Author: Deepak Jain

  • Pennsylvania’s $517M Tax-Break Vote Moves AI Data Center Leverage to Local Zoning

    Pennsylvania’s $517M Tax-Break Vote Moves AI Data Center Leverage to Local Zoning

    TL;DR · 30-second read

    The Short Version

    Pennsylvania’s state House voted 201 to 1 to let towns stop considering new data centers for six months while they write rules for them. Data centers are the huge warehouses of computers that run artificial intelligence and online services.

    Lawmakers also voted to end a tax break for these buildings that was expected to cost the state $517 million a year by 2030. That break was the reward in the governor’s plan to get builders to promise jobs, clean power and water limits.

    If the break disappears, local town boards become the place these projects are decided. The state Senate still has to act.

    With the June 30 state budget deadline days away, the Pennsylvania House passed three bills on data center development on June 24 and 25, 2026, the Pennsylvania Capital-Star reported. House Bill 2496 lets municipalities pause consideration of data center applications for 180 days and passed 201-1. House Bill 2650 writes Gov. Josh Shapiro’s Governor’s Responsible Infrastructure Development (GRID) Standards into law and passed 134-68. House Bill 2198 repeals the state’s 6% sales and use tax exemption on data center computer equipment and passed 197-5.

    The tax exemption is projected to cost the commonwealth $517 million a year by 2030. On the afternoon of the House vote, the Senate adopted the repeal language as an amendment to a separate tax code bill. The other two bills now sit with the Republican-led Senate, which could send them to Shapiro or leave them unconsidered.

    Executive Summary

    Pennsylvania’s House has moved from courting large data centers to regulating them. In two days it voted for a local construction-review pause, a certification regime that ties tax benefits to disclosure and job commitments, and the repeal of the tax benefit that regime depends on.

    The votes point in different directions. The GRID Standards are voluntary. Developers who meet targets on in-state clean energy, investment, prevailing-wage construction jobs and permanent jobs would get a sales tax exemption and faster permits. The repeal bill, which drew a 197-5 vote and quick Senate action, would eliminate that exemption. That leaves the standards with little to offer beyond permitting speed.

    The measure with the broadest support was the local pause, which passed 201-1. For developers of AI-scale facilities in one of the country’s most active data center markets, the practical gate is moving from state tax terms to municipal land-use calendars. That shift is not yet final, because the Senate has not acted on the pause or the GRID bill.

    The Incentive Is Going Before the Standards Arrive

    Pennsylvania’s 6% sales and use tax exemption on data center computer equipment is about five years old. House Environmental Resources Committee Chairperson Greg Vitali, who sponsored the repeal, said it was passed “during an era when lawmakers were trying to attract them to the commonwealth.” It is now projected to cost $517 million a year by 2030. For a facility filled with servers, networking gear and storage, a sales tax on equipment is a real line item. Across a whole state’s pipeline, it has become a budget question.

    The repeal has the most momentum of the three bills. It passed the House 197-5, and the Senate adopted the same language as an amendment to another tax code bill within hours. That creates a sequencing problem. HB 2650, passed a day earlier, is built on the exemption: it would make the tax break contingent on certification from the Department of Revenue. Rep. Craig Williams (R-Delaware) put it bluntly: “Like Lucy holding the football for Charlie Brown, we’re going to pull that football away.” Repeal supporters such as Rep. Kyle Mullins (D-Lackawanna) described the industry as “no longer an emerging sector looking for a foothold.” In their view the subsidy has simply outlived its purpose.

    Why Leverage Shifts From Harrisburg to the Township Board

    The GRID Standards are designed as a trade. A developer discloses its ownership structure, energy procurement strategy, water usage, community engagement and environmental plans, and meets investment and job floors. In return it gets the sales tax exemption and expedited permitting. Remove the exemption and the state’s offer shrinks to faster permits. That is a thinner inducement for a program that is voluntary. Critics were already skeptical that the largest companies would participate even with the tax benefit.

    The tool that drew near-unanimous support is local. Under HB 2496, a municipality could stop considering data center applications for 180 days and use that time to adopt, amend or repeal the parts of its land use ordinance that cover data centers. Any application received during the pause would be treated as received the day after the pause ends. That detail matters. The pause does not reject projects. It holds them until new rules exist, so a proposal filed mid-pause would be judged against whatever ordinance the township writes. Rep. Paul Friel (D-Chester) noted that many rural municipalities have no zoning ordinance at all, and others lack provisions for facilities of this scale.

    The people most affected are data center developers and their site-selection teams. Their schedule risk would be set by municipal meeting calendars rather than state tax terms. Landowners holding land options are also exposed, as are the utilities and power suppliers whose planning depends on when large new loads actually arrive. One borough in Lackawanna County faces 16 data center applications, according to Mullins, which shows how concentrated these decisions can become. There is an important caveat. If the repeal becomes law and the pause bill stalls in the Senate, Pennsylvania would have neither the state incentive nor the new local pause. Projects would proceed under existing ordinances, which Friel says many townships lack.

    What the GRID Terms Say About Power and Scale

    The certification requirements go beyond jobs. Developers would need to increase their use of clean energy from Pennsylvania sources step by step, reaching 32% by 2035. That makes the standards a power procurement rule, not just an economic development one. The other floors are:

    • at least $250 million in cumulative new investment;
    • at least 200 prevailing-wage construction jobs;
    • 50 permanent jobs paying at least 125% of the statewide average wage within four years;
    • at least $1.5 million a year in compensation to full-time employees after that.

    Energy procurement and water usage both appear on the application. That reflects the pressures lawmakers cited: large AI facilities draw heavily on water resources, energy infrastructure and public services. Pennsylvania sits inside PJM Interconnection, the regional grid operator for much of the mid-Atlantic and Midwest, where large new loads weigh directly on grid planning. The state’s most visible project shows how developers are responding to power constraints. The former Homer City Generating Station in Indiana County, once Pennsylvania’s largest coal-burning power plant, is being turned into a natural gas-powered data center campus, and construction was under way as of mid-May.

    Testing Both Sides’ Claims

    Williams argued that companies such as Google and the parent company of X “don’t care” about the sales tax exemption and would forgo it rather than accept conditions. The $517 million projection suggests the break is worth a lot in total, so it is hardly trivial. But total cost does not show that the exemption decides where any particular facility goes, and the debate did not include a developer saying either way.

    Repeal supporters argue the break is “not needed,” and the volume of applications supports the claim that demand exists without it. Current filings, though, do not show what happens to marginal projects once the benefit is gone. Neither does the phrase “speculative gold rush,” which one legislator used, tell us how many of the dozens of hyperscale applications have tenants, power and financing behind them. A 180-day pause may be the mechanism that sorts firm projects from speculative land positions.

    Background

    Pennsylvania spent much of the past half-decade trying to attract data centers, including through a sales and use tax exemption on their computer equipment. The race to build artificial intelligence capacity has since produced hyperscale proposals across the commonwealth, and local officials are now weighing dozens of applications. Some of the most prominent projects repurpose power-plant sites, such as the former Homer City Generating Station, which is being converted from coal to a gas-powered data center campus.

    Gov. Josh Shapiro outlined his approach to data center development in his February 2026 budget address and released details in May as the GRID Standards. The state is part of PJM Interconnection, the regional grid operator where the arrival of large new electricity loads has become a central planning concern. That places Pennsylvania’s policy choices at the intersection of land use, tax policy and power supply.

