jain.com — data centers, connectivity and security, analyzed daily

Author: Deepak Jain

  • Everyone Is Watching Meta’s $50B Campus. The Ratepayer Question Is Its 5 GW Math.

    Everyone Is Watching Meta’s $50B Campus. The Ratepayer Question Is Its 5 GW Math.

    TL;DR · 30-second read

    The Short Version

    Meta is building a giant artificial intelligence computing complex in rural Louisiana, estimated at $50 billion. The local power company, Entergy Louisiana, plans to build new gas-fired generators to supply it.

    On August 12, Louisiana’s utility regulators voted 3 to 1 to let Meta keep secret how it worked out how much electricity it needs. Officials have put that need at roughly several large power plants’ worth.

    That matters to ordinary households. An adviser to the regulators estimated the deal would raise average monthly power bills by $8 to $13.

    The Louisiana Public Service Commission voted 3-1 on Wednesday, August 12, 2026, to overturn an administrative law judge’s order that would have required Meta to disclose how it calculated the job projections and electricity needs for Hyperion, its artificial intelligence data center campus in Richland Parish, the Louisiana Illuminator reported. The project has been estimated at $50 billion. Republicans Jean-Paul Coussan, Mike Francis and Eric Skrmetta voted to reverse the order, Democrat Davante Lewis voted to uphold it, and Democrat Foster Campbell was absent.

    The Alliance for Affordable Energy, the Union of Concerned Scientists, the Sierra Club and other groups had sought the records before the commission approves Entergy Louisiana’s plans to build generation for the campus: three natural gas-fired turbines initially, plus seven more generating units for a planned expansion. Administrative Law Judge Melanie Verzwyvelt approved their subpoena on July 10. Meta appealed, arguing the information contains trade secrets.

    Executive Summary

    The ruling keeps confidential the detailed load forecast behind one of the largest single-customer power buildouts in the country. The headline figures are public. Meta, Entergy and state officials have estimated Hyperion will need 5 gigawatts of power and support 1,000 permanent jobs. The derivation of those numbers is not public, and that includes how quickly demand ramps and what assumptions drive it. That derivation is what Entergy relied on when it asked regulators to approve new generation.

    This matters because the generation will be built and paid for through a regulated utility. A consultant who advises the commission, Lane Sisung, filed an analysis in June estimating that the deal, if approved, would raise Entergy Louisiana customer bills by $8 to $13 per month on average. The Alliance for Affordable Energy has warned that customers could bear additional costs if Meta later backs out of its agreements.

    The vote may be the final word on disclosure within the commission. The Alliance’s executive director, Logan Burke, said the group’s attorneys are weighing whether they can act before the full case concludes, including potentially in state court.

    The Contested Number Is the Forecast, Not the Turbines

    Most coverage of AI data centers focuses on the dollar figure, here $50 billion, or on the generation hardware. In Louisiana, the dispute that reached the commission was narrower: how Meta calculated its electricity needs. That calculation matters because a utility sizes new generation around a customer’s load forecast. Load means the amount of electricity a customer draws, and the forecast describes how much it will draw and when. Entergy’s applications to build turbines for Hyperion include power-need figures derived from Meta’s calculations, and the intervening groups argued Entergy never adequately explained how those figures were produced.

    The scale explains why the forecast carries so much weight. Officials have put Hyperion’s eventual demand at 5 gigawatts, a gigawatt being a thousand megawatts. The build plan runs to ten generating units: three gas turbines to start and seven more for the announced expansion. Suppose the forecast overstates demand, or the ramp arrives later than modeled. Then capacity gets built ahead of the load it was meant to serve. Under a regulated model, the cost of stranded or underused plants can end up in the rate base shared by all customers unless contract terms prevent it. That is the concern the Alliance for Affordable Energy has raised. The commission’s own consultant has already put the average bill impact at $8 to $13 per month if the deal is approved, and that estimate itself rests on the load assumptions now kept confidential.

    The people affected are Entergy Louisiana’s residential and business customers, whose bills move with the approved generation, and the commissioners, who must approve that generation without the underlying forecast on the public record. Entergy is affected too: its applications will be judged partly on figures it cannot fully show.

    Two Arguments, Each With Unanswered Questions

    The majority’s position has substance. Commissioner Coussan argued that the general figures for jobs and power demand are already public, that the granular data is proprietary, and that letting a third party not directly involved in commission proceedings subpoena sensitive information would set a bad precedent and could have “a chilling effect” on companies considering Louisiana. Detailed load and staffing projections can reveal commercial plans, and that concern is not unique to Meta.

    The intervenors’ position has substance too. A headline figure like 5 gigawatts says little about timing, ramp rate or how firm the demand is, and regulators need those factors to judge whether ten generating units are the right size and pace. The groups have not shown that Meta’s forecast is wrong. Their argument is that it has not been explained, and that distinction matters when weighing their case. Coussan also described the effort as politically motivated, citing speakers at the hearing. That characterization does not address the cost-allocation question the filings actually raise, and the question stands on its own regardless of who asks it.

    The ruling also leaves one question open: whether a middle path was available, such as review of the data under confidentiality protections by commission staff or intervenors. Neither side is reported to have publicly addressed that option, and it is the most obvious way to reconcile trade-secret protection with ratepayer scrutiny.

    What It Signals for Single-Customer Power Buildouts

    Hyperion concentrates two risks in one project: a single customer’s demand forecast and a regulated utility’s capital program. When generation is built mainly for one large load, the protections in the supply contract matter as much as the forecast. These include minimum payment terms, exit fees and collateral. None of those terms has been made public, and the non-disclosure agreements signed by state and local officials, including Gov. Jeff Landry, have kept much of the deal’s detail out of view.

    This is one commission’s ruling, and it would be an overreach to call it a national pattern. But any regulator asked to approve generation built primarily for a single data center customer faces the same question Louisiana just answered: how much of the demand case must be public before costs can be assigned to the broader customer base. The next test here is whether the intervening groups take the matter to state court, and how the commission rules on Entergy’s generation applications themselves.

    Background

    Meta selected Richland Parish in northeast Louisiana for Hyperion, an artificial intelligence data center campus projected at about $50 billion. Meta, Entergy and state officials have estimated it will support 1,000 permanent jobs and require 5 gigawatts of power. Entergy Louisiana, the regulated utility serving the region, contracted to supply the campus and applied to the Louisiana Public Service Commission to build natural gas-fired generation for it. Many details of the arrangement have been covered by non-disclosure agreements signed by state and local officials.

    The Louisiana Public Service Commission regulates the state’s investor-owned utilities, including approving new generation and the rates customers pay for it. Because generation approved for one customer can affect the rates of all customers, intervening groups have pressed for the demand calculations behind Entergy’s applications to be disclosed.

    Sources

    Source: Meta can keep its data center details private, PSC decides, from the Louisiana Illuminator, on the commission’s 3-1 vote overturning a subpoena for Meta’s Hyperion job and power calculations.

  • Two of Australia’s 16 Cables Broke. For Data Centers, the Gap Is Fault Attribution

    Two of Australia’s 16 Cables Broke. For Data Centers, the Gap Is Fault Attribution

    TL;DR · 30-second read

    The Short Version

    Two of the undersea internet cables that link Australia to the rest of the world stopped working off Perth on the same weekend. Only 16 such cables connect the entire country.

    At the same time, a tanker ship managed from the United Arab Emirates was tracked sailing back and forth in loops over the protected strip of seabed where the cables lie. The cable company says this could be coincidence and has asked federal police to investigate.

    The bigger problem is that the cables had no built-in fault sensors. Nobody can say what broke them until a repair crew physically inspects them.

    The Maritime Executive reported on August 13, 2026, that fiber operator Subco lost two subsea cables back-to-back over the preceding weekend. Both breaks occurred inside a government-designated cable protection zone off Perth, Western Australia. At about the same time, a UAE-managed product tanker (a ship that carries refined fuels) sailed an unusual racetrack pattern over the charted cable area and crossed it several times. Security consultancy Windward independently validated the vessel’s presence and identity.

    Subco founder and CEO Bevan Slattery said the ship’s activity may be a coincidence. He nonetheless urged the Australian Federal Police to investigate urgently and said this could be the second such incident in a federally declared cable protection zone. The cables had no automatic diagnostic systems, so Subco will not know the root cause, including whether an anchor was dragged across the seabed, until it arranges a physical inspection.

