentives appeared first on Techstrong IT.

Our goal in these challenging economic times is to make it just that much easier for IT professionals to advance their careers.

Of course, the pool of available IT talent is still relatively constrained, so when one IT professional takes on a new role, it tends to create opportunities for others.

The ten job postings shared this week are selected based on the company looking to hire, the vertical industry segment and naturally, the pay scale being offered.

We’re also committed to providing additional insights into the state of the IT job market. In the meantime, for your consideration.

Greenhouse

SpaceX
Palo Alto, CA
Principal Network Engineer (Starlink Ground Network)
$200,000 to $285,000

Lever

Commonwealth Fusion Systems
Milpitas, CA
IT Infrastructure Engineer
$150,000 to $250,000

Ashby

Crusoe
San Francisco, CA
Senior Network Engineer, Deployment
$162,000 to $196,000

Gamma
San Francisco, CA
Site Reliability Engineer
$230,000 to $310,000

Sonio
Remote, US
Site Reliability Engineer (SRE)
$165,000 to $190,000

ALSO
Palo Alto, CA
Staff Site Reliability Engineer – Cloud Platform & Vehicle Telemetry
$200,000 to $245,000

Superhuman
Remote, US
Site Reliability Engineer
$214,000 to $260,000

Favorited
Santa Monica, CA
Senior Site Reliability Engineer (SRE)
$150,000 to $200,000

Vultr
Remote, US
Senior Software Engineer, Cloud Networking
$128,000 to $145,000

Meter
San Francisco, CA
Software Engineer, Models
$170,000 to $250,000

The post Ten Great IT Job Opportunities appeared first on Techstrong IT.

Captains of Industry

This strategy is a clear pivot from simple hardware sales to a cohesive enterprise ecosystem. They are introducing UniFi Fabrics to handle centralized policy orchestration across thousands of sites, backed by integrated identity management and zero touch provisioning workflows. There is even an enhanced partner program designed to provide certified support and priority equipment allocations, which signals a mature understanding of how enterprise purchasing actually works. The goal is to offer a license-free management experience that scales up to meet the demands of global organizations without the standard recurring cost penalties.

Nowhere is this engineering ambition more obvious than in their new hardware lineup, particularly the U7 Industrial access point. Dubbed the “Batman AP” during the presentation due to its wing-like side panels, this device tackles the chaotic RF environments of warehouses and manufacturing plants through physical engineering. Instead of relying purely on software beamforming, which often struggles in environments with shifting physical obstacles. Administrators can manually adjust these integrated reflectors to set beam widths at sixty, seventy, or ninety degrees. In practice, this mechanical approach acts like an old-school radar waveguide, allowing engineers to physically isolate and direct signals away from metallic obstructions.

The hardware implementation includes intelligent internal sensors that read the exact physical placement of these wings. This structural data is sent back to the UniFi software, allowing the AP to automatically tune its transmit power to maintain strict regulatory compliance based on the configured antenna shape. This monitoring capability also serves as a brilliant real-world operational tool. If a warehouse forklift strikes the access point or a bird lands on the chassis and alters the reflector position, the system can immediately trigger an administrative alert. The radio itself is a dual-band design that intentionally omits the 6 GHz spectrum, which simplifies worldwide regulatory compliance and speeds up international deployments. It also includes standard SMA connectors for environments that demand custom external cabling, and it runs on standard PoE+ power.

More Than Just Wi-Fi

This ruggedized philosophy extends down to the routing and switching layers with the new UniFi Industrial lineup. The UCG Industrial Gateway is a temperature-hardened unit designed for punishing deployments like broadcast trucks and mobile roadie cases. It features 10G WAN and LAN ports alongside flexible power options, including a four-pin Molex connector and a fifty-four volt DC input. It also integrates physical SIM slots that interface directly with the external U5G Outdoor Max modem. This configuration allows you to mount the cellular modem high up on an outdoor tower while keeping the physical SIM cards safely accessible in the ground-level gateway. Companion industrial switches are also joining the family, bringing DIN-rail mounting, dual DC inputs, and fundamental enterprise features like VLAN tagging and 802.1x authentication to the plant floor.

Ubiquiti is also addressing high-availability campus needs with their new ECS Stackable Switches. This line introduces true physical stacking to the UniFi portfolio, supporting up to eight units in a resilient loop or ring topology. The front panel delivers modern access speeds at 2.5 Gbps and 10 Gbps, while the backplane relies on dedicated 100 Gbps interconnects for the stack fabric. This architecture allows administrators to aggregate ports across different physical switches in the stack, providing robust link redundancy and load balancing. Interestingly, Ubiquiti has embraced the tongue-in-cheek marketing term PoE+++ for these campus units to denote 90-watt 802.3bt Type 4 power, simply because their engineering team found repeating the official IEEE nomenclature too tedious.