    Sources

    Source: Pa. House passes data center ‘pause’ along with Shapiro’s plan for ‘responsible’ development (Pennsylvania Capital-Star), covering House passage of HB 2496, HB 2650 and HB 2198 on data center regulation and tax policy.

  • Five Eyes Warn: AI Is Reshaping Cyber Risk, Act Now

    Five Eyes Warn: AI Is Reshaping Cyber Risk, Act Now

    The cybersecurity agencies of the Five Eyes intelligence alliance — the United States, United Kingdom, Canada, Australia, and New Zealand — issued a joint statement on AI-related shifts in cybersecurity risk, telling organizational leaders to act now rather than wait for guidance to mature.

    The statement, surfaced through the Inside Privacy legal publication on 25 June 2026, is directed at boards and executives across critical infrastructure and enterprise sectors rather than at technical staff alone.

    Executive Summary

    Joint Five Eyes statements are relatively rare and typically signal that member agencies see a risk landscape shifting faster than existing guidance and procurement cycles can absorb. In this case, the subject is artificial intelligence — both as a capability defenders can deploy and as a set of systems attackers can target or abuse.

    The act now framing is the notable editorial choice. Rather than a technical bulletin aimed at security operations centers, the statement targets organizational leaders, implying that governance, procurement, and risk-tolerance decisions — not just tooling — are what member agencies believe are lagging.

    For infrastructure operators, cloud tenants, and the vendors supplying them, the message is that AI-related cybersecurity risk is now a board-level topic in five major English-speaking economies simultaneously, which tends to precede regulatory attention and customer contract changes.

    Why A Joint Statement, And Why Now

    The Five Eyes is a signals-intelligence sharing arrangement dating to the postwar UKUSA Agreement. Its civilian cybersecurity arms — CISA in the United States, the NCSC in the United Kingdom, the CCCS in Canada, the ASD’s ACSC in Australia, and New Zealand’s NCSC — have increasingly co-signed technical advisories over the past several years. A joint statement addressed to leadership, rather than a technical advisory addressed to defenders, suggests the agencies see the gap as one of executive urgency and organizational readiness rather than missing detection signatures.

    The phrasing shifts in cybersecurity risks is deliberately broad. It can cover attacker use of large language models for phishing and social engineering, model and data-pipeline security within enterprises adopting AI, exposure of sensitive data through third-party AI services, and the emerging attack surface of AI-enabled software supply chains. Without the underlying document text, it is not possible to say which of these the agencies weight most heavily.

    What Changes For Infrastructure Buyers

    For operators of data centers, networks, and cloud platforms, a coordinated Five Eyes push tends to translate into three practical pressures within twelve to eighteen months: customer questionnaires expand to include AI governance and model-security controls; regulated customers in finance, health, and government begin requiring contractual assurances about how AI features process their data; and insurance underwriters recalibrate cyber policies to reflect AI-related exposure. Vendors that can point to concrete controls — data segregation, model access logging, red-team results — will have an easier renewal cycle than those still describing intent.

    The economics are not neutral. Meeting a rising bar on AI security controls favors larger providers with dedicated security engineering capacity and disadvantages smaller vendors that ship AI features by wrapping third-party APIs. That concentration effect is a recurring pattern whenever cybersecurity expectations step up, and it deserves scrutiny on its own terms rather than being treated as an unambiguous good.

    Reading The Statement Carefully

    A leadership-level act now statement is useful precisely because it is short and non-technical, but that brevity is also its limitation. Boards asked to act now reasonably want to know: act on what, measured how, and against what threshold. Without accompanying technical annexes or a maturity model, well-intentioned organizations can respond with procurement activity — buying tools labeled AI-secure — that does not change their actual risk posture.

    It is also fair to ask whether coordinated agency messaging is the most effective channel. The Five Eyes agencies bring credibility and reach, but their remit is advisory in most member countries; the operative levers on organizational behavior remain domestic regulators, sector supervisors, and, increasingly, insurers. A statement of this kind is best read as a signal that those levers are likely to move, not as a substitute for them.

    Background

    The Five Eyes alliance traces to the 1946 UKUSA Agreement on signals-intelligence sharing among the United States, United Kingdom, Canada, Australia, and New Zealand. Its civilian cybersecurity agencies have progressively taken on a public advisory role, co-publishing technical advisories on ransomware, state-linked intrusion sets, and secure-by-design software practices.

    Coordinated statements on artificial intelligence sit at the intersection of two trends: the rapid enterprise adoption of generative AI since 2023, and a broader policy shift toward holding software and service providers — not only end users — accountable for the security properties of what they ship.

    Source: Five Eyes Cybersecurity Agencies Issue Statement Regarding AI-Related Shifts in Cybersecurity Risks, Urging Organizational Leaders to “Act Now” – Inside Privacy — legal-industry summary of a joint Five Eyes cybersecurity statement on AI risk directed at organizational leaders.

  • FERC Aims to Cut Data Center Grid Queues and Electricity Bills: What It Means

    FERC Aims to Cut Data Center Grid Queues and Electricity Bills: What It Means

    IEEE Spectrum reported on June 25, 2026, that the Federal Energy Regulatory Commission (FERC) — the U.S. agency that oversees the interstate power grid and wholesale electricity markets — aims to cut the queues that data centers face when seeking grid connections, while also containing electricity bills. The syndicated item carries only the headline, so the specific mechanism, docket, and timeline are not detailed in the material available here.

    The framing itself is significant: the regulator is treating slow grid interconnection and rising consumer power costs as a single, linked problem — the two pressures the AI data center boom has placed on the U.S. electric system.

    Executive Summary

    According to the report, FERC is moving to shorten the waits that large new loads — chiefly AI data centers — endure before they can connect to the grid, and to do so in a way that limits the impact on ordinary electricity bills. Interconnection is the process by which a new generator or major customer is studied, assigned any needed grid-upgrade costs, and physically wired into the transmission system; the backlog of these requests is widely regarded as one of the tightest bottlenecks on U.S. data center growth.

    Why it matters: hyperscale operators can erect a building in 18 to 24 months, but securing hundreds of megawatts of firm grid power can take far longer, and utilities in several regions have quoted multi-year waits. At the same time, household and business electricity prices have become politically charged in data-center-heavy regions, with debates over how much of the grid buildout ordinary ratepayers should fund. A federal move that credibly addresses both — speed and cost — would be the single biggest regulatory lever on how fast AI infrastructure can actually energize.

    What is and is not substantiated: the available source confirms the regulator’s stated aim but not the instrument. Whether this is a formal rulemaking, a policy statement, or guidance to grid operators — and whether it is binding — cannot be determined from the headline alone, and readers should weight it accordingly until the underlying FERC documents are public.

    Why the Interconnection Queue Is the Real Bottleneck

    Every large project that wants to plug into the high-voltage grid — a solar farm, a gas plant, or increasingly a gigawatt-scale data center campus — must file an interconnection request and wait for engineering studies that determine what upgrades the grid needs and who pays for them. By the end of 2023, Lawrence Berkeley National Laboratory counted roughly 2,600 gigawatts of generation and storage capacity waiting in U.S. queues — more than double the nation’s entire installed generating fleet — with typical waits stretching toward five years from request to operation.