    Executive Summary

    Australia’s international connectivity runs through 16 subsea cables, and two of them went down together off Perth. The breaks happened in a zone where anchoring and certain fishing are restricted precisely to protect cables. A tanker was moving back and forth overhead at the time, and surface conditions were benign. That combination made the operator raise the possibility of a security incident rather than treat the outage as routine maintenance.

    The more consequential detail is operational. Because the cables lacked automatic diagnostics, Subco cannot yet tell an accidental anchor drag from deliberate interference. Anchor drag alone causes about a third of cable breaks worldwide, according to the International Cable Protection Committee. Until a repair ship reaches the fault, investigators, insurers, customers and policymakers have to act without knowing which kind of event this was.

    For anyone running data centers or cloud services that depend on international routes, the difference matters. An accident is a random event that route diversity absorbs. A deliberate act can recur and can target several cables at once, and that changes how much redundancy is enough.

    Two Out of Sixteen Is Not a Rounding Error

    Subsea cables carry the overwhelming majority of intercontinental data traffic. That includes the traffic that lets Australian businesses reach cloud regions and AI services hosted overseas, and lets local data centers serve customers across the Indian Ocean and Asia. With just 16 cables linking Australia to the world, losing two at once removes a significant share of the country’s international paths. The Maritime Executive described that as a substantial security event.

    Geography sharpens the problem. Perth is Australia’s western gateway, and cables landing there take routes toward Asia and the Middle East that east-coast systems do not. When two breaks cluster in the same protected corridor, traffic has to shift onto fewer routes. That can mean longer paths, higher latency (the delay data experiences in transit) and less spare capacity if anything else fails before repairs are complete.

    The Real Gap: Nobody Can Yet Say What Happened

    The detail that should concern infrastructure operators most is that Subco’s cables had no automatic diagnostic systems. Without them, the operator knows the cables stopped carrying light but not why. It has to send a vessel to locate the fault, lift the cable and examine the damage before it can distinguish an anchor strike from a snag, a fishing gear incident or deliberate cutting. Until then, the circumstantial evidence is all anyone has: back-to-back breaks, a ship making multiple crossings in a racetrack pattern, and calm surface conditions that remove bad weather as an easy explanation.

    This attribution delay has direct consequences for data center and cloud operators. Resilience planning for subsea routes has traditionally assumed faults are independent accidents. Under that assumption, buying capacity on two or three diverse cables makes a simultaneous outage very unlikely. That math weakens if breaks can be deliberate and coordinated, because a single actor can hit several cables in one corridor, as two breaks in one protection zone illustrates. An operator that cannot tell which kind of event it is facing cannot tell whether its redundancy is adequate. It also cannot tell whether the same corridor is safe to rely on while repairs proceed.

    The same gap slows everyone downstream. Police cannot easily build a case, insurers cannot assign a cause, and governments cannot calibrate a response. Slattery asked the Australian Federal Police to move urgently, and the physical evidence on the seabed is what any investigation will ultimately depend on.

    Weighing the Evidence on Both Sides

    The case for concern is real but circumstantial. The vessel’s track and timing match the breaks, Windward validated its identity and presence, and Slattery noted this could be the second incident in a federally declared zone. Researchers are also warning about the broader pattern. ASPI’s David Brewster told ABC that many recent cable cuts in Europe and around Taiwan were, in his view, fairly obviously the work of Russia and China. A new review from the Australian National University’s National Security College concludes that state actors appear to be probing cable infrastructure in peacetime so they could degrade connectivity quickly at the start of a conflict.

    The case for caution is equally worth stating. Slattery himself twice said the ship’s behavior may be a coincidence. Anchor drag is a common, ordinary cause of cable damage. Brewster’s remarks concerned incidents elsewhere, not this one, and he acknowledged it is difficult to be certain in some cases. A tanker managed from the UAE is not evidence of any state’s involvement. Unusual vessel movements can have commercial or navigational explanations that have not yet been aired. The fair reading is that the incident warrants investigation, not that its cause is known.

    What Operators and Buyers Should Take From It

    For cable owners, the lesson is that monitoring is now part of the security posture, not just a maintenance convenience. Fault-location and fiber-sensing systems that can flag disturbances as they happen would shorten the window in which an event goes unexplained. Tracking vessel behavior over protection zones as it happens would help too. Protection zones deter accidental damage by lawful mariners, but they do not stop a ship that ignores them. Enforcement depends on noticing quickly.

    For data center and enterprise buyers of international capacity, the practical questions are about correlation, not just redundancy. Are the backup routes physically separate from the primary ones, or do they share the same seabed corridor or landing point? How quickly can capacity be restored, and how many repair vessels are available in the region? Those answers now matter as much as headline bandwidth.

    Background

    Subsea cables are bundles of optical fiber laid along the seabed that carry most of the world’s intercontinental internet and data traffic. They are built to last decades but are physically exposed. Ship anchors and fishing gear are the most common causes of damage, and repairs require specialized vessels that are in limited supply. Australia’s geographic isolation makes it unusually dependent on a small number of these systems, and the government declares cable protection zones off major landing points to restrict risky activity near them.

    Subco is an Australian subsea cable operator founded by Bevan Slattery, a long-time builder of Australian digital infrastructure. Concern about deliberate cable damage has grown since a series of incidents in the Baltic Sea and around Taiwan over the past several years. Security researchers there have pointed to state actors seeking disruption while keeping deniability, and that context is why every unexplained break now draws closer scrutiny.

    Sources

    Source: Suspicious Subsea Cable Breaks Prompt Calls for Investigation in Australia (The Maritime Executive): Subco reports two cable breaks off Perth coinciding with a tanker’s movements over a cable protection zone.

  • Mistral’s 1 GW Plan Shows Why Sovereign AI Compute Needs Buyers Before Builders

    Mistral’s 1 GW Plan Shows Why Sovereign AI Compute Needs Buyers Before Builders

    TL;DR · 30-second read

    The Short Version

    Mistral, a French artificial intelligence company, says it plans to build up to one gigawatt of computing capacity by 2030. That is about the output of one large nuclear reactor, all feeding data centers that run artificial intelligence.

    The unusual part is how it wants to pay for it. It is lining up big European companies to commit to years of computer time in advance, so the buildings have customers before they go up.

    The goal is artificial intelligence in Europe that does not depend on American tech giants.

    Mistral AI said in an August 11 blog post that it is taking three steps toward what it calls AI sovereignty. First, Mistral Regional Endpoints are now generally available, letting customers choose whether inference (the work of running a trained model to answer requests) happens in Europe or the US. Second, a new Mistral Priority Tier, in public preview, offers custom rate limits backed by an uptime service-level agreement. Third, Mistral’s platform will host third-party open models, starting with Z.ai’s GLM-5.2.

    Mistral is also assembling an anchor group of enterprises whose multi-year commitments will be converted into European Compute Units (ECUs), giving access to Mistral-built infrastructure over several years. The company says it plans to build up to 1 GW of capacity by 2030. The chief executives of Amadeus, ASML, Capgemini, Caisse des Dépôts and CMA CGM are quoted in support.

    Executive Summary

    The headline products are incremental: region selection for inference and an SLA-backed service tier are features enterprise buyers already expect from large cloud providers. The strategically significant piece is the ECU program, which changes the order in which AI capacity gets built. Rather than constructing data centers and then selling time on them, Mistral is gathering multi-year enterprise commitments first and letting that pooled demand decide what is built, where, and for whom.

    That matters because the 1 GW target is large by European AI-infrastructure standards, and capacity at that scale is hard to finance without contracted revenue behind it. The announcement also broadens Mistral from a model developer into a platform that hosts other developers’ open models on its own regional infrastructure, a move that widens the pool of workloads that can fill the capacity it plans to build.

    What the announcement does not yet show is scale of commitment: no ECU volumes, prices, sites or financing structure were disclosed, and the endorsing CEOs describe support rather than signed quantities.

    Why the 1 GW Plan Starts With Buyers, Not Buildings

    AI capacity is expensive long before it earns anything. Sites, grid connections, cooling systems and accelerators (the specialised chips that run models) are paid for years before the revenue matures, and the chips lose value as newer generations arrive. Lenders typically size financing against contracted revenue, and grid operators and utilities plan more readily around connection requests backed by credible demand. A provider that builds speculatively carries the risk of idle capacity on depreciating hardware.

    European Compute Units are Mistral’s answer to that problem. The company says its anchor group’s multi-year commitments “can support infrastructure in Europe at a scale no participant could secure alone,” and that aggregating long-term demand “will help determine what capacity is built, where it is located, and whom it serves.” In effect, this is an offtake model, familiar from power projects where a plant is financed against buyers who have agreed to take its output, applied to AI compute. The commitments come first; the build follows them.

