Author: Deepak Jain

  • Iran-Linked Cyberattack Forces UK Power Plant Offline: A Wake-Up Call for OT Security

    Iran-Linked Cyberattack Forces UK Power Plant Offline: A Wake-Up Call for OT Security

    A small power plant in the United Kingdom was taken offline following a cyberattack that has been linked to Iran, according to a report by The Telegraph carried by CNBC on July 6, 2026. The facility’s name, capacity, and the duration of the shutdown were not disclosed in the report.

    If confirmed, the incident would join a very short list of cyberattacks anywhere in the world that have resulted in the loss of physical power-generation capacity — a category of event that grid operators and security agencies have long warned about but rarely seen materialize.

    Executive Summary

    According to the reporting, hackers attributed to Iran compromised systems associated with a small UK generating facility, and the plant was subsequently shut down. That one sentence contains nearly everything that is publicly known — and that brevity is itself significant. Neither the operator, the attack method, nor the official basis for the Iran attribution has been made public in the source material.

    Why it matters: the vast majority of cyberattacks on energy companies hit their corporate IT — email, billing, customer data. What makes this report notable is the claimed crossing into the physical domain, where an intrusion ends with turbines stopping rather than data leaking. Confirmed cyber-physical grid incidents are so rare that the canonical examples remain the 2015 and 2016 attacks on Ukraine’s grid. A confirmed case in the UK, a G7 economy with mature critical-infrastructure regulation, would mark a meaningful escalation in what operators must plan for.

    For the infrastructure industry — utilities, data center operators, and anyone whose business depends on reliable power — the practical takeaway does not depend on the attribution being right. The incident, as described, is a live test of assumptions about how well operational technology is separated from the internet-facing systems attackers can reach.

    From Stolen Data to Stopped Turbines

    Security professionals draw a sharp line between IT (information technology — the email servers, databases, and laptops every company runs) and OT (operational technology — the industrial control systems that open valves, spin generators, and switch breakers). Attacks on energy-sector IT are routine; attacks that reach OT and cause physical consequences are exceptionally rare, because control systems are typically segmented from corporate networks and because causing physical effects requires specialized knowledge of industrial equipment.

    The report does not say whether the attackers actually manipulated control systems, or whether the operator shut the plant down as a precaution after detecting an intrusion elsewhere. That distinction matters enormously. A precautionary shutdown means defenses worked as designed — disruptive, but contained. Direct manipulation of control systems would put the incident in the same category as Ukraine 2015, where attackers remotely opened breakers and blacked out roughly a quarter-million customers. Until the mechanism is disclosed, both readings remain open, and honest analysis has to hold them both.

    Attribution Is a Claim, Not Yet a Conviction

    The Iran link originates with The Telegraph’s reporting rather than, so far as the source material shows, a formal government attribution. Cyber attribution is genuinely hard: attackers reuse each other’s tools, route through third countries, and sometimes deliberately imitate rival groups. Western agencies have previously documented Iranian-linked activity against industrial control systems — including the 2023 compromises of Unitronics controllers at US water utilities — so the claim is plausible. Plausible, however, is not proven, and the geopolitical stakes of naming a state actor make the evidentiary bar higher, not lower.

    Fair questions cut in every direction here. What forensic indicators support the Iran link, and will the UK’s National Cyber Security Centre confirm it? Equally, if the attribution is later walked back, was the initial linkage sourced from officials, from the operator, or from third-party researchers? Early attribution reporting on infrastructure incidents has a mixed track record — the 2019 claims around a US grid ‘attack’ that turned out to be a firewall flaw are a cautionary example — which is reason for patience, not dismissal.

    Why Small Plants Are the Soft Underbelly

    It is no accident that the target described is a small power plant. Large transmission operators and major generators sit under heavy regulatory scrutiny and can amortize security operations centers across billions in revenue. Small generators — peaking plants, biomass and waste-to-energy sites, independent operators — run thin staffs, often rely on remote-access links for vendor maintenance, and operate control equipment that predates modern security design. They are individually low-value targets but collectively numerous, and in an increasingly decentralized grid their aggregate capacity matters.

    The economics are unforgiving: a security program that is table stakes for a gigawatt-scale utility can be a material fraction of a small plant’s operating budget. That gap is precisely where regulation, insurance requirements, and shared-service security models will be contested in the years ahead. An incident like this one strengthens the argument that minimum OT-security standards need to reach the long tail of generation, not just the giants.