    Data centers sit on the demand side of this equation, and large-load interconnection has historically been even less standardized than the generator process, handled utility by utility and state by state. For AI operators, the queue — not chips, land, or capital — is frequently the schedule-defining constraint. That is why a federal regulator signaling it wants to compress these timelines matters more to data center delivery dates than most technology announcements.

    Two Goals in Tension: Faster Hookups and Lower Bills

    Cutting queues and cutting bills pull in different directions, and the report’s pairing of them is the most analytically interesting element. Connecting multi-hundred-megawatt loads quickly often requires transmission upgrades whose costs, under traditional utility ratemaking, are spread across all customers. Consumer advocates in several data-center-heavy states have argued that households are subsidizing the grid expansion that serves hyperscale computing; utilities and data center operators counter that large, steady loads can spread fixed grid costs over more sales and put downward pressure on rates.

    Both claims can be true depending on how cost allocation is structured — which is precisely the kind of question FERC decides. Mechanisms observers have debated in recent years include dedicated large-load rate classes, requirements that data centers fund their own upgrades or bring their own generation, and co-location arrangements that place computing directly at power plants. Which of these, if any, the regulator is now advancing is not specified in the available source.

    What a Federal Regulator Can — and Cannot — Fix

    FERC has a track record here: its Order 2023 overhauled the generator interconnection process, replacing first-come-first-served study lines with clustered, first-ready-first-served batches, backed by deposits and readiness requirements to flush speculative projects from the queue. Extending comparable discipline to large loads would be a logical next step, and FERC has also been drawn into the co-location debate through disputes over data centers sited at existing power plants.

    But the agency’s jurisdiction has hard edges. States control retail rates, generation siting, and most permitting; regional grid operators run their own study processes; and no order can conjure the transformers, turbines, and skilled crews that are in genuinely short supply worldwide. A FERC action can remove procedural delay — often years of it — but the physical buildout still moves at the pace of supply chains and state approvals. Expectations should be calibrated to that split.

    Winners, Losers, and What to Watch

    If queue reform for large loads materializes and works, the clearest beneficiaries are hyperscalers and data center developers with projects stalled behind study backlogs, along with the transmission engineering firms and equipment suppliers that would see demand pulled forward. Utilities face a mixed outcome: faster load growth boosts their invested capital base, but tighter federal timelines and cost-assignment rules constrain how they manage it. Generation developers could gain if load and supply requests are studied more coherently together.

    The unresolved variable is the ratepayer. If the regulator pairs faster interconnection with cost rules that make large loads bear the upgrades they cause, the political friction around data center power could ease; if speed comes without that discipline, bill impacts could intensify the local backlash that has already slowed projects in several markets. The details — still unpublished in the material available here — will determine which scenario unfolds.

    Background

    FERC is the century-old independent agency that governs the U.S. interstate grid, and interconnection reform has been its defining workstream of the 2020s. After two decades of essentially flat electricity demand, AI data centers, manufacturing, and electrification pushed load growth back onto utility planning maps around 2023–2024, colliding with queue backlogs that Lawrence Berkeley National Laboratory measured at roughly 2,600 gigawatts of waiting capacity by the end of 2023. Order 2023 tackled the generator side of the problem; large loads — the data centers themselves — remained governed by a patchwork of utility and state processes.

    Through 2024 and 2025, disputes over co-locating data centers at power plants and over who pays for grid expansion made large-load policy one of the most watched dockets in U.S. energy. The June 2026 report places FERC’s next move squarely in that lineage: an attempt to standardize and speed how the grid absorbs its biggest new customers without letting the cost land on everyone else’s bill.

    Source: U.S. Regulator Aims to Cut Data Center Queues and Electricity Bills — IEEE Spectrum report, June 25, 2026, on FERC’s effort to speed data center grid interconnection while containing consumer electricity costs.

  • Why Liquid Cooling Is Non-Negotiable for High-Density AI Factories

    Why Liquid Cooling Is Non-Negotiable for High-Density AI Factories

    Data Center Dynamics published an analysis on 25 June 2026 contrasting the cooling demands of AI factories — facilities purpose-built for dense GPU training and inference clusters — with those of conventional cloud data centers, arguing that liquid cooling is now essential for high-density AI workloads rather than an optional upgrade.

    The piece lands amid an industry-wide retooling: operators worldwide are redesigning halls, mechanical plants, and supply chains around direct-to-chip and other liquid cooling approaches as accelerated computing outgrows the air-cooled designs that served the cloud era.

    Executive Summary

    The core claim is straightforward: the data center designs that carried the cloud computing era are hitting a physical ceiling. Conventional cloud halls were engineered around air cooling — moving chilled air through racks drawing power in the single-digit-to-low-double-digit kilowatt range. AI training clusters concentrate far more power in each rack, because modern GPU systems pack accelerators tightly together to keep them on fast, short interconnects. At those densities, air simply cannot carry heat away fast enough, and liquid — which is far denser and holds vastly more heat per unit volume than air — becomes the only practical medium.

    Why it matters: cooling is no longer a back-of-house mechanical detail but a gating factor for who can host AI workloads at all. Operators with liquid-ready facilities can court the highest-value tenants; operators with legacy air-cooled halls face expensive retrofits or a narrowing addressable market. For enterprises buying AI capacity, a provider’s cooling architecture is now a proxy for whether it can actually deliver current-generation GPU infrastructure.

    The analysis frames this as a structural divide — ‘AI factory’ versus ‘cloud hall’ — rather than a spectrum, which is a useful lens even if real-world facilities often blend both.

    The Physics Sets the Deadline, Not the Marketing

    Air cooling works by blowing large volumes of conditioned air through servers, and it has a well-understood practical ceiling: as rack power climbs, the airflow, fan energy, and temperature gradients required become unmanageable. Liquid cooling — most commonly direct-to-chip cold plates, where coolant flows across a metal plate bonded to the processor, or immersion, where hardware is submerged in a dielectric (electrically non-conductive) fluid — removes heat at the source with far greater efficiency. This is not a vendor preference; it is thermodynamics. Water-based coolants can absorb on the order of thousands of times more heat per unit volume than air, which is why every leading accelerated-computing platform roadmap now assumes liquid at the high end.

    The important nuance is that the ceiling is not a single number. Well-engineered air systems with hot-aisle containment can stretch surprisingly far, and many inference and enterprise workloads will remain comfortably air-coolable for years. The ‘non-negotiable’ framing applies specifically to dense training clusters, where chips must sit physically close together for interconnect performance. Density is a networking decision as much as a thermal one — and that is precisely why it cannot be relaxed just to make cooling easier.

    Economics: Liquid Costs More Up Front and Less to Run

    Liquid cooling shifts spending from operations to capital. Cold plates, coolant distribution units, manifolds, leak detection, and plumbing add up-front cost and engineering complexity that air systems avoid. In exchange, operators typically get lower fan energy, better power usage effectiveness (PUE — the ratio of total facility power to IT power, where closer to 1.0 is better), and the ability to run warmer coolant loops that reduce or eliminate energy-hungry chillers. Heat captured in liquid at useful temperatures is also far easier to reuse — for district heating or industrial processes — than diffuse warm air.

    The strategic consequence is that cooling architecture now shapes site selection and facility economics together. A liquid-cooled AI factory can put more revenue-generating compute on the same power envelope, which matters enormously when grid connections — not land or capital — are the scarcest input in the industry. That said, buyers should treat sweeping efficiency claims with care: realized PUE depends on climate, design discipline, and utilization, and figures quoted for flagship builds do not automatically transfer to retrofits.