    The consequences run in several directions. Enterprises trade flexibility for assured access, making a multi-year bet on their own AI usage. For Mistral, pooled commitments are what can turn a target as large as 1 GW, roughly the output of a large nuclear reactor, into phases that can be financed rather than a speculative build. For site developers, utilities and grid operators, one aggregated buyer is easier to plan around than dozens of small ones. Mistral has not said how much has been committed so far, so how much of the 1 GW is already spoken for remains open.

    In-Region Inference, With a Stated Caveat

    Regional Endpoints let customers pin inference to Europe or the US, aligning processing with data-residency rules, sector regulation and latency needs. For European firms handling personal or regulated data, keeping processing in-region simplifies compliance conversations, and Mistral notes that most of its customers already run its models in their own data centers or cloud environments. The endpoints serve those who want Mistral-operated capacity alongside that.

    Mistral is precise about the limits: processing in the chosen region is “subject to limited, safeguarded transfers to sub-processors that may occur outside that region,” as detailed in its Trust Center. That candour is useful, and regulated buyers will want the sub-processor list and the nature of those transfers before treating residency as absolute.

    The Priority Tier adds custom rate limits and an uptime SLA for mission-critical workloads, though the uptime figure and pricing were not published. Mistral’s claim to be the only European AI lab offering both region choice and an SLA-backed tier is scoped to European labs and not independently verifiable from the announcement; large US cloud providers already offer regional, SLA-backed hosting of many models in Europe, which is the comparison many buyers will actually make.

    Third-Party Open Models Help Fill the Capacity

    Mistral’s platform will now run open-weight models from other developers, starting with Z.ai’s GLM-5.2, under the same infrastructure, regional controls and service commitments as Mistral’s own models. Open weights mean the trained model parameters are published, so a customer can inspect, adapt and host the model where it chooses.

    The commercial logic ties directly to the capacity plan. A platform that hosts several models keeps a customer’s workloads, and spend, on Mistral infrastructure even when another developer’s model suits a task better. Factory’s CEO, Matan Grinberg, frames it this way: different workloads need different models, and running them under one set of regional controls eases compliance. Utilisation is what makes data center capacity pay for itself, and a broader catalogue widens the demand that can fill 1 GW.

    There is a nuance buyers will weigh. Z.ai is a Chinese developer. Running its open weights on European infrastructure means prompts and data need not reach the model’s creator, but some public-sector and regulated buyers apply provenance policies to models themselves, not just to where they run. How much that matters will depend on each organisation’s own rules.

    Endorsements Are Not Yet Contracts

    The breadth of names quoted is notable: travel technology (Amadeus), semiconductor equipment (ASML), IT services (Capgemini), French public finance (Caisse des Dépôts) and shipping and logistics (CMA CGM). That spread suggests appetite for European AI capacity well beyond the tech sector.

    The quotes, however, describe ambition and confidence rather than quantities. None specifies a capacity figure, contract length or value, and the announcement does not say which of these organisations hold ECU commitments. ASML is also a significant investor in Mistral, which is relevant context for its endorsement. The most concrete usage claim comes from CMA CGM, which says deployment is already under way among thousands of employees. Caisse des Dépôts describes Mistral Compute as a “European neocloud,” a term for newer providers that specialise in renting AI compute.

    The test between now and 2030 is conversion: whether endorsements become signed ECU volumes large enough to anchor each successive phase of the build.

    Background

    Mistral AI is a Paris-based AI developer founded in 2023 and best known for releasing open-weight models alongside commercial ones. Its product line now spans Studio for building agents and apps, Forge for training and customising models, the Vibe agents, and an AI Cloud business, Mistral Compute, that provides infrastructure for training and inference. Most of its customers today run its models inside their own data centers and cloud environments.

    The announcement lands in a long-running European debate over dependence on US cloud providers for critical digital infrastructure. European organisations operate under data-protection rules such as GDPR and the EU AI Act, and many public bodies and regulated industries want assurance about where data is processed and who controls the underlying systems. Demand for AI capacity has made that question more pressing, because the compute behind modern AI is scarce, capital-intensive and concentrated among a few large providers.

    Sources

    Source: In-region inference, open models, and new European infrastructure for sovereign AI. Mistral AI’s announcement of Regional Endpoints, the Priority Tier, third-party open models and the European Compute Units program.

  • Imperial Valley’s 330 MW Lawsuit Shows Water Rights Can Gate AI-Era Capacity

    Imperial Valley’s 330 MW Lawsuit Shows Water Rights Can Gate AI-Era Capacity

    TL;DR · 30-second read

    The Short Version

    A company planning California’s largest data center, a huge building full of computers like the ones that run AI tools, was refused the water it needs to keep those machines cool. It is now suing the public agency that controls the local water.

    The water comes from the Colorado River. About 40 million people in seven states depend on that river, and drought has strained it. The company wants to take over water that farms use today.

    The court case could help decide whether water, not only electricity, limits where the AI boom can be built.

    BGR reported on August 12, 2026 that Imperial Valley Computer Manufacturing has sued the Imperial Irrigation District (IID), the public agency that delivers Colorado River water to California’s Imperial Valley. IID had denied the developer’s request to use the district’s water supply and wastewater to cool a planned 330-megawatt data center, which would be the largest in California. The facility is expected to require 287 million gallons of water.

    The developer plans to source that water by reallocating it from farmland. Its lawsuit argues there is “no rational basis” for IID to bar an industrial company from combining several fields into a single unit for water reallocation while allowing agricultural users to do the same.

    Executive Summary

    Most debate about where AI data centers can be built focuses on electricity: whether the grid can deliver hundreds of megawatts, and how fast. The Imperial Valley dispute puts a different constraint in front of a court. The developer’s cooling plan depends on moving existing farm water allocations to an industrial user, and the agency that controls those allocations has said no.

    The legal question is narrow, whether an industrial user may aggregate and transfer water the way farmers already can. The practical stakes are broad. If water rules can block a 330-megawatt project in one of the most water-contested basins in the country, developers eyeing arid regions will have to treat water rights as a gating item on par with a power connection, not as a permit to settle later.

    Sebastian Rucci, the figure leading the project, argues the data center would use about as much water as a 160-acre farm and would benefit the local economy. Critics counter that reallocating farm water puts rural communities at risk, and local residents broadly favor a moratorium on new data center construction.

    Why Water Rights, Not Just Megawatts, Gate This Project

    A data center’s electrical rating, here 330 megawatts, describes how much power its computers and support systems can draw. Nearly all of that electricity ends up as heat, which has to be removed continuously. Many large facilities do this with evaporative cooling, in which water is evaporated in cooling towers to carry heat away. That design is efficient on electricity but consumes water in proportion to the heat load, which is why a 330-megawatt campus arrives with a 287-million-gallon water requirement attached.

    In the Imperial Valley, that water has to come from somewhere already spoken for. IID delivers Colorado River water to the region, and the developer’s plan is to reallocate water currently tied to farmland. IID’s denial covered both its water supply and its wastewater, closing off reclaimed water as well as fresh. As things stand, the project’s cooling design cannot proceed on the terms the developer proposed, whatever its power situation. That is the mechanism the title points to: a local water agency’s allocation rules, not the grid, are the decision point currently holding up this capacity.

    The people affected extend beyond one developer. Anyone siting large facilities in the arid Southwest, where land and solar power are attractive, now has a live example of water access acting as a hard stop. Water agencies are affected too, since a ruling will signal how much discretion they have to treat industrial and agricultural transfers differently.

    The Legal Question: Can Industry Aggregate Water Like Farms Do?

    The lawsuit’s core argument is about equal treatment. According to the complaint as described, IID lets agricultural users combine multiple fields into a single unit for water reallocation but refused to let an industrial company do the same. The developer says there is “no rational basis” for the distinction.

    That phrasing echoes a well-established legal standard under which a government body’s line-drawing generally survives if any reasonable justification exists. It is typically a demanding standard for challengers. IID may point to its mandate to serve agriculture, drought conditions on the Colorado River, or the difference between water that stays in farming and water permanently moved to a new industrial use. The developer will argue that water is water and that the district cannot favor one class of user without a principled reason.

    Either outcome sets a marker. A win for the developer would suggest farm water allocations in the district can be assembled and redirected to data centers, which would change land and water economics in the valley. A win for IID would confirm that water agencies can effectively decide whether water-intensive data centers get built in their service areas.