    What Operators — Including Data Centers — Should Take From This

    For data center and cloud operators, this story is about the other side of the meter. Facilities that promise 99.999% availability model grid failure as a weather or equipment problem; a world where generation can be taken offline by remote adversaries changes the risk calculus for utility redundancy, on-site generation, and fuel reserves. It also lands amid record data-center-driven load growth, which is already straining grid planning in the UK and elsewhere.

    For anyone running OT: the defensive playbook this incident points to is well established, if unevenly applied — rigorous segmentation between IT and OT networks, multi-factor authentication on every remote-access path, monitoring inside the control network rather than only at its edge, and rehearsed manual-operation procedures so a plant can run or shut down safely when its digital systems cannot be trusted. None of that is exotic. The persistent gap is investment and follow-through, and events like this are what close it.

    Background

    Power plants and grid operators have digitized steadily over three decades, layering remote monitoring and control onto industrial equipment that was designed long before modern cyber threats. Security agencies have warned since at least the Stuxnet operation of 2010 — which physically damaged Iranian centrifuges via malicious code — that industrial control systems can be weaponized, but confirmed grid consequences have remained rare: the 2015 and 2016 Ukraine blackouts are the textbook cases.

    The UK regulates its critical energy infrastructure under the NIS Regulations of 2018, with the National Cyber Security Centre as technical authority, and both UK and US agencies have repeatedly warned of Iranian-linked interest in Western critical infrastructure amid broader geopolitical tensions. A confirmed cyber-induced plant shutdown on British soil would be the first incident of its kind publicly acknowledged in the country.

    Source: Small UK power plant shut down after cyberattack linked to Iran: Telegraph — CNBC’s July 6, 2026 report of The Telegraph’s account of an Iran-linked cyberattack that forced a small UK power plant offline.

  • Sysdig Documents First Fully Autonomous AI-Agent Ransomware Attack

    Sysdig Documents First Fully Autonomous AI-Agent Ransomware Attack

    Security vendor Sysdig has reported what it characterizes as the first documented instance of a ransomware attack executed end-to-end by an autonomous AI agent, according to a July 5, 2026 write-up in The HIPAA Journal. In this framing, the agent — not a human operator following a runbook — made the tactical decisions from initial access through encryption.

    The claim is being circulated widely because it marks a symbolic threshold in the offensive use of large language model-based agents, systems that can chain tools, reason about goals, and take multi-step actions with limited human oversight.

    Executive Summary

    The announcement, as relayed by The HIPAA Journal, positions Sysdig’s finding as a landmark in cybersecurity: an intrusion in which an AI agent, rather than a human ransomware operator, drove the attack chain. That is a meaningful shift in threat modeling. Where traditional ransomware crews rely on human affiliates to move laterally, escalate privileges, and stage encryption, an autonomous agent could theoretically compress those stages into machine time and run them in parallel across many victims.

    For infrastructure operators — data centers, cloud tenants, connectivity providers, and their customers — the practical implication is that assumptions built around human attacker tempo may need revisiting. Runbooks that count on hours of dwell time to detect and evict an intruder become weaker when the intruder is a piece of software that never sleeps and does not tire of retrying.

    That said, the summary made available in this feed is thin. The claim of “first fully autonomous” is a strong one, and the industry should read the underlying Sysdig research carefully before treating the milestone as settled fact rather than a plausible and important report.

    Why “Autonomous” Is The Word That Matters

    Ransomware crews have used automation for years — mass scanners, exploit kits, off-the-shelf loaders. What Sysdig is reportedly describing is different in kind: an AI agent that plans and adapts rather than executing a fixed script. In agent architectures, a language model is given a goal, a set of tools (shell access, network utilities, credential stores) and permission to iterate until it succeeds or gives up. If the report holds up, the notable step is not that malware ran on its own, but that decision-making — normally the human’s contribution — was delegated to software.

    The distinction matters because defenders have historically exploited the human bottleneck. Every hour an operator spends deciding what to do next is an hour a SOC can use to detect them. Autonomous agents narrow that window.

    Economics: Scaling Attacks Without Scaling Headcount

    Ransomware is a business, and its unit economics are constrained by affiliate labor. Recruiting, vetting, and paying human operators is expensive and risky for the crews at the top of the pyramid. An autonomous agent, if it works reliably, lowers that cost floor. The same operator could in principle run many concurrent intrusions, each customized to the victim environment, without a proportional increase in staff.

    The flip side is reliability. Language model agents are known to hallucinate, loop, and make confidently wrong choices. Whether Sysdig’s observed agent achieved its objective through skill or luck is the kind of detail that separates a novelty from a business model. The public summary does not settle that question.