    Winners, Losers, and the Retrofit Question

    The clearest winners are operators and builders that committed early to liquid-ready designs — reinforced floors for heavier racks, space for coolant distribution, higher-capacity power delivery — along with the supply chain behind them: cold-plate and CDU manufacturers, fluid suppliers, and mechanical contractors with liquid experience. Chipmakers benefit too, since liquid cooling removes a constraint on how much power their next generations can draw.

    The harder story is the installed base. Thousands of existing air-cooled halls cannot be casually converted: adding liquid means new piping, floor loading analysis, leak-management protocols, and often a rethink of the entire mechanical plant. Some facilities will be retrofitted profitably, some will serve the still-large market for air-coolable workloads, and some will be stranded relative to AI demand. For colocation providers, the honest question customers should ask is not ‘do you support liquid cooling?’ but ‘how many megawatts of it can you deliver, at what density, and by when?’

    Operational Risk: New Skills, New Failure Modes

    Bringing liquid into the white space introduces failure modes the air-cooled era rarely faced: leaks near live electronics, coolant chemistry maintenance, and the coordination of facility water loops with IT equipment loops. None of these are exotic — mainframes were water-cooled decades ago, and modern systems are engineered with negative-pressure loops and leak detection — but they demand skills that many data center operations teams are still building. Expect certification programs, standardized quick-disconnect fittings, and reference designs to matter as much as raw technology in determining who executes this transition smoothly. The industry’s real constraint may be trained people, not parts.

    Background

    Data center cooling has followed computing density for decades: water-cooled mainframes gave way to air-cooled commodity servers in the client-server and cloud eras, when racks drawing modest power made air the cheap, simple choice. The generative AI boom reversed the trend — modern accelerator systems concentrate unprecedented power in single racks, and leading GPU platform roadmaps now assume liquid cooling at the high end, pulling the entire industry’s mechanical design along with them.

    Data Center Dynamics, the publication behind this analysis, is a long-established trade outlet covering data center design and operations. Its framing of ‘AI factories’ versus conventional cloud facilities echoes terminology popularized by the accelerated-computing industry to describe purpose-built AI infrastructure — a sign of how thoroughly that vocabulary has permeated the sector.

    Source: AI factory cooling vs cloud data centers: Why liquid cooling is essential for high-density AI workloads — a Data Center Dynamics analysis, published 25 June 2026, on why liquid cooling has become a baseline requirement for dense AI infrastructure.

  • Hyperscale Data’s $1.2B, 20-Year AI Data Center Services Deal, Explained

    Hyperscale Data’s $1.2B, 20-Year AI Data Center Services Deal, Explained

    Hyperscale Data has signed a $1.2 billion AI data center services agreement, reported June 25, 2026 via Investing.com. The contract is structured over a 20-year term — an unusually long commitment in an industry where colocation and cloud deals typically run three to ten years.

    The announcement positions the company as a beneficiary of surging demand for AI compute capacity, with a single long-dated services relationship underwriting future campus development.

    Executive Summary

    The headline facts are simple: a $1.2 billion total contract value, a 20-year duration, and AI data center services as the product. Averaged across the term, that works out to roughly $60 million per year — meaningful, recurring revenue for a company of Hyperscale Data’s size, if the contracted volumes materialize as projected.

    Why it matters is the structure, not just the size. AI infrastructure operators increasingly need anchor tenants — customers who commit to capacity years before it is fully built — to justify the enormous capital costs of power, land, and cooling. A 20-year services agreement is a signal to lenders and investors that demand exists beyond the current AI investment cycle. The announcement, as reported, does not name the counterparty or detail the commercial terms, so the durability of that signal depends on specifics the headline does not provide.

    Why Anchor Deals Now Run Decades, Not Years

    Data center economics have always depended on matching long-lived assets to shorter-lived contracts. A campus takes years to permit, power, and build, and the shell and electrical infrastructure depreciate over decades — yet traditional colocation leases (renting space, power, and cooling to a customer’s own equipment) often ran only three to five years. The AI buildout has inverted that mismatch: operators now seek contracts as long as the assets themselves, and customers desperate for scarce GPU-ready capacity are willing to sign them. A 20-year term puts this deal at the far end of that trend, closer to a power purchase agreement or an infrastructure concession than a conventional hosting contract.

    For the operator, the appeal is financing. Lenders and infrastructure investors price projects on contracted cash flow; two decades of committed revenue can unlock construction debt that a merchant (uncontracted) facility could never raise. For the customer, locking in capacity and pricing hedges against a market where AI-grade space and power remain supply-constrained.

    The Neocloud Layer in the AI Stack

    The demand behind deals like this increasingly comes from so-called neoclouds — specialized GPU cloud providers that rent AI compute to enterprises and model developers, sitting between the chip makers and end users. Unlike the hyperscale giants, neoclouds typically do not build their own campuses; they lease capacity from data center operators and fill it with accelerators. That makes them natural anchor tenants for second-tier and emerging operators that cannot land a hyperscaler directly.

    The trade-off is counterparty quality. Hyperscalers carry investment-grade balance sheets; many neoclouds are young companies whose own revenue depends on continued AI demand. A 20-year commitment is only as strong as the customer’s ability to pay in year eight or year fifteen. Without the counterparty’s identity and credit profile — which the reported announcement does not supply — the $1.2 billion figure describes the contract’s ambition more than its guaranteed value.

    Reading a Total Contract Value Honestly

    Total contract value, or TCV, is the standard way these announcements are framed, and it deserves careful reading in every case, from any operator. $1.2 billion over 20 years averages about $60 million annually, but real contracts rarely pay evenly: they typically ramp as capacity is delivered, may include usage-based components, and can carry termination or renegotiation provisions. The material questions are how much of the value is a firm, take-or-pay minimum (payment owed whether or not capacity is used) versus a projection, and what milestones the operator must hit to earn it.

    None of that skepticism is unique to Hyperscale Data — it applies to the entire wave of multibillion-dollar AI capacity announcements across the industry. The pattern to watch, here and elsewhere, is whether contracted revenue converts into financed construction, energized power, and recognized revenue on subsequent earnings reports.

    Background

    Hyperscale Data is a diversified, US-listed holding company that rebranded from Ault Alliance as it repositioned around data centers and AI infrastructure. Like several smaller operators, it is pursuing the AI buildout from outside the ranks of the established wholesale data center giants, which makes long-dated anchor contracts especially consequential for its growth story.

    The market context is a historic capacity crunch: demand for GPU-ready power and space has outrun supply since the generative-AI investment wave began, pushing customers toward earlier and longer commitments and giving emerging operators a route to bankable projects that would have been unattainable in the pre-AI colocation market.

    Source: Hyperscale Data signs $1.2B AI data center services agreement — Investing.com report, June 25, 2026, on the company’s 20-year AI data center services contract.

  • Qualcomm’s Dragonfly Bid: A Third Path in AI Inference Silicon

    Qualcomm’s Dragonfly Bid: A Third Path in AI Inference Silicon

    On June 24, 2026, Qualcomm announced a comprehensive data center roadmap built around a new product family it calls Dragonfly, positioning the portfolio for what the company describes as the agentic AI era — workloads where AI systems act autonomously across chained tasks rather than answering single prompts.

    The announcement marks Qualcomm’s most explicit push yet into data center silicon, a market currently dominated by Nvidia with AMD as the principal challenger.