    Testing the 160-Acre Farm Comparison

    Rucci’s claim that the facility would have as much impact as a 160-acre farm can be checked against the stated figure. Dividing 287 million gallons across 160 acres works out to roughly 1.8 million gallons per acre, or about 5.5 acre-feet, the unit water managers use, where one acre-foot covers an acre of land a foot deep. Whether that matches a typical Imperial Valley farm depends on the crop and on whether 287 million gallons is an annual figure, which has not been specified.

    The comparison also answers a different question from the one critics are asking. Their objection is less about the total volume than about where it comes from: water moved out of farming, in a region where agriculture sustains rural livelihoods, from a river that serves 40 million people across seven states. It also leaves open how much of the data center’s water would be consumed through evaporation versus returned to the system, a distinction that matters for downstream users.

    Local Opposition Is Part of the Siting Math

    Residents in the area generally favor a strict moratorium on new data center construction, according to the reporting, reflecting concerns about utility prices and water strain that have followed data center projects in other regions. Rucci has argued the project would bring economic benefits, but the specifics of jobs, tax revenue or community agreements have not been laid out.

    For the industry, the lesson is that community sentiment and water governance are converging into a single siting risk. A project that must win both a court case over water allocation and a skeptical local public faces a longer and less predictable path than one that secures water and community support before it announces its scale.

    Background

    The Imperial Valley sits in California’s southeastern desert and depends on Colorado River water delivered by the Imperial Irrigation District, much of it for farming. The Colorado River supplies roughly 40 million people across seven U.S. states and has faced years of drought, making any new large draw on it contentious.

    Data centers have become one of the fastest-growing sources of industrial demand for both electricity and water as companies build capacity for artificial intelligence. Their water use comes mainly from cooling, and communities in several regions have pushed back on new facilities over utility costs and water strain, making water sourcing an increasingly visible part of the siting process.

    Sources

    Source: The Largest Data Center In California Has Been Denied Water For Cooling — Now It’s Suing A Public Agency (BGR), on Imperial Valley Computer Manufacturing’s lawsuit against the Imperial Irrigation District over cooling water for a planned 330 MW data center.

  • Virginia’s 3.8 GW Data Center Trip Shows Why Ride-Through Is Becoming a Grid-Access Test

    Virginia’s 3.8 GW Data Center Trip Shows Why Ride-Through Is Becoming a Grid-Access Test

    TL;DR · 30-second read

    The Short Version

    On July 22, an ordinary electrical fault on a power line in northern Virginia caused a cluster of data centers, the buildings full of computers that run the internet, to disconnect from the grid almost at once.

    In minutes, about 4% of all the electricity being used across 13 states and Washington, D.C. vanished from the system.

    The organization that runs that grid says the data centers were too jumpy. It is now considering rules requiring them to stay connected through brief power glitches. If those rules arrive, how a data center reacts to a glitch could decide whether it gets connected at all.

    Utility Dive reported that PJM Interconnection, the grid operator for 13 Mid-Atlantic and Midwest states and the District of Columbia, is considering new interconnection reliability requirements for “computational loads” such as data centers and crypto-mining facilities. The requirements under consideration include “ride-through” standards, which require equipment to stay connected through brief voltage and frequency disturbances rather than disconnecting.

    The trigger was a July 22 event in which about 3,800 MW of data center load in northern Virginia tripped offline in two waves after a fault on a 230-kV transmission line in Dominion Energy’s zone. PJM’s total load fell about 3.8%, from 99,984 MW to 96,205 MW. PJM called it the largest such event in its history.

    Executive Summary

    PJM is evaluating whether to expand its reliability requirements for large computing loads after data centers disconnected from the grid in response to what its Operating Committee chair called “a normally cleared fault.” In other words, the grid did what it was supposed to do, and the data centers dropped off anyway. PJM staff told an August 6 Operating Committee meeting that the facilities appear to be disconnecting too early.

    This matters because it shifts part of the grid-reliability burden onto the load side. Until now, the obligation to stay online through disturbances has fallen mainly on power plants. PJM’s position is that a data center cluster large enough to remove nearly 4% of system demand in minutes now needs the same kind of behavioral rules. As David Souder, PJM’s executive director of system operations, put it, those requirements need to be “in place before the load comes on the system.”

    The timing also matters. The North American Electric Reliability Corp. (NERC) is building national standards for computational loads under a Federal Energy Regulatory Commission directive. PJM said, however, that NERC will not address voltage and frequency ride-through until sometime next year. That leaves PJM weighing whether to act in the meantime.

    How a Routine Fault Became a 3,800 MW Problem

    Transmission faults happen constantly. A line gets struck by lightning or contacts a tree, protective relays isolate it within a fraction of a second, and voltage briefly sags on nearby parts of the network before recovering. Power plants are expected to “ride through” that sag, meaning they stay connected and keep producing. Data centers are generally built with the opposite instinct. Their electrical systems are designed to protect servers from bad power, so when voltage moves outside a set tolerance window, they switch to batteries and on-site generators and stop drawing from the grid.

    For a single facility, that is sensible engineering. For a cluster, it becomes a system event. When thousands of megawatts of demand leave the grid at the same moment while generation keeps running, supply suddenly exceeds demand. Frequency and voltage swing, and the operator has to rebalance quickly. PJM said the July 22 event caused “a high imbalance between generation and load” and large swings in both voltage and frequency, and that it dispatched reactive power resources to bring system voltage down. PJM’s Operating Committee chair, Emanuel Bernabeu, described the data centers as “too sensitive to the kind of voltage that they like.”

    The Event Sizes Are Growing Faster Than the Rules

    PJM listed two earlier events in the same Dominion zone. Each caused a sudden load transfer of about 1,500 MW, one in 2024 and one in 2025. The July 22 event was roughly two and a half times that size. Three incidents in one region do not make a statistical trend, but the direction is consistent with how northern Virginia is building out: more and larger facilities, clustered on the same transmission network and often reacting to the same disturbance at once. For scale, 3,800 MW is on the order of the output of three large nuclear reactors.

    The grid handled it. PJM said it brought its Area Control Error, the running measure of the mismatch between supply and demand, back within limits in nine minutes, against a NERC standard of 30 minutes. That is a good operational result, and it should not be read as a near miss. The concern PJM raised is forward-looking: each new facility connected under current practice adds to the amount of demand that could disappear in a single disturbance. At some point the operator has to either hold more balancing capacity in reserve or require the load to behave differently. PJM’s move signals it prefers the second option.

    Ride-Through Moves From Design Choice to Condition of Connection

    This is the substance behind the headline. PJM is framing the fix as an interconnection reliability requirement, a condition attached to connecting to the grid, rather than a best-practice guideline. Souder’s remark that requirements must be in place “before the load comes on the system” points to the interconnection queue as the enforcement point. If PJM follows through, projects now in development would need to show that their electrical design keeps them connected through a normally cleared fault. Retrofitting that behavior after energization is harder and more contested.

    The practical effects would be felt in specific places. Developers and colocation operators would need to revisit the voltage and frequency tolerance settings on their uninterruptible power supplies and switchgear, and possibly the hardware itself. Hyperscale tenants would need to accept that their facilities ride out disturbances their servers were previously shielded from. Equipment vendors whose systems can demonstrate grid-friendly behavior would gain a selling point. Utilities such as Dominion, whose zone hosts the world’s largest data center cluster, would gain a clearer technical basis for connection studies. The rules would cover crypto miners too, because PJM used the broader category of “computational loads.” Mining loads are generally more flexible about interruption, but they are no less able to trip off in a block.

    There is a real engineering tradeoff, and it deserves an even-handed hearing. Data center operators set tight tolerances to protect equipment and meet uptime commitments to customers, and loosening them carries risk for the operator. PJM’s argument is that the risk of undisciplined load, which falls on everyone connected to the grid, now outweighs the private risk. Where the thresholds land, and whether PJM offers alternatives such as coordinated or staggered transfer settings, will determine how costly the change becomes.

    A Gap Between Regional Urgency and National Standards

    NERC is scheduled to finalize registry criteria and initial reliability standards for computational loads by December 31, and to produce a plan by March for additional standards. According to PJM, ride-through for minimum voltage and frequency will not be addressed until sometime next year. Meanwhile, Northern Virginia continues to add capacity. That leaves PJM choosing between acting regionally now and risking a mismatch with a later national standard, or waiting and connecting more load under today’s practices. Wharton’s reference to “existing and future industry ride-through standards and practices” suggests PJM intends to align with NERC where it can, without waiting for NERC to finish.

