Compute Exchange and Silicon Data, Bochev added “are also working on developing clearer benchmarks for the compute market, and will have more details to share on that in the coming weeks.” PIC ‘should serve to keep suppliers honest ..’ Scott Bickley, an advisory fellow at Info-Tech Research Group, said he views the offering “as a way for enterprises to source short-term GPU capacity and possibly get a deal, especially if it is stranded capacity from the neocloud providers.” This, he said, “would also help to benchmark costs when purchasing this capacity in general, so it’s good, but it is also straightforward in terms of the value proposition.” He also noted that most companies are not buying GPU capacity directly; “This is for those that are building their own models or deploying their own AI applications atop existing models.” Bickley added, “it should serve to keep suppliers honest to some degree in terms of the floors and ceilings of the price to access GPU capacity.” Soon after Compute Exchange first launched in February, Matt Kimball, VP and principal analyst for data center compute and storage at Moor Insights & Strategy, described the GPU compute situation as “pretty dire. This is driven by what most view as a single supplier (Nvidia) selling GPUs before they can even be made to a market that has an insatiable thirst.” On Tuesday, following the announcement, he said that the concept of PIC is appealing: “I really like the idea of PIC as a tool for customers and seeing the compute exchange become an arbitrageur of sorts. This delivers a real value to [anyone] who is looking to utilize AI infrastructure,” he said.

Data center operators reported limited advances—and even some declines—in energy efficiency, carbon tracking, and water usage due in part to rising power demand and easing regulatory pressure in some regions, according to the recently released results of the Uptime Institute’s 15th Annual Global Data Center Survey 2025. As artificial intelligence workloads continue to grow and legacy data centers remain operational, sustainability initiatives have stalled, according to the Uptime Institute, which attributes this in part to reporting challenges. Uptime Institute’s 2025 data center survey was conducted online from April 2025 to May 2025 and collected responses from more than 800 data center owners and operators and more than 1,000 vendors and consultants.  “What’s interesting this year is that we have seen a far from startling increase over the last few years of the data being collected, but this year it actually fell. And this obviously led to some speculation that there is a backing off of sustainability, and that it is no longer a high priority,” said Andy Lawrence, executive director of research at Uptime Institute, during a webinar sharing the survey results. “I think that the data center industry has not yet adapted to being very good at sustainability reporting.”

“We also think UALink is another spec that’s coming out, and that may run as an overlay on top of an Ethernet underlay. There needs to be some firm standards there because today, scale-up is frankly all proprietary NV Link. And we’re encouraged by—just like we worked hard to found the Ultra Ethernet Consortium as a member for some of the back-end Ethernet, and the migration from InfiniBand to Ethernet is literally happening in 3 to 5 years. We expect the same phenomenon on scale-up,” Ullal said. “The rise in Agentic AI ensures any-to-any conversations with bidirectional bandwidth utilization. Such AI agents are pushing the envelope of LAN and WAN traffic patterns in the enterprise,” Ullal said. Work to do on VeloCloud integration The recent acquisition of VeloCloud was also a hot topic of the second quarter results that included the introduction of former Cisco exec and industry veteran Todd Nightingale, as its newly appointed President & COO.  “It’s only been a month, but I can’t tell you how impressed I am with the passion and focus of the team, the trust that Arista customers have in the technology and the enormous opportunity we have ahead of us in data center, AI, and in the campus,” Nightingale said. “VeloCloud’s secure AI optimized WAN portfolio offers seamless application-aware solutions to connect customer branch sites, complementing Arista’s leading spines in the data center and campus,” Ullal said.  “In a classic leaf-spine atomic identifier, we are enabling multipathing, encryption, in-band network telemetry, segmentation, application identification, and traffic engineering across distributed enterprise sites. We are so excited to fill this missing void in our distributed enterprise puzzle to bring that holistic branch solution.” “We also intend to work closely with best-of-breed security partners to enable SASE overlays. Please do note that VeloCloud is not