    Executive Summary

    Qualcomm is best known for smartphone modems and mobile system-on-chip designs. With Dragonfly, the company is signaling that it intends to translate its low-power, inference-oriented engineering heritage into a full data center accelerator roadmap aimed at agentic AI — inference workloads that are longer-running, more memory-intensive, and more sensitive to cost-per-token than the training runs that made Nvidia’s H100 and Blackwell generations famous.

    Why it matters: hyperscalers, sovereign cloud buyers, and neocloud operators have been vocal about wanting a viable third source for AI accelerators to ease supply constraints and pricing power. A credible Qualcomm entry, alongside AMD’s Instinct line and in-house silicon from AWS, Google, and Microsoft, would reshape purchasing leverage across the data center stack. Whether Dragonfly clears that bar depends on details the June 24 release does not fully disclose.

    For infrastructure operators, the immediate question is not whether Qualcomm can build competitive silicon — it has a strong NPU (neural processing unit) track record in mobile — but whether it can deliver the software stack, systems integration, and multi-year supply commitments that hyperscale procurement demands.

    Why Inference, and Why Now

    The AI silicon market has bifurcated. Training the largest models remains a specialized, capital-intensive workload where Nvidia’s CUDA software moat and networking assets (NVLink, InfiniBand via Mellanox) give it a durable lead. Inference — actually running trained models to serve users — is a larger and faster-growing spend line, and it is more fragmented technically. Different model sizes, latency targets, and cost envelopes favor different silicon architectures. Qualcomm’s positioning of Dragonfly around agentic inference is a rational reading of where the addressable market is opening up: agentic workloads chain many inference calls together, making cost-per-token and energy-per-token the metrics that matter most to operators.

    Qualcomm’s mobile heritage is genuinely relevant here. The company has shipped billions of NPU-equipped chips optimized for running neural networks under tight power budgets — a discipline the data center now needs as grid capacity, not GPU supply, becomes the binding constraint on AI buildouts.

    The Third-Source Thesis

    Buyers of AI infrastructure have made no secret of wanting alternatives to Nvidia. AMD has partially filled that role with its Instinct MI300 and successor accelerators, and hyperscalers have invested heavily in custom silicon — AWS Trainium and Inferentia, Google TPU, Microsoft Maia. Qualcomm’s Dragonfly enters a field that is crowded but still supply-constrained, and where any credible merchant-silicon alternative can command attention simply by existing. The commercial question is whether Qualcomm can win design wins at hyperscalers that already have in-house programs, or whether its natural customers are tier-two clouds, sovereign AI initiatives, and enterprise on-premises deployments where a turnkey vendor stack is more valuable than bespoke silicon.

    The competitive risk cuts both ways. If Dragonfly ships on schedule with competitive performance-per-watt and a workable software stack, it pressures Nvidia’s pricing on inference SKUs and validates AMD’s playbook. If it slips or underdelivers on software, it joins a long list of ambitious accelerator programs — from Intel’s Gaudi to various startups — that failed to convert silicon competence into share.

    Software Is Where Accelerator Roadmaps Live or Die

    The unspoken subject of any new AI silicon announcement is the software stack. Nvidia’s advantage is not primarily transistors; it is CUDA, cuDNN, TensorRT, and a decade of framework integration that makes developers productive on day one. Any Dragonfly evaluation by a serious buyer will focus on how well Qualcomm supports PyTorch, vLLM, TensorRT-equivalent inference runtimes, and increasingly the open standards like OpenAI-compatible APIs and the emerging agentic frameworks. The June 24 release frames Dragonfly as a portfolio and roadmap rather than a single product, which suggests Qualcomm is aware that ecosystem depth matters as much as peak throughput numbers.

    For infrastructure operators evaluating Dragonfly, the practical checklist is well-established: what models run out of the box, what quantization formats are supported, how does the compiler handle novel architectures, and what is the update cadence when a new model family lands. None of these are answered in the announcement itself.

    Power, Density, and the Data Center Fit

    Modern AI accelerators are increasingly constrained by rack-level power and cooling rather than chip-level cost. A meaningful Dragonfly value proposition would show up in performance-per-watt at realistic inference batch sizes, and in the thermal envelope that determines whether the parts drop into air-cooled facilities or require liquid cooling retrofits. Qualcomm’s mobile pedigree suggests an efficiency-first design philosophy, which aligns with where the industry’s power problem is heading, but the announcement does not disclose the numbers that would let operators model total cost of ownership.

    Background

    Qualcomm built its business on wireless modems and Snapdragon system-on-chip designs that power much of the global smartphone market. Its neural processing units have delivered on-device AI in mobile phones for years, giving the company deep expertise in low-power inference. A prior effort to enter the server market with the Centriq Arm CPU in the late 2010s was ultimately discontinued, making Dragonfly the company’s most substantial data center push since.

    The AI accelerator market took its current shape after 2022, when generative AI demand made Nvidia’s data center GPUs the scarcest resource in enterprise computing. AMD’s Instinct MI300 series became the primary merchant-silicon alternative, while AWS, Google, and Microsoft accelerated in-house silicon programs. Buyers across hyperscale, sovereign cloud, and enterprise segments have consistently signaled that a credible third source would be welcome — the question Dragonfly will answer over the coming quarters is whether Qualcomm can be that source.

    Source: Qualcomm Unveils Comprehensive Data Center Roadmap for the Agentic AI Era with New Qualcomm Dragonfly Portfolio — Qualcomm’s June 24, 2026 announcement of its Dragonfly data center product family for agentic AI inference.

  • Microsoft’s Mount Pleasant AI Campus Reaches Full Operation in Wisconsin

    Microsoft’s Mount Pleasant AI Campus Reaches Full Operation in Wisconsin

    Microsoft’s AI data center campus in Mount Pleasant, Wisconsin is now fully operational, according to a June 24, 2026 report from Data Center Knowledge. The milestone marks the completion of the commissioning phase for one of the most closely watched hyperscale AI sites in the United States — a campus Microsoft has publicly positioned as a flagship of its AI infrastructure program since announcing a $3.3 billion investment there in May 2024.

    Executive Summary

    The report that Microsoft’s Wisconsin campus has gone fully operational converts years of announcements into working capacity. “Fully operational” in hyperscale terms means the facility has moved past construction and phased commissioning — the staged process of energizing electrical systems, validating cooling loops, and bringing compute halls online rack by rack — into steady-state production service.

    It matters for three reasons. First, the site is a bellwether: Microsoft branded its Mount Pleasant build “Fairwater” and described it as among the most powerful AI data centers in the world, purpose-built for training large AI models on massive GPU clusters. Second, the location carries unusual economic symbolism, occupying land originally assembled for Foxconn’s largely unrealized 2017 manufacturing project. Third, it is a data point on whether the AI capital-expenditure cycle is delivering finished, revenue-generating infrastructure on schedule — a question investors and utilities are asking with increasing urgency.

    One caveat readers should hold onto: the source is a headline-level trade report. Specific operational figures — megawatts energized, GPU counts in service, final headcount — are not independently confirmed in it, and we flag below what remains unverified.

    From Foxconn’s Ghost Site to an AI Flagship

    Few parcels of American industrial land carry as much narrative weight as Mount Pleasant. In 2017, Foxconn pledged a $10 billion LCD manufacturing campus there with talk of up to 13,000 jobs; the project was dramatically scaled back, leaving the village and Racine County with prepared land, water infrastructure, and unmet expectations. Microsoft’s arrival in 2023–2024 — culminating in the $3.3 billion commitment announced in May 2024 — recast the site as AI infrastructure rather than manufacturing.