    Background

    Northern Virginia, largely within Dominion Energy’s service zone, hosts the largest concentration of data centers in the world. Rapid growth in data center demand there and elsewhere in PJM’s 13-state footprint has made large computing loads a central grid-planning issue. PJM has reported sudden data center load transfers of about 1,500 MW in both 2024 and 2025, and NERC issued alerts on sudden data center load losses earlier in 2026, including a Level 3 alert in May calling for mandated actions.

    Grid reliability rules in North America are set by NERC and approved by the Federal Energy Regulatory Commission. Regional operators such as PJM can add their own interconnection requirements. Historically those rules focused on generators. The rise of very large, clustered loads has prompted FERC to direct NERC to develop registry criteria and reliability standards specifically for computational loads.

    Sources

    Source: PJM eyes data center, crypto reliability requirements after 3.8 GW of load trips offline (Utility Dive, August 12, 2026), covering PJM’s response to the July 22 northern Virginia load-loss event.

  • Gunra’s 10 p.m.–6 a.m. Pattern Shows Why Edge Appliance Logs Belong Off the Box

    Gunra’s 10 p.m.–6 a.m. Pattern Shows Why Edge Appliance Logs Belong Off the Box

    TL;DR · 30-second read

    The Short Version

    Government agencies in the United States and South Korea have jointly warned about a criminal hacking group called Gunra. It breaks into organizations, copies their private files, locks everything up, then demands payment, threatening to publish or sell the stolen files if victims refuse.

    Its main way in is the security equipment that lets employees log in to work networks from home. Once inside, it erases its tracks and works overnight, between 10 at night and 6 in the morning.

    Hospitals, banks, utilities and universities are among its targets. One bright spot: a flaw in one version lets some victims unlock files without paying.

    Industrial Cyber reported that the FBI, the Cybersecurity and Infrastructure Security Agency (CISA), the Department of Defense Cyber Crime Center (DC3), the National Security Agency (NSA), the US Secret Service and South Korea’s National Police Agency (KNPA) released a joint advisory Monday on Gunra, a ransomware operation derived from leaked Conti source code. Gunra steals data, encrypts systems and threatens to publish or sell the stolen data on its leak site, with victims listed across the Americas, Europe, the Middle East, Africa and the Asia-Pacific region.

    The advisory says Gunra gets in mainly by exploiting known flaws in internet-facing firewall and VPN appliances, including two authentication-bypass vulnerabilities in specific FortiOS and FortiProxy versions (CVE-2024-55591 and CVE-2025-24472), deletes logs to hide, and works primarily between 10 p.m. and 6 a.m.

    Executive Summary

    Six US and South Korean agencies have published a detailed account of how Gunra ransomware operates, from initial access to extortion. Gunra emerged in April 2025, added a Linux variant by mid-2025, and by early 2026 had become a ransomware-as-a-service operation that recruits affiliates on dark web forums. Its leak site lists victims in more than ten sectors, including healthcare, finance, utilities, transportation and government.

    For infrastructure operators, the advisory’s significance is where the attack begins and how it hides. Entry runs through the edge: firewall and VPN appliances with known vulnerabilities, default credentials or weak access controls. Once inside, the actors delete system and network access logs, clear command history, work overnight and run an encryptor that produces no observable network indicators. That combination leaves appliance logs held somewhere the attacker cannot reach, and watched after hours, among the most useful signals a defender has.

    The advisory also contains a rare point in defenders’ favor: as of March 2026, Gunra’s Linux variants use weak, time-seeded randomness that lets encryption keys be reconstructed and files recovered without payment, though that does nothing about data already stolen.

    The Front Door Is the Security Appliance Itself

    The most consequential detail for infrastructure operators is how Gunra gets in. The FBI observed the group obtaining initial access primarily by exploiting known vulnerabilities in internet-facing devices, including firewall and VPN appliances, and named two authentication-bypass flaws in specific FortiOS and FortiProxy versions: CVE-2024-55591 and CVE-2025-24472. An authentication bypass lets an attacker reach functions that should require a login without valid credentials. South Korea’s National Police Agency separately observed Gunra exploiting credential exposure and Secure Shell (SSH) access-control weaknesses on internet-facing VPN gateways.

    The intrusion the agencies describe shows how modest the first step can be: default credentials on an SSL-VPN appliance, the encrypted gateway remote staff use to reach internal systems, with no account lockout to stop repeated guessing. After compromising an administrator workstation, the actors reached the appliance’s administrative console, found an unused account with access to both external and internal networks, and modified it to skip a mandatory password change, creating a persistent backdoor that looks like an ordinary account.

    For data center, colocation and network operators, firewalls and VPN concentrators sit at the boundary of nearly everything they run, including remote administration paths into their own environments. Because both named CVEs are known vulnerabilities, the defense here is patch cadence and asset inventory rather than anything exotic. The practical test is whether an operator can list every internet-facing appliance it owns, its firmware version and every account configured on it, including dormant ones.

    Why Edge Appliance Logs Belong Off the Box

    Gunra is built to leave few traces where defenders usually look. According to the advisory, the actors delete system and network access logs and clear command history while active, conduct reconnaissance and encryption primarily between 10 p.m. and 6 a.m., and run the encryptor as a self-contained binary that produces no observable network indicators. Data is exfiltrated before encryption, so by the time files lock, the extortion leverage has already left the network.

    Taken together, those facts narrow the useful evidence to a short list, and the advisory’s own incident-response guidance points at it: review network appliance logs for anomalous privileged activity and audit file creation to detect exfiltration. Logs that exist only on the compromised appliance or host are exactly what the actors delete. Logs forwarded in near real time to a separate, access-controlled collector survive that deletion, and they only help if someone, or an automated rule, is watching during the overnight window the group favors.

    This falls hardest on organizations that run round-the-clock infrastructure with thin overnight staffing, and on managed service providers that administer appliances for many customers. An after-hours login to a VPN administrative console, a dormant account suddenly modified, or a change to appliance traffic controls are low-volume events that are straightforward to alert on, provided the record sits where the attacker cannot erase it. None of this is new practice, but Gunra’s combination of edge entry, log wiping and night-time activity turns off-device logging from a best-practice line item into the difference between catching the intrusion before encryption and discovering it from a ransom note.

    MFA and Credentials Fail at the Seams

    The advisory is a reminder that multi-factor authentication (MFA) protects a login step, not the systems behind it. Against one victim, Gunra modified files on the virtual desktop infrastructure (VDI) authentication portal, the gateway to remotely hosted desktops, so it would accept a designated one-time password. The actors also manipulated SSL-VPN traffic controls to harvest credentials and session cookies, hijacked sessions to impersonate legitimate users, and entered the internal VDI environment with stolen session tokens.

    Inside, the playbook targets the identity core. Using Impacket libraries over SMB, the Windows file-sharing protocol, the actors moved laterally to the VDI authentication server, the Active Directory server that governs who can access what, and IT staff desktops. They ran secretsdump.py against the NTDS database where domain controllers store password hashes, enabling pass-the-hash and pass-the-ticket attacks that reuse stolen credentials without cracking them. At one victim they reached a Hiware system access control server over SSH and stole the symmetric key needed to decrypt enterprise server credentials stored in its database.

    The operational lesson is that authentication portals, privileged-access systems and key stores are high-value targets in their own right and warrant the same integrity monitoring and segmentation as production systems. The agencies’ recommendations map onto each step: least-privilege administrative rights, audits for unrecognized accounts, MFA on all services with emphasis on VPNs and webmail, network segmentation to restrict lateral movement, and disabling command-line and scripting activity where possible.

    A Flaw in the Linux Encryptor, With Limits

    One finding cuts in defenders’ favor. As of March 2026, the advisory says, Gunra’s Linux ELF variants use a weak pseudorandom number generator seeded with predictable system time, allowing defenders to mathematically reconstruct keys from file timestamps and recover files without paying. That is relevant to infrastructure operators because Linux underpins much of the server estate, and Gunra introduced its Linux variant in mid-2025 as it moved toward cross-platform targeting.

    The limits matter. The weakness is described for the Linux variants as of a specific date, and ransomware-as-a-service operations, which rent their tooling to affiliates, can ship revised builds. Key recovery also does nothing about stolen data: the threat to publish or sell it stands regardless. Offline, immutable backups kept in a physically separate, segmented location, and tested, remain the control that makes recovery independent of both the attacker’s cooperation and the attacker’s mistakes.