The remaining tips were cited by roughly two-thirds of the enterprises. Tip number three is to look especially at applications whose users are widely dispersed. And by “widely” here, they mean on different continents, not just different neighborhoods. The reason is that quality of experience and even availability can be compromised when work has to transit a lot of networks just to get to where it’s processed. This can lead to user dissatisfaction, and dispersing resources closer to the users may be the only solution. If an enterprise doesn’t already have their own data center located close to each user concentration, chances are that putting a new hosting point in themselves couldn’t achieve reasonable economy of scale in capex, power and cooling, and operations costs. The cloud would be cheaper. A qualifying comment here is to take great care in evaluating the real impact of dispersion of application users. In some cases, there may not be enough of a difference in QoE or availability to require dispersing hosting points, and in fact it may be that where the application is hosted isn’t even the problem. “The cloud may look like the easy way out,” one enterprise said, “but it may not be the economical way.” See where your QoE issues really lie before you go to the cloud’s distributed hosting to fix them. Tip four is to examine the user-to-application interaction model carefully, to see if there’s a large non-transactional component. Mission-critical business systems, and business core databases, are almost always in the data center. The stuff that changes them are the transactions that add, update, and delete records. If an application’s user interaction is tightly coupled to the creation of transactions, then its processing is tied to those data center resources. That makes it harder to move the user-interface

The CFO emphasized that SoftBank remains committed to its original target of $346 billion (JPY 500 billion) over 4 years for the Stargate project, noting that major sites have been selected in the US and preparations are taking place simultaneously across multiple fronts. Requests for comment to Stargate partners Nvidia, OpenAI, and Oracle remain unanswered. Infrastructure reality check for CIOs These challenges offer important lessons for enterprise IT leaders facing similar AI infrastructure decisions. Sanchit Vir Gogia, chief analyst and CEO at Greyhound Research, said that Goto’s confirmation of delays “reflects a challenge CIOs see repeatedly” in partner onboarding delays, service activation slips, and revised delivery commitments from cloud and datacenter providers. Oishi Mazumder, senior analyst at Everest Group, noted that “SoftBank’s Stargate delays show that AI infrastructure is not constrained by compute or capital, but by land, energy, and stakeholder alignment.” The analyst emphasized that CIOs must treat AI infrastructure “as a cross-functional transformation, not an IT upgrade, demanding long-term, ecosystem-wide planning.” “Scaling AI infrastructure depends less on the technical readiness of servers or GPUs and more on the orchestration of distributed stakeholders — utilities, regulators, construction partners, hardware suppliers, and service providers — each with their own cadence and constraints,” Gogia said.

Across the United States, states are rolling out a wave of new tax incentives aimed squarely at attracting data centers, one of the country’s fastest-growing industries. Once clustered in only a handful of industry-friendly regions, today’s data-center boom is rapidly spreading, pushed along by profound shifts in federal policy, surging demand for artificial intelligence, and the drive toward digital transformation across every sector of the economy. Nowhere is this transformation more visible than in the intensifying state-by-state competition to land massive infrastructure investments, advanced technology jobs, and the alluring prospect of long-term economic growth. The past year alone has seen a record number of states introducing or expanding incentives for data centers, from tax credits to expedited permitting, reflecting a new era of proactive, tech-focused economic development policy. Behind these moves, federal initiatives and funding packages underscore the essential role of digital infrastructure as a national priority, encouraging states to lower barriers for data center construction and operation. As states watch their neighbors reap direct investment and job creation benefits, a real “domino effect” emerges: one state’s success becomes another’s blueprint, heightening the pressure and urgency to compete. Yet, this wave of incentives also exposes deeper questions about the local impact, community costs, and the evolving relationship between public policy and the tech industry. From federal levels to town halls, there are notable shifts in both opportunities and challenges shaping the landscape of digital infrastructure advancement. Industry Drivers: the Federal Push and Growth of AI The past year has witnessed a profound federal policy shift aimed squarely at accelerating U.S. digital infrastructure, especially for data centers in direct response both to the explosive growth of artificial intelligence and to intensifying international competition. In July 2025, the administration unveiled “America’s AI Action Plan,” accompanied by multiple executive orders that collectively redefined