    Full operation closes that redemption arc, at least physically. For local officials who financed roads, water mains, and land assembly for Foxconn, a running hyperscale campus finally puts heavy, long-lived capital on the tax rolls. It is worth being precise about what changed, though: a data center campus employs far fewer people per dollar of investment than the factory once promised. The win for the region is tax base, grid and fiber investment, and anchor-tenant credibility — not mass employment.

    What “Fully Operational” Actually Means at Hyperscale

    Hyperscale campuses do not flip on like a light switch. They are commissioned in phases: substations and switchgear are energized, cooling plants are load-tested, and data halls are accepted one at a time, often over 12 to 24 months. A “fully operational” declaration means the last planned phase of the current build has passed acceptance and is carrying production workloads — in this case, most likely AI training and inference for Microsoft’s own models and its Azure cloud customers.

    Microsoft has said the Wisconsin facility was designed around dense GPU clusters — the specialized processors that do the mathematical heavy lifting of AI — networked into effectively one giant computer for training large models. That design choice matters commercially: a training-oriented campus is measured less by how many customers it hosts and more by how fast it lets its owner iterate on frontier models. Full operation here is capacity Microsoft has been publicly hungry for throughout the AI demand surge.

    Power and Cooling: The Real Constraints on the AI Buildout

    The binding constraints on AI infrastructure are no longer chips alone but electricity and heat. Microsoft has described the Mount Pleasant design as using closed-loop liquid cooling — water is filled once and continuously recirculated to carry heat away from densely packed GPUs, rather than being evaporated and replaced as in traditional cooling towers. If it performs as described, that design substantially reduces ongoing water draw, a sensitive issue in any community hosting a large data center near the Lake Michigan basin.

    Electricity is the harder question. Facilities of this class draw utility-scale power measured in the hundreds of megawatts, and Wisconsin utilities have been planning generation and transmission additions with data center demand explicitly in view. Who pays for that grid expansion — hyperscalers through special tariffs, or ratepayers broadly — is one of the live policy debates of the AI era, in Wisconsin as elsewhere. A fully operational campus moves that debate from the hypothetical to the measurable: actual load data now exists, even if it is not yet public.

    A Bellwether for the AI Capex Cycle

    The AI buildout is one of the largest private capital deployments in history, and skeptics reasonably ask whether announced projects become working assets or stall in permitting, power queues, and supply chains. Mount Pleasant going fully operational is evidence for the “it’s getting built” side of the ledger — a site that went from announcement to full operation in roughly two years, and which Microsoft subsequently doubled down on with a second announced facility that pushed its stated Wisconsin commitment past $7 billion.

    For competitors and suppliers, the milestone sharpens the map. Rivals racing to stand up comparable training capacity now face a Microsoft with another flagship online. For the ecosystem of electrical contractors, cooling vendors, and fiber providers, a completed phase means crews and supply chains roll to the next site — including, presumably, the second Wisconsin building. And for enterprise buyers of AI services, more training capacity upstream generally translates, with a lag, into more capable models and more available GPU capacity downstream.

    Background

    Microsoft is one of the world’s largest cloud and AI providers, and since 2023 it has led one of the largest infrastructure buildouts in corporate history to supply computing capacity for AI model training and services delivered through its Azure cloud. Data centers — warehouse-scale buildings packed with servers, specialized AI processors, power distribution, and cooling — are the physical foundation of that effort, and Microsoft has announced multibillion-dollar campuses across the United States and abroad.

    The Mount Pleasant, Wisconsin site carries particular history. It was assembled for Foxconn’s heavily subsidized 2017 manufacturing project, which largely failed to materialize. Microsoft began acquiring land there in 2023, announced a $3.3 billion AI data center investment in May 2024, later unveiled the campus under the “Fairwater” banner as a flagship AI training facility with closed-loop liquid cooling, and announced a second Wisconsin data center that raised its stated commitment in the state above $7 billion. The June 2026 report that the campus is fully operational marks the completion of that first flagship build.

    Source: Microsoft’s Wisconsin AI Data Center Campus Now Fully Operational — Data Center Knowledge, June 24, 2026, reporting that Microsoft’s Mount Pleasant AI campus has completed commissioning and entered full production service.

  • Nvidia’s Hot-Water Cooling Claims Up to 100% Water-Use Reduction for AI Data Centers

    Nvidia’s Hot-Water Cooling Claims Up to 100% Water-Use Reduction for AI Data Centers

    Nvidia has announced a liquid cooling system for AI data centers that circulates water described as running “hotter than a hot tub,” a design the company says can reduce electricity consumption and cut water use by up to 100%. The announcement, reported June 24, 2026 by Tom’s Hardware, targets one of the AI build-out’s most scrutinized side effects: the enormous water and energy appetite of the facilities that host Nvidia’s chips. The same report notes that sustainability challenges remain despite the headline claims.

    Executive Summary

    Nvidia, the dominant supplier of AI accelerators, is moving further down the stack — from chips and rack-scale systems into the cooling infrastructure that keeps them running. The newly announced system uses hot-water liquid cooling: instead of chilling coolant to low temperatures before it reaches the hardware, the loop runs deliberately warm, hotter than the roughly 40°C (104°F) at which a typical hot tub is kept, which is the comparison Nvidia’s framing invites.

    Why does that matter? Warmer coolant is the key that unlocks both of the claimed benefits. If the water returning from the chips is already hot, a facility can often reject that heat to the outside air with simple dry coolers rather than energy-hungry chillers — cutting electricity — and without evaporative cooling towers, which consume water by design. That is the engineering logic behind the “up to 100%” water-reduction figure. The claim is significant if it holds up at scale, but as reported it is a vendor claim with important qualifiers, and the source coverage itself flags that sustainability challenges remain.

    Water Is Becoming AI’s Second Resource Fight

    Electricity has dominated the AI infrastructure debate, but water is close behind. Many conventional data centers cool themselves with evaporative systems: they literally evaporate water to carry heat away, because evaporation is cheap and effective. As hyperscale and AI campuses have multiplied, their water draw has become a flashpoint in drought-prone regions and a recurring obstacle in permitting and community relations.

    Nvidia has a direct commercial stake in defusing that fight. Its rack-scale AI systems concentrate so much heat that air cooling is no longer practical, which already pushed the industry toward liquid cooling. If the company can also credibly claim its reference designs eliminate on-site cooling water, it removes an objection that slows down the very data center projects that buy its chips. In that sense this is as much a market-access play as an engineering one.

    The Counterintuitive Physics of Cooling with Hot Water

    “Hot-water cooling” sounds like a contradiction, but it rests on straightforward thermodynamics. A chip does not need cold coolant; it needs coolant that is cooler than the chip and flowing fast enough to carry heat away. Liquid is far denser than air as a heat-transfer medium, so even warm water can hold chip temperatures within limits.

    The payoff comes at the other end of the loop. Cold-water systems need chillers — essentially industrial refrigerators — whose compressors are among the largest energy consumers in a data center. Evaporative towers avoid some of that electricity but spend water instead. A loop that returns water hotter than the outdoor air can shed its heat through dry coolers, closed radiators that use neither compressors nor evaporation. That is the mechanism behind both claims in the announcement: less electricity because chillers shrink or disappear, and less water because nothing is evaporated. Hotter return water is also more useful for heat reuse, such as district heating, though the reporting here does not say whether Nvidia is claiming that benefit.