    Background

    Gunra traces its code to Conti, a prolific ransomware operation whose source code leaked in 2022 and has since been reused by several groups. Gunra appeared in April 2025 focused on Windows systems, introduced a Linux variant by mid-2025, and by early 2026 had adopted a ransomware-as-a-service model, in which core developers supply the malware and infrastructure while recruited affiliates carry out intrusions in exchange for a share of ransoms.

    Joint advisories from US agencies and foreign partners are a standard way of publishing attacker techniques, indicators and mitigations for a specific ransomware family. They typically align recommendations with CISA and NIST’s cross-sector cybersecurity performance goals, a baseline of practices such as patching internet-facing systems, maintaining offline backups, enforcing multi-factor authentication and segmenting networks.

    Sources

    Source: Gunra ransomware expands globally as US, South Korean agencies warn of data theft, encryption and extortion tactics — Industrial Cyber’s report on the joint FBI, CISA, DC3, NSA, Secret Service and Korean National Police Agency advisory on Gunra ransomware.

  • OpenAI’s 95% vs 1.5% Cyber Test Moves AI Safety From Refusals to Access Control

    OpenAI’s 95% vs 1.5% Cyber Test Moves AI Safety From Refusals to Access Control

    TL;DR · 30-second read

    The Short Version

    OpenAI, the company behind ChatGPT, has released a version of its artificial intelligence built for computer security professionals.

    Its regular version turns down almost every request to help break into computer systems. In the company’s own tests it finished just 1.5 of every 100 such tasks. The new security version finished 95 of every 100.

    The reasoning is that defenders have to think like attackers to find weak spots first. That shifts the safety question from what the software refuses to who is allowed to use it, and OpenAI is widening that circle.

    SecurityWeek reported that OpenAI announced GPT-5.6-Cyber on Monday, August 10, 2026. It is a cybersecurity-focused AI model that OpenAI describes as designed for “advanced, authorized cybersecurity work.” The model is built on GPT-5.6-Sol and trained for specialized tasks such as finding zero-days (previously unknown software flaws) and building exploit chains (sequences of attacks that together compromise a system). OpenAI also configured it with a lower refusal rate for dual-use requests.

    OpenAI also expanded its Daybreak Cyber Partner program, the channel through which it gives more organizations access to its security AI. In the company’s own testing, GPT-5.6-Cyber completed 95% of prompts involving exploit chain development, privilege escalation and authentication bypass. GPT-5.6-Sol completed 1.5% of the same prompts, and the earlier GPT-5.5-Cyber completed 57.3%.

    Executive Summary

    OpenAI has built a second version of its flagship model that differs from the first mainly in what it will agree to do. On offensive-security tasks, the general-purpose GPT-5.6-Sol refuses almost everything. GPT-5.6-Cyber, built on the same model, completes nearly everything. OpenAI says the change responds to security researchers who hit persistent refusals with the previous cyber model.

    This matters because it changes where the safeguard sits. A model that completes 95% of exploit-development requests cannot rely on refusals to keep it out of the wrong hands. The protection now depends on authorization: which organizations are admitted through the Daybreak Cyber Partner program, and under what conditions. OpenAI is expanding that program at the same time it lowers refusals, which makes the admission process the central control.

    For defenders, including the teams that secure data centers, cloud platforms and network infrastructure, the release promises a capable assistant for penetration testing and vulnerability research. The performance claims come from OpenAI’s own evaluation, and the public account does not specify how the vetting works.

    The Guardrail Moved From the Model to the Gate

    The most revealing figures in the announcement are completion rates, not capability scores. On prompts covering exploit chain development, privilege escalation (turning limited access into administrator control) and authentication bypass (getting past a login), GPT-5.6-Sol completed 1.5%. GPT-5.6-Cyber, built on Sol, completed 95%. The underlying model family is the same. What changed is specialized training and a deliberately lower refusal threshold for dual-use work, meaning tasks that serve a penetration tester and a criminal equally well.

    Consumer AI models have long relied on refusal as a safety mechanism: the model declines requests that look dangerous. That approach cannot tell a bank’s red team, hired to attack its own systems, from an intruder asking the same question. OpenAI’s answer is to stop asking the model to make that judgment for this product. GPT-5.6-Cyber is described as intended for “authorized” work, and access runs through the Daybreak Cyber Partner program. In practice, a 95% completion rate means the protection is who gets in, not what the model will say.

    That has consequences for everyone involved. Security teams gain a tool that no longer argues with them. OpenAI takes on a gatekeeping role closer to that of a controlled-software vendor than a chatbot provider, which means vetting, monitoring and revoking access. Because the partner program is expanding in the same announcement, the strength of that admission process becomes the key variable in whether the model’s net effect favors defense.

    Why Refusals Were a Real Cost for Defenders

    OpenAI presents the release as a response to user friction. It said GPT-5.5-Cyber completed only 57.3% of these requests, and that GPT-5.6-Cyber addresses “feedback from security researchers who encountered persistent refusals with the earlier model.” It also said the general-purpose Sol model has a high refusal rate for penetration testing against production systems. Production systems are the live environments where real vulnerabilities, and real consequences, exist.

    For a professional tester, a tool that refuses four requests in ten is hard to build a workflow around. Engagements run on tight schedules, and every refusal means rephrasing, switching tools or doing the work by hand. Moving from 57.3% to 95% is the difference between an occasional assistant and one a team can rely on, assuming the output is sound. The security industry has long used offensive tooling for defensive ends, and this release applies that practice to frontier AI.

    What a 95% Completion Rate Does and Does Not Measure

    A completion rate measures willingness, not correctness. A model can complete an exploit-development request by producing code that does not compile, targets the wrong version or fails against a patched system. The published figures show GPT-5.6-Cyber attempts nearly everything it is asked. They do not show how often the attempts work.

    OpenAI also says the model has superior performance in developing arbitrary code execution exploits (attacks that let an intruder run their own commands on a target) and in finding and assessing known vulnerabilities and new zero-days. As reported, those claims come without the benchmark figures, test sets or comparison baselines that would let outside practitioners judge them. The comparative numbers that were published come from OpenAI’s own testing. That is normal for a product launch, but it means the capability case awaits independent evaluation and field results from partner organizations.

    What It Means for Infrastructure Operators

    Data centers, cloud platforms and network operators are high-value targets, and much of their defense depends on finding weaknesses before someone else does. If GPT-5.6-Cyber performs as described, vetted partners could run more penetration tests and vulnerability searches with the same staff. That could shorten the time between a weakness existing and someone noticing it.

    The same speed brings operational pressure. More findings mean more patches to prioritize, test and deploy, often on systems that cannot easily go offline. Operators whose vendors or security providers join the Daybreak program may receive more vulnerability reports, faster, and will need remediation processes that can absorb them. The benefit goes to organizations that can fix quickly, not only to those that can find quickly.

    Background

    OpenAI, the developer of ChatGPT, trains general-purpose AI models that it has increasingly adapted for specific professional fields. GPT-5.6-Cyber follows an earlier cybersecurity variant, GPT-5.5-Cyber, and is built on the general-purpose GPT-5.6-Sol. General-purpose AI models usually decline requests that resemble hacking, which protects against casual misuse but frustrates security professionals whose job is to attack systems with permission.

    The security industry has a long history of defenders using offensive techniques. Penetration testers and red teams are hired to break into an organization’s own systems, and vulnerability researchers look for flaws before criminals or state-backed groups do. Specialized AI models for this work raise the same question the industry has always faced with powerful offensive tools: who gets access, and under what controls.

    Sources

    Source: OpenAI Unveils New Cybersecurity Model GPT-5.6-Cyber (SecurityWeek). OpenAI launches a cybersecurity model for authorized work and expands its Daybreak Cyber Partner program.

  • Aalo’s 50 MW Pod Shows Small Nuclear’s AI Benchmark Is On-Site Gas

    Aalo’s 50 MW Pod Shows Small Nuclear’s AI Benchmark Is On-Site Gas

    TL;DR · 30-second read

    The Short Version

    Crusoe, a company that rents out computing power for artificial intelligence, will spend six months testing a small nuclear reactor from startup Aalo alongside one of its compact data centers.

    Aalo’s reactor is sized to match the gas-fired generators many of these sites already run on. The test is about whether it can keep the computers running reliably, even as their demand for electricity swings up and down.

    If it works, the two companies aim to start building for real in 2029, but first Aalo needs a green light from federal nuclear safety regulators.