Taken together, these developments show an organization that is listening closely to the pain points of field engineers. Ubiquiti is no longer just building inexpensive access points; they are designing specialized tools tailored for environments where standard enterprise hardware often fails.

Bringing IT All Together

Ubiquiti is making a sophisticated play for the enterprise market by solving physical, real-world deployment challenges rather than just competing on feature checklists. The mechanical innovation of the U7 Industrial AP demonstrates that they understand how unpredictable warehouse RF can be, and the inclusion of true 100GbE stacking switches bridges a gap in their enterprise campus portfolio. By pairing this rugged, high-capacity hardware with license-free orchestration software, Ubiquiti is positioned to seriously disrupt traditional enterprise pricing models. Traditional networking incumbents should stop treating them like a prosumer alternative and start recognizing them as a legitimate architecture competitor.

To learn more about Ubiquiti’s solutions for campus, enterprise, and industrial applications, make sure to check out their website at https://ui.com. To see their entire Mobility Field Day presentation, make sure to check out the presentation appearance page here.

The post Super Industrial Networking from Ubiquiti appeared first on Techstrong IT.

Bloom Energy’s latest Data Center Power Report finds that access to electricity has emerged as the dominant site-selection factor. Bloom found that 84% of operators now identify power availability as the most important consideration when evaluating new locations.

The challenge has become significant enough that developers are now more often planning to generate electricity themselves. The survey found that 61% of developers would deploy onsite power systems if utility providers cannot deliver power on required timelines.

That shift is occurring at a pace few industry observers expected. Last year’s survey projected that only 13% of facilities would depend primarily on onsite generation by 2030. The latest survey now forecasts between 27% and 38% of data centers will rely on onsite power as their primary energy source by the end of the decade.

Even more dramatically, 27% of facilities are expected to operate entirely on self-generated power by 2030. In the previous survey, only 1% of operators anticipated fully onsite-powered facilities.

Environmental considerations are also becoming part of energy planning. Nearly one-third of facilities using onsite power generation are expected to incorporate carbon-capture technology by 2030.

AI Data Centers

The Bloom report projects that AI-focused facilities will account for a growing share of future capacity additions, increasing from 13% of new deployments in 2026 to 23% by 2030.

Demand is being driven largely by inference workloads. According to the report, inference now represents more than 50% of AI computing activity, demonstrating the shift from training models to running AI applications in production environments, and the adoption of agentic AI.

The scale of future projects is also getting larger. Bloom found that one in five data center campuses is expected to exceed one gigawatt of capacity by 2030. By 2035, that figure rises to one in three campuses.

Texas is becoming one of the industry’s most important markets. The report projects that data center-related electricity demand in the state could exceed 40 gigawatts by 2028.

Possibly slowing all of this growth: developers are encountering increasing resistance from local governments and residents concerned about resource consumption. Bloom reported that at least 18 state-level bills and 86 local moratorium proposals related to data center development had been introduced across the US as of May 2026. Respondents identified electricity prices, water usage and grid reliability as the issues most likely to influence community opposition to future projects.

Hardware Disconnect

The report also revealed a growing disconnect between data center development schedules and advances in AI hardware. Chip manufacturers expect high-density computing architectures and rack-level direct-current power systems to become mainstream in 2028. In contrast, data center developers do not expect to deploy those technologies until approximately one year later.

That timing gap could complicate future deployments as AI hardware requirements continue to escalate.

The survey was conducted in April 2026 and included 156 decision-makers across the data center ecosystem, including hyperscalers, colocation providers, chip developers and data center operators. Of those surveyed, 79% were based in the US.

Bottom line, the Bloom survey reveals a data center industry experiencing major growing pains, with challenges ranging from limits on power generation, hardware issues, and public scrutiny at greater levels than ever before.

The post Survey: AI Data Center Boom Runs Into Power Shortages and Local Resistance appeared first on Techstrong IT.

According to multiple reports, OpenAI is discussing a 20-year lease arrangement for a campus designed to deliver up to 10 gigawatts of power. At current estimates for hardware, energy infrastructure, construction, and labor, total development costs could exceed $500 billion.

The facility would be developed by SB Energy, a SoftBank-owned energy and infrastructure company. The site would span both federal property managed by the US Department of Energy and adjacent privately owned land in southern Ohio.

Under the reported structure, OpenAI would lease the campus rather than own it outright. The company would manage and operate the computing equipment deployed at the facility, with lease payments beginning once operations commence. Initial capacity is expected to come online in 2028.

NVIDIA is expected to play a central role in the project. Reports indicate the company would supply the AI hardware used throughout the campus and could also provide financial support by guaranteeing portions of OpenAI’s lease commitments.