    Reading the “Up to 100%” Claim Carefully

    “Up to 100%” is a ceiling, not a promise. Real-world results will depend on climate — dry cooling gets harder on very hot days, when some designs fall back on water assist — as well as on facility design and how much of a site’s load actually sits on the new system. The reported claim does not, on its face, distinguish between a best-case new build in a favorable climate and a typical deployment.

    There is also a boundary question. Eliminating on-site cooling water does not eliminate a data center’s water footprint, because the power plants that generate its electricity often consume water themselves. Reduced electricity consumption helps on that front too, but “water-free” at the fence line is not the same as water-free end to end. The source’s own caveat — that sustainability challenges remain — is best read in this light: the announcement addresses a real problem without dissolving it.

    Who Feels This Announcement

    Cooling incumbents and the liquid-cooling supply chain feel it first. When the dominant chip vendor blesses a particular thermal architecture, it tends to become the default for new AI capacity, shaping demand for cold plates, coolant distribution units, and dry coolers, and putting pressure on vendors invested in evaporative or chilled-water designs. Operators, meanwhile, gain a potential permitting and siting advantage: a campus that can credibly promise near-zero cooling-water draw is an easier sell to water-stressed municipalities.

    The open competitive question is whether this arrives as an open reference design others can build on or as another layer of the Nvidia-specified stack. The reporting available here does not say. Either way, buyers should expect warm-water readiness — higher allowable coolant temperatures across IT hardware — to show up in procurement requirements, because the economics above only materialize if the whole rack tolerates the heat.

    Background

    Nvidia is the world’s leading supplier of the GPUs (graphics processing units) that train and run modern AI models, and its data center business has grown into one of the largest in the technology industry. As its systems evolved from individual chips into full pre-integrated racks drawing unprecedented power, the company has taken an increasingly active role in specifying the surrounding infrastructure — power delivery and cooling included — because its hardware roadmap now depends on facilities that can handle the heat.

    Data center cooling has historically split between air cooling, chilled-water systems, and evaporative designs that trade water for electricity. AI’s density has pushed the industry rapidly toward direct liquid cooling, and water consumption has become a headline issue in siting battles. Warm-water liquid cooling — long used in some high-performance computing installations — is the established engineering idea this announcement scales up and brands for the AI era.

    Source: Nvidia announces liquid cooling system that runs ‘hotter than a hot tub’ — promises to reduce electricity consumption and cut water use by up to 100%, but sustainability challenges remain — Tom’s Hardware coverage, June 24, 2026, of Nvidia’s hot-water liquid cooling announcement for AI data centers.

  • OpenAI and Broadcom Unveil LLM-Optimized Inference Chip

    OpenAI and Broadcom Unveil LLM-Optimized Inference Chip

    OpenAI and Broadcom announced an inference chip optimized for large language models (LLMs) — the AI systems behind products like ChatGPT — in a release dated June 24, 2026. The unveiling is the visible next step in the partnership the two companies disclosed in October 2025, under which Broadcom is co-developing and deploying racks of OpenAI-designed accelerators targeting some 10 gigawatts of computing capacity, with deployments slated to begin in the second half of 2026.

    Executive Summary

    The announcement marks OpenAI’s transition from designing custom silicon on paper to unveiling a product: a chip built specifically for inference, the work of running a trained AI model to answer queries, as distinct from the training runs that build the model in the first place. Inference is where the ongoing operating cost of AI lives — every user prompt consumes it — so a chip tuned to OpenAI’s own models attacks the largest recurring line item in the company’s cost structure.

    For Broadcom, the chip validates its custom-accelerator (XPU) business model: rather than selling merchant chips as Nvidia does, Broadcom co-designs silicon to a single customer’s workload and pairs it with its Ethernet networking portfolio. For the broader market, the announcement escalates a race in which nearly every hyperscaler — Google, Amazon, Meta, Microsoft — now fields in-house AI silicon aimed at reducing dependence on Nvidia’s GPUs. What the headline announcement does not yet substantiate, based on the source available, is performance data, manufacturing details, or deployment volumes; we flag those open questions below.

    Why Inference Is the Battleground

    Training a frontier model is a periodic, enormous expense; serving it to hundreds of millions of users is a continuous one. Industry economics increasingly hinge on the cost per generated token — the small units of text an LLM produces — and general-purpose GPUs carry silicon and features that inference of a known model family doesn’t need. A chip co-designed around OpenAI’s own model architectures can, in principle, strip that overhead: right-sized memory bandwidth, dense low-precision math, and interconnects matched to how the models are actually sharded across racks.

    That logic explains why the first unveiled product of the partnership is an inference part rather than a training part. It is the safer engineering bet — inference workloads are more predictable than training — and the faster payback. It also preserves a pragmatic split: OpenAI can keep buying Nvidia and AMD hardware for training frontier models while shifting the high-volume serving fleet onto silicon it controls.

    Broadcom’s Quiet Counter-Model to Nvidia

    Broadcom does not sell a rival to Nvidia’s GPU catalog. Instead it builds custom accelerators — the model proven over roughly a decade with Google’s TPUs — supplying design expertise, chip infrastructure such as serializer/deserializer (SerDes) and packaging technology, and the Ethernet switching that ties accelerators together. The October 2025 agreement made OpenAI the marquee addition to that franchise, with racks scaled entirely on Ethernet rather than Nvidia’s proprietary NVLink interconnect.

    That networking detail matters more than it may appear. If the industry’s largest inference fleets standardize on open Ethernet for chip-to-chip traffic, the moat around Nvidia’s full-stack platform — GPU plus NVLink plus InfiniBand plus the CUDA software layer — narrows at exactly the layer where Broadcom is strongest. A working, unveiled chip converts that thesis from investor-deck material into deployable hardware.

    The Custom-Silicon Race Nobody Can Sit Out

    Every major AI buyer now hedges the same way: Google with TPUs, Amazon with Trainium and Inferentia, Meta with MTIA, Microsoft with Maia. OpenAI joining that club is notable because it is not a cloud provider — it is the highest-profile pure consumer of AI compute, and its willingness to fund custom silicon signals that even Nvidia’s best customers see strategic risk in single-vendor dependence. None of this displaces Nvidia in the near term; demand still outstrips everyone’s supply, and custom chips typically serve internal workloads rather than the open market.

    The realistic effect is on the margin: each gigawatt of inference that moves to custom silicon is pricing leverage for buyers and a ceiling on how much of the AI build-out flows through one vendor. For data-center operators, the practical takeaway is architectural diversity — facilities must now plan for heterogeneous racks, Ethernet-based scale-up fabrics, and the power and cooling densities these custom systems demand, rather than a single GPU-defined template.

    Background

    OpenAI, the developer of ChatGPT and the GPT model family, has pursued an aggressive infrastructure expansion as usage of its models has grown, layering large compute agreements with cloud and chip partners. In October 2025 it announced a partnership with Broadcom — a semiconductor and networking company best known in AI for co-designing Google’s TPU accelerators and for its data-center Ethernet switch silicon — to build and deploy OpenAI-designed accelerator racks totaling roughly 10 gigawatts, connected with Broadcom’s Ethernet technology.