    Crusoe, a cloud provider built specifically for AI workloads (often called a neocloud), and reactor developer Aalo Atomics will pair Aalo’s small modular reactor (SMR) technology with Crusoe’s modular Spark AI data center for a six-month trial, Fierce Network reported on August 11, 2026. The test will examine the reactor’s capacity factor, meaning how consistently it can run without downtime, along with its ability to follow the fluctuating power draw of AI workloads when combined with battery storage.

    Both companies are targeting commercial nuclear-powered AI data center deployments in 2029. Aalo’s first product is the Aalo One Reactor in a 50-megawatt Aalo Pod configuration. Commercial deployment in the United States depends on approval from the Nuclear Regulatory Commission.

    Executive Summary

    The Crusoe–Aalo trial is a small-scale engineering test, not a power purchase deal. The goal is to find out whether a factory-built reactor and a factory-built data center can be tightly coupled on one site. According to Aalo CEO Matt Loszak, this will be the first time the two types of infrastructure are so closely integrated. Crusoe co-founder Cully Cavness described the work as mostly electrical engineering: controls, monitoring, load profiles and fault conditions running in both directions.

    The strategic significance lies in how the reactor is sized. The large nuclear deals signed by the major hyperscalers mostly secure output from existing large plants. Aalo, by contrast, has built its 50-megawatt pod to fit the 10-to-50-megawatt range where on-site gas turbines currently power many data centers, and says it can make nuclear’s economics very competitive with natural gas. If that holds, small reactors would be competing for the on-site generator slot rather than against utility-scale nuclear. That claim is still a target and has not been demonstrated.

    The Benchmark Is a Gas Turbine, Not a Gigawatt Plant

    Hyperscaler nuclear deals have mostly involved large legacy plants that deliver power over the grid. Aalo is aiming at a different market. Loszak said many data centers are being powered by gas turbines and small modular gas generators in the 10-to-50-megawatt range, and that this is the reason Aalo chose a 50-megawatt pod. He described the design as a balance between economy of scale, which argues for larger units, and economy of number, which argues for mass-producing many identical ones. He added that a reactor as small as 1 megawatt would suffer on economics.

    This sizing decision identifies the real competitor. A 50-megawatt pod that can be stacked is meant to fill the same role in a site plan that on-site gas fills today: dedicated, behind-the-meter capacity, sized in blocks, added as a campus grows. It is not meant to replace a grid connection to a 1-gigawatt plant. The people affected are data center developers and neoclouds that currently use gas to get around grid interconnection delays. Crusoe is one of them, and Cavness said the company already pursues natural gas, solar and batteries as a portfolio.

    The comparison also sets the standard the technology must meet. On-site gas is a known quantity with established supply chains. Aalo’s statement that it can be very competitive with natural gas is a stated goal, and neither company has published cost-per-megawatt-hour figures to back it up. The benchmark is clear. Whether Aalo can hit it has not yet been shown.

    Why the Trial Is About Load Following, Not Reactor Physics

    Loszak said the first priority is capacity factor: can the reactor run for months with very limited downtime? The second priority is load following, meaning how well the power supply tracks demand as it changes. AI workloads are not flat. Loszak said Crusoe may simulate the different fluctuation patterns it sees across training and inference jobs. The setup will include battery storage so the companies can test how nuclear and batteries share the job of absorbing those swings.

    This is the practical challenge of tight coupling. When a data center draws from the grid, the grid absorbs its variability. When a reactor sits beside the load with no grid in between, any mismatch between supply and demand has to be handled on site, through controls, storage, or the reactor’s own response. Cavness’s emphasis on fault conditions in both directions points to the same problem: a disturbance on either the data center side or the reactor side has to be contained without taking down the other.

    This is also where the gas comparison is tested in practice. On-site generators are often chosen because operators can dispatch them to match load. A nuclear pod paired with batteries needs to show similar operational behavior before a buyer will treat it as a replacement. Six months of data on that pairing is exactly what a prospective buyer would want to see.

    Time Is the Cost Line Both Companies Are Trying to Cut

    Cavness described nuclear’s historic problem as time and money. Build timelines of a decade or more lead to cost overruns, which raise the price per kilowatt-hour. The proposed solution is modular, mass-manufactured units that bring down both the timeline and the cost. Aalo reached criticality, a sustained nuclear chain reaction, within three years of its founding, which is fast by the standards of the nuclear industry. Crusoe’s modular data center approach follows the same manufacturing-led thinking.

    The 2029 target is aggressive, and several dependencies stand in the way. Aalo needs Nuclear Regulatory Commission approval before any domestic commercial deployment. The companies also acknowledge they will need to win over local communities, which are increasingly skeptical of data centers even without a reactor involved. Their stated answer is careful site selection and public education. A successful trial would reduce engineering risk, but it would not settle licensing, siting or cost, and those are the factors that will determine whether small nuclear actually takes a share of the on-site power market from gas.

    Background

    Crusoe is a neocloud, a cloud provider focused on AI computing, that builds its own data centers, including its modular Spark design. It has pursued several energy sources, including natural gas, solar and batteries, to power those sites as grid capacity for data centers has become harder to secure. Aalo Atomics is a nuclear developer building factory-made reactor pods. Its first product is the Aalo One Reactor in a 50-megawatt Aalo Pod configuration, and the company reached criticality within three years of its founding.

    Nuclear power has become a major topic in data center planning. All three major hyperscalers have signed nuclear supply agreements, mostly with large existing plants. Small modular reactors are a newer, largely unproven category. Proponents argue they can avoid the decade-plus build times and cost overruns of traditional nuclear, but they have yet to demonstrate that at commercial scale.

    Sources

    Source: Neocloud Crusoe puts Aalo nuclear power to the AI data center test (Fierce Network), covering Crusoe and Aalo’s six-month trial pairing a small modular reactor with Crusoe’s Spark AI data center.

  • West Virginia’s 20-Year Data Center Plan Turns AI Site Selection Into a Power Project

    West Virginia’s 20-Year Data Center Plan Turns AI Site Selection Into a Power Project

    TL;DR · 30-second read

    The Short Version

    West Virginia wants the giant computer warehouses that power artificial intelligence, and it has none of them today.

    Its new 20-year plan sets the terms. Companies must supply their own electricity instead of drawing on the power households use, and they must pay for any grid upgrades themselves. Half the money the projects bring in is set aside to cut, and eventually end, the state income tax.

    The catch is that none of this has been tested yet. No company has signed on, and no one knows how much money will actually arrive.

    On August 11, 2026, West Virginia Governor Patrick Morrisey, Senate President Randy Smith, House Speaker Roger Hanshaw and Department of Commerce leadership announced the West Virginia Responsible Data Center Development Plan, the Office of the Governor said. The framework relies on House Bill 2014, the Power Generation and Consumption Act, to set state-level oversight and development criteria for hyperscale data centers, the very large facilities built to run cloud and artificial-intelligence workloads.

    The state says it starts with zero active hyperscale facilities. The plan is a 20-year strategy managed by a Data Economy Office. It requires developers to supply their own energy, and it dedicates 50 percent of project revenue to reducing the state personal income tax.

    Executive Summary

    West Virginia has published the rules it will use to attract AI-scale data centers. Before any hyperscale campus exists in the state, it has set four conditions. Developers must prove financial capacity, committed capital and clear power load projections to be certified. They must build, bring or procure their own energy and pay for their own grid and microgrid infrastructure. They must operate under a demand-response model that curtails grid use during peak stress. And project revenue is split by formula, with half going to income tax relief and none to the state general fund.

    The central design choice is the energy requirement. Many jurisdictions compete for data centers on land, tax incentives and fiber. West Virginia is asking developers to arrive with a power solution already in hand. That changes who can realistically build in the state and what a site proposal has to contain. It also shifts the risk of new generation away from existing utility customers, at least on paper.

    The release is strong on principles and light on mechanics. The revenue base, the enforcement of the ratepayer pledge, water targets and the advisory council’s standards are all still to come. The state’s projections of billions in investment and thousands of jobs have no figures behind them yet.

    Bring Your Own Power Is the Real Entry Fee

    The plan’s most consequential clause is the Ratepayer Protection Pledge. Developers must build, bring or procure their own energy resources and fully fund their own infrastructure and microgrid costs. A microgrid is a local power system that can run alongside the main grid or apart from it. The state says utility customers will not bear construction, operational or grid upgrade expenses. Certification also requires clear power load projections, meaning a forecast of how much electricity a facility will draw. As a result, a developer cannot secure a site first and look for power later. The energy plan is part of the application.