The project’s planned 10-gigawatt scale matches the capacity target originally outlined for Stargate, the AI infrastructure initiative announced by OpenAI, SoftBank, and Oracle in early 2025.

For OpenAI, securing that capacity could become a critical step in supporting the next generation of compute-intensive AI models.

A Public-Private Initiative

The location carries historical significance as well. The proposed development is tied to the Portsmouth Site in Pike County, Ohio, a former uranium enrichment facility that operated during the Cold War before ending operations in 2001.

Earlier this year, the US Department of Energy announced a public-private initiative involving SoftBank and AEP Ohio aimed at redeveloping the area. Plans include construction of 10 gigawatts of new power generation capacity and approximately $4.2 billion in transmission upgrades to support large industrial and technology projects.

Most of the planned electricity generation would come from natural gas facilities. Public announcements tied to the redevelopment project indicated that much of the planned generation would be supported by Japanese investment capital.

Clearly, leading AI model developers are investing heavily to secure access to computing resources as demand for AI training and inference continues to rise. Data center construction has become a top priority as companies compete to build not just better AI models but the larger data facilities to support them.

OpenAI has been particularly focused on expanding infrastructure access. Earlier this year, the company paused a proposed data center initiative in the UK, citing regulatory challenges and energy costs. The Ohio project would provide a substantially larger domestic alternative if negotiations ultimately lead to a finalized agreement.

The development would also reinforce growing ties between OpenAI, SoftBank, and NVIDIA. SoftBank has expanded its involvement in AI infrastructure investments, while NVIDIA not only remains the dominant supplier of advanced chips for AI models but also works to expand its product offerings across the data center sector. Neither OpenAI nor NVIDIA has publicly confirmed the reported discussions.

The post OpenAI Explores AI Data Center That Could Cost More Than $500 Billion appeared first on Techstrong IT.

Working with support from Google, the team plans to deploy a computing environment built from approximately 2,000 retired Pixel smartphones. Rather than using the devices in their original form, researchers remove components unnecessary for data center operations, including displays, batteries, cameras, speakers, and outer casings. The remaining motherboard, which contains the processor, memory, storage, and other computing elements, becomes the foundation of the computing platform.

The unusual approach addresses a major environmental problem associated with computing infrastructure. While the tech sector focuses on reducing the energy consumed by data centers, a substantial portion of technology’s environmental impact comes from manufacturing new hardware. Researchers believe extending the useful life of existing devices, even basic smartphones, could help reduce that burden.

Comparable to Servers

The project is based on a finding that surprised many observers: processor cores in relatively recent smartphones can deliver single-threaded benchmark performance comparable to, and in some cases higher than, individual cores in server processors.

Servers, of course, maintain major advantages in memory capacity, scalability, and throughput, but the research results suggest older phones remain capable of supporting a range of computing workloads.

To make the hardware suitable for cloud-style deployment, researchers replace Android with a general-purpose Linux OS. The switch eliminates mobile-specific software limitations and allows the phones to run standard infrastructure tools. Kubernetes is then used to coordinate workloads across large groups of devices, enabling clusters of smartphones to function as a unified computing platform.

Benchmark testing indicates that between 25 and 50 smartphones can provide computing performance comparable to a modern server CPU for certain applications. Researchers organize the devices into clusters of that size, allowing software workloads to be distributed across multiple phones.

The initial focus is on educational and research use cases rather than demanding commercial workloads. Early testing showed that a cluster of roughly 20 smartphones could handle peak assignment submission activity for a university course with more than 75 students. Still, researchers reported performance that was lower than the AWS-based backend previously used for the course.

At full scale, the planned 2,000-phone deployment is expected to support roughly one hundred similar courses simultaneously. The system will be used for computer science programs including systems programming and parallel computing, while also serving as a research platform for studying smartphone-based infrastructure.

Beyond cost savings, the project offers a chance to examine how consumer-grade hardware performs under continuous operation. Smartphones are designed for personal use rather than around-the-clock data center workloads. Researchers will evaluate reliability and maintenance needs as the system operates over longer periods.

The motherboard-focused approach is important because researchers estimate the motherboard accounts for roughly half of a smartphone’s manufacturing-related carbon footprint. Reusing the motherboard preserves a large portion of the embedded resources and emissions associated with the device.

This new use of smartphones is not meant to replace data centers or the specialized infrastructure used for AI training. Hyperscalers require dense, highly reliable hardware optimized for massive workloads. But researchers believe smartphone clusters may offer a practical alternative for educational institutions and organizations with limited budgets.

The full 2,000-device platform is expected to enter operation in 2026.

The post Researchers Turn Retired Smartphones Into Low-Cost Computing Clusters appeared first on Techstrong IT.