    The move places OpenAI in a well-established industry pattern: Google, Amazon, Meta, and Microsoft have all built in-house AI chips to supplement Nvidia GPUs, control costs, and secure supply. The June 2026 unveiling of an LLM-optimized inference chip is the first public product milestone of the OpenAI–Broadcom program.

    Source: OpenAI and Broadcom unveil LLM-optimized inference chip — announcement dated June 24, 2026, carried via Google News; analysis draws on the companies’ previously disclosed October 2025 partnership.

  • EPA’s AI Data Center Exit: 57 Disputed Turbines Show Where Permit Risk Lands

    EPA’s AI Data Center Exit: 57 Disputed Turbines Show Where Permit Risk Lands

    TL;DR · 30-second read

    The Short Version

    The federal government’s main pollution regulator said on June 10 that it will not write national environmental rules for the giant computer warehouses that run artificial intelligence. States and local towns will decide instead.

    That does not mean fewer fights. Some of these sites burn natural gas on their own property to get electricity quickly. In April, civil rights and environmental groups sued the company xAI, saying it ran 57 gas-fired generators near Memphis without the pollution permits the law requires. How that case ends will decide who gets to challenge these projects in court.

    US Environmental Protection Agency (EPA) Administrator Lee Zeldin said at the Politico Energy Summit on June 10, 2026, that the agency will not pursue any nationwide environmental requirements or recommendations aimed at the AI industry, Crowell & Moring reported. Zeldin acknowledged that technologies exist to reduce data centers’ water consumption and air pollution, but said deciding the right practices for each facility is a matter for states and local communities, not EPA.

    The announcement follows a year of EPA actions easing permitting for power generation and data centers, and it lands amid active litigation: an April 14 Clean Air Act citizen suit by the NAACP, the Southern Environmental Law Center and Earthjustice alleging that xAI and its subsidiary MZX Tech operated 57 unpermitted portable methane gas turbines to power the Colossus 2 data center in South Memphis, and a June 16 Department of Justice motion to intervene and have that suit dismissed.

    Executive Summary

    EPA has taken itself out of the business of writing AI-specific environmental rules. There will be no national standard for data center water use, air emissions or cooling practices, and no federal list of recommended best practices. That removes a layer of regulation developers had reason to expect, but it does not suspend the Clean Air Act or the Clean Water Act, and it does not change the fact that most day-to-day permitting decisions sit with state and local agencies.

    The practical effect is to relocate risk rather than remove it. The part of an AI campus most exposed is on-site combustion — gas turbines and backup engines that make electricity on the property — because that equipment is what triggers air permits. With no federal template, those permits will be judged against differing state and local standards, and enforcement increasingly runs through private lawsuits, as the xAI case shows.

    Whether that private enforcement route stays open is now itself in question. The Department of Justice has argued that private groups cannot pursue Clean Air Act suits over the federal government’s objection. The outcome of that argument will shape the legal risk attached to fast-tracked on-site power across the industry.

    Stepping Back Is Not the Same as Fewer Permits

    Zeldin’s June 10 statement rules out new federal requirements and recommendations aimed at AI. It does not rewrite existing law. Data centers that burn fuel on site still fall under the Clean Air Act, and those that discharge water still fall under the Clean Water Act. New Source Review (NSR) — the Clean Air Act program requiring a permit before a large new pollution source is built — still applies. Until EPA finalizes the rule it proposed in May 2026, a major NSR permit is still required before any construction begins.

    Read in sequence, EPA’s recent actions share a pattern. In May 2025 it issued guidance allowing certain emergency engines to run up to 50 hours a year in non-emergency conditions to support grid reliability for data centers. In September 2025 it gave developers flexibility to start non-emissions-related building work before obtaining a Clean Air Act permit. In May 2026 it proposed redefining when construction legally ‘begins’ so non-emitting structures can go up earlier. Each step lets the concrete and steel move faster. None of them exempts the equipment that actually emits.

    Why 57 Turbines Mark Where the Risk Now Sits

    The server halls of an AI campus do not need an air permit; the generators feeding them can. That is the mechanism behind the title’s claim. When grid connections lag behind construction schedules, some developers bridge the gap with on-site gas generation, and that generation is precisely where Clean Air Act obligations bite. The citizen suit filed April 14 alleges that xAI and MZX Tech ran 57 portable methane gas turbines without required air permits to power Colossus 2 — a fleet the complaint treats as the data center’s power supply, not an ancillary backup. The allegations have not been proven in court, and the plaintiffs seek substantial penalties and cite health risks for families in South Memphis and North Mississippi.

    With EPA declining to set a national standard, the questions that decide cases like this — whether equipment crosses major source thresholds (the emissions levels above which a facility needs the most demanding permits), what controls are required, how long ‘temporary’ generation may run — are answered by state air agencies and local boards, and tested by private plaintiffs in federal court. A citizen suit is a lawsuit a private party can bring to enforce a federal environmental law after giving 60 days’ notice. The people most affected are developers using on-site generation to beat grid timelines, the lenders and customers relying on those schedules, and the communities living next to the equipment.

    The Justice Department’s Intervention Is the Swing Variable

    On June 16, the Department of Justice moved to intervene in the xAI case and have it dismissed. Its filing argues that the suit threatens national security by seeking to cut power to AI infrastructure that supports the Department of War’s operations, and, more broadly, that the executive branch holds exclusive authority to enforce federal law, so private parties cannot pursue Clean Air Act citizen suits over the government’s objection.

    If a court accepted that theory, it would sharply narrow a private right of action Congress wrote into the statute and raise separation-of-powers questions, since the executive branch would effectively decide which cases reach a judge. Crowell & Moring’s attorneys consider it unlikely the court will accept the argument. Either way, the ruling matters well beyond Memphis: it will tell developers whether federal non-enforcement actually lowers their exposure, or whether private enforcement fills the space EPA has left.

    A Patchwork Rewards Early Paperwork

    Without a national baseline, requirements will diverge by state and locality, and a multi-state developer will face different rules for the same turbine or cooling design. That favors operators who secure permits before breaking ground, audit whether their generation crosses major source thresholds, and treat compliance with federal law as protection against both citizen suits and a future administration that enforces more aggressively.

    Zeldin’s own acknowledgment that pollution- and water-reducing technologies exist points to a hedge: voluntary adoption of measures such as closed-loop cooling, which recirculates water rather than evaporating it, may reduce the odds of state action and local opposition. The competitive edge in this environment belongs less to whoever builds fastest than to whoever can show a regulator, a neighbor or a judge that the power plant behind the data center was permitted from day one.

    Background

    The US Environmental Protection Agency sets national rules under laws such as the Clean Air Act and Clean Water Act, but state and local agencies issue and enforce many of the permits that apply to individual facilities. The rapid growth of AI data centers — and the electricity they need — has drawn activity from Congress, the executive branch, federal agencies and the courts, without yet producing a cohesive regulatory framework.

    Since May 2025, EPA has issued a string of actions aimed at speeding data center and power plant construction, and in January 2026 it held a roundtable with the Data Center Coalition, the industry’s trade association. xAI’s Colossus 2 in South Memphis has become the most visible test of how environmental law applies to AI campuses that generate their own power on site.

    Sources

    Source: EPA Steps Back from AI Data Center Regulation: What Developers Must Know Now – Crowell & Moring LLP — EPA’s June 10, 2026 decision not to set national environmental standards for AI data centers, and the Clean Air Act litigation over xAI’s Colossus 2 turbines.