    This is what makes site selection in West Virginia a power project. An AI campus proposal there needs the same pieces as a generation development: a fuel or supply source, generation or procurement contracts, and the capital to build them. It also has to coordinate capacity with the local utility and comply with the reliability frameworks of PJM (the regional grid operator for much of the Mid-Atlantic and Midwest), FERC (the federal energy regulator) and NERC (the body that sets grid reliability standards). On top of that, facilities must operate under demand response. During peak stress periods they either switch to on-site backup generation or cut their consumption.

    The requirement is likely to favor well-capitalized developers and those with established generation partners. It is likely to filter out speculative land plays, which the release says it intends to reject. It also changes the conversation for AI tenants. Uptime at a West Virginia campus will depend partly on assets the developer owns and runs, not only on the utility. Demand response adds a further consideration: a facility must be able to curtail or self-supply during grid emergencies. Buyers will want to know how that interacts with their own service-level expectations.

    A Revenue Formula Designed for Voters, Not Just Developers

    The state has specified exactly where project revenue goes. Fifty percent is dedicated to reducing and ultimately eliminating the state personal income tax. Thirty percent goes to host counties for schools and government operations. Ten percent is spread across all other counties, and ten percent funds local water, wastewater and electrical upgrades. None enters the general revenue fund, and host county budgets cannot be negatively affected.

    This structure answers a common local objection to data centers, which is that the host community carries the disruption while the benefits flow elsewhere. Tying half of the revenue to a statewide income tax cut also builds a constituency for the program beyond host counties. Whether the income tax goal is realistic depends on figures the state has not published. The plan does not say what counts as project revenue under the High Impact Data Center designation. It also gives no estimate of how much revenue a typical facility would generate. With zero hyperscale facilities today, the tax promise is a long-horizon aspiration rather than a budget line.

    A Clean Slate, and an Unfinished Rulebook

    The release says that before HB 2014, more than 50 West Virginia counties had no mechanism to approve, condition or deny large data center projects. That governance gap exists in many places that have seen sudden data center interest. Closing it at the state level before the first hyperscale project arrives is the plan’s clearest structural advantage. Developers get one certification path. Communities get a screening step that did not exist before.

    Much of the detail is still to be written. A Data Center Advisory Council will recommend standards for setbacks, noise, site security and proximity to sensitive areas. On water, the state keeps existing oversight and says it is pioneering closed-loop systems, immersion cooling and reclaimed water. It aims to lead on Water Usage Effectiveness, a measure of water consumed per unit of computing energy. It has not set a numeric target or a requirement. Two claims in the release are also unquantified: that development will help modernize the grid and drive down residential rates over time, and that it will bring billions in investment and thousands of jobs. Both are reasonable goals, and neither is substantiated yet.

    Background

    West Virginia has long been an energy-producing state, and it sits inside the PJM grid region. That region’s reliability and capacity rules shape how any large new electricity user connects. The state enters the AI data center race without a single hyperscale campus. Until House Bill 2014, more than 50 of its counties had no formal mechanism to approve, condition or deny such projects.

    The August 2026 plan builds on that law. It combines a Data Economy Office, a certification process for developers, a fixed revenue-sharing formula and a requirement that developers supply their own energy. Governor Morrisey described the approach as leading without repeating the mistakes of other states, which he did not name.

    Sources

    Source: Governor Morrisey, Legislative Leaders Announce Unified Plan for Responsible Data Center Development in West Virginia, the West Virginia Office of the Governor’s announcement of the state’s 20-year data center framework under House Bill 2014.

  • IBM’s $240M Together AI Deal Treats AI Inference Like Reserved Capacity

    IBM’s $240M Together AI Deal Treats AI Inference Like Reserved Capacity

    TL;DR · 30-second read

    The Short Version

    IBM, the century-old computing company, has signed a $240 million, multi-year deal with Together AI, a younger firm that runs freely available artificial intelligence programs for other businesses.

    IBM’s cloud, meaning its rentable computers in data centers, will host a large set of powerful Nvidia chips set aside for one job: answering people’s questions to those programs, the everyday work of a chatbot.

    The surprising part is how it is bought. Instead of paying per use, the computing power is booked years ahead, like leasing a whole building instead of renting hotel rooms by the night.

    IBM and Together AI have signed a $240 million multi-year agreement for an Nvidia-powered AI inference cluster that will run on IBM Cloud, Reuters reported on August 11, 2026. Inference is the stage where a trained AI model answers requests, as opposed to training, where the model is built.

    Together AI operates a cloud platform that serves open-weight AI models, meaning models whose parameters are publicly released, to developers and enterprises. Under the deal, a dedicated cluster of Nvidia accelerators on IBM’s infrastructure is committed to that inference work for several years.

    Executive Summary

    The headline number is $240 million, but the structure matters more. This is not a customer swiping a card for GPU hours. It is a multi-year commitment to a specific, dedicated cluster, which is how data center space and power have long been contracted, and increasingly how AI compute is contracted too.

    For IBM, the deal places IBM Cloud in the business of hosting third-party AI inference at scale, a market where the large hyperscalers and specialist GPU clouds have been most visible. For Together AI, it secures a block of capacity for serving open models at a time when access to Nvidia hardware still depends heavily on who has committed to it in advance.

    The companies have not disclosed the hardware generation, the size of the cluster, where it sits, or how the $240 million is paid over time. Those details determine whether this is a modest capacity reservation or a meaningful new source of inference supply.

    Inference Is Being Booked Like a Lease

    Most people meet AI inference as a metered service: a developer sends a prompt, a provider returns an answer, and the bill is calculated per token, a token being a small chunk of text. That pricing looks elastic, but the hardware underneath it is not. Nvidia accelerator clusters are expensive to buy, take time to deliver and install, and lose value as each new chip generation arrives. Someone has to commit to them before the first token is served.

    A $240 million, multi-year contract for a dedicated cluster is that commitment made explicit. It shifts inference from the on-demand model of cloud computing toward the reserved-capacity model long familiar from colocation and wholesale data center leasing, where a tenant contracts a defined block of infrastructure for a defined term. The provider gets revenue visibility that justifies deploying hardware; the tenant gets guaranteed supply it can resell to its own customers with predictable performance.

    This is one deal, and it does not prove an industry-wide shift on its own. But it shows clearly that open-model inference, often described as a commodity service racing to the lowest per-token price, is being underwritten by the same kind of long-dated capacity contracts as training clusters. The per-token price a developer sees sits on top of a fixed, multi-year obligation somewhere in the stack.

    What IBM Gains by Hosting Someone Else’s Models

    IBM has spent recent years positioning around open and hybrid AI: its Granite models are openly released, and its Red Hat business is built on open-source software. Hosting a platform whose entire proposition is serving open-weight models fits that posture more naturally than competing head-on with the largest clouds for frontier-lab training contracts.

    It also gives IBM Cloud a substantial, recurring AI workload with an anchor tenant. For a cloud provider, a committed customer reduces the risk of buying accelerators that might otherwise sit underused. For enterprise buyers who already run on IBM infrastructure, the arrangement could, in principle, place open-model inference closer to their existing systems, though neither company has described how enterprise customers would access the cluster.

    The Risk Sits in the Utilisation Curve

    Reserved capacity moves risk; it does not remove it. Inference demand can shift quickly as new models arrive, and per-token prices for open models have been falling as providers compete. A multi-year commitment to a fixed pool of hardware exposes whichever party carries it to two pressures: whether the cluster stays busy, and whether the hardware stays competitive as Nvidia releases newer, more efficient generations.

    There is also a physical side. Dense Nvidia clusters draw significant power and increasingly require liquid cooling, which constrains which data halls can host them and how quickly. The value of a contract like this depends on the cluster actually being energised and in service on schedule, and none of those milestones have been published.

    Background

    IBM has operated a public cloud for more than a decade and in recent years has concentrated its AI strategy on enterprise and hybrid deployments, including its watsonx platform, its openly released Granite models and the open-source software business of Red Hat. IBM Cloud offers Nvidia GPU infrastructure alongside its traditional enterprise services.

    Together AI emerged during the generative AI boom as a platform for running open-weight models, giving developers an alternative to closed models from the largest AI labs. Like other inference providers, it depends on securing large volumes of Nvidia accelerators, whose supply has been shaped by long-term commitments from cloud providers and AI companies.

    Sources

    Source: IBM, Together AI ink $240 million deal for Nvidia-powered AI inference cluster (Reuters): report on a multi-year agreement for a dedicated Nvidia inference cluster on IBM Cloud.