    Implications For Infrastructure Buyers

    For enterprises buying cloud, colocation, and connectivity, the near-term takeaway is not panic but pressure on already-known controls. Identity hygiene, least-privilege access, tested backups, egress monitoring, and behavioral detection at the workload layer — the fundamentals Sysdig itself sells into — matter more, not less, if attacker tempo increases. Providers that offer runtime detection, immutable backups, and rapid isolation of compromised workloads have a clearer story to tell.

    There is also a governance dimension. If an attack is driven by an AI agent, questions of attribution, evidence preservation, and even insurance coverage become murkier. Incident responders will want to capture not just the malware artifacts but the agent’s prompt history, tool calls, and model provenance where possible.

    Reading The Claim Fairly

    “First” claims in security are notoriously hard to verify. Autonomous or semi-autonomous offensive tooling has been demonstrated in research settings and hinted at in underground forums for at least two years. Sysdig may well have observed the first in-the-wild case that meets a strict definition of full autonomy, but the industry should ask what that definition is: Did a human select the target? Approve the ransom demand? Handle negotiation? Each answer changes how landmark the milestone really is.

    None of that diminishes the direction of travel. Whether this specific case is the first or the fifth, agent-driven intrusions are a plausible near-term trajectory, and treating the report as a prompt to stress-test defenses is a reasonable response even before every detail is independently confirmed.

    Background

    Ransomware has evolved over the past decade from opportunistic file-encrypting malware into an organized affiliate economy, in which core developers license their tooling to human operators who conduct intrusions and split proceeds. Detection and response strategies have been built largely around the pace and habits of those human affiliates.

    In parallel, the rise of large language models has produced “agent” frameworks that let AI systems use tools, browse, execute code, and pursue goals across many steps. Security researchers have warned since at least 2024 that the same capabilities that make agents useful for legitimate automation make them attractive for offensive operations. Sysdig’s reported finding, if it holds up to scrutiny, marks the point at which that warning moves from theory into documented practice.

    Source: AI Agent Conducts First Fully Autonomous Ransomware Attack – The HIPAA Journal — reporting on Sysdig’s research documenting what it describes as the first end-to-end ransomware intrusion driven by an autonomous AI agent.

  • Anthropic’s $19B TeraWulf Lease Reroutes Miner Into AI Landlord

    Anthropic’s $19B TeraWulf Lease Reroutes Miner Into AI Landlord

    Anthropic, the AI lab behind the Claude model family, has signed a data center lease valued at roughly $19 billion with TeraWulf (Nasdaq: WULF), a bitcoin miner that has been repositioning itself as an AI infrastructure host. The agreement was reported by SiliconANGLE on July 5, 2026.

    The transaction makes Anthropic a long-duration anchor tenant on TeraWulf’s power-rich footprint, and it ranks among the largest single AI hosting commitments disclosed to date.

    Executive Summary

    The headline number — about $19 billion — is what an AI lab would normally spend building its own campus, not renting one. By pushing that spend into a lease with a listed bitcoin miner, Anthropic is trading capex for speed: TeraWulf already controls interconnected sites and substation capacity, which is the scarce input in the current AI build-out.

    For TeraWulf, the contract is a category change. A company whose revenue has been tied to bitcoin’s price now has a multi-year, investment-grade-style cash flow tied to a frontier AI customer. That is why WULF sits on many investor watchlists as a proxy for the miner-to-AI-landlord thesis.

    The deal also sharpens a broader trend: hyperscalers and AI-native labs are no longer waiting on traditional colocation supply. They are contracting directly with whoever holds the two things that matter most right now — energized land and a grid connection.

    Why an AI Lab Rents from a Bitcoin Miner

    Bitcoin miners spent the last cycle acquiring the exact ingredients AI now needs: cheap power contracts, substation rights, and shells that can dissipate very high rack densities. Retooling those shells for GPUs is non-trivial — liquid cooling, tenant-grade redundancy, and network fiber all have to be added — but it is far faster than greenfield permitting. For Anthropic, leasing from TeraWulf compresses time-to-first-megawatt in a market where a new build can take three to five years.

    The economics also matter. A lease shifts risk: Anthropic pays for capacity as it is delivered rather than tying up cash in construction, while TeraWulf finances the fit-out against a signed contract. That is the same playbook enterprise tenants use with traditional colocation providers; what is new is the scale and the counterparty.

    What $19 Billion Actually Buys

    The release frames the commitment as a lease value rather than an upfront payment, which typically means it spans many years of rent, power pass-through, and services. Without disclosed megawatts, PUE assumptions, or a term length, the figure is best read as a ceiling on Anthropic’s obligation and a floor on TeraWulf’s backlog — not a check written on day one.

    Even so, a nine- or ten-figure annualized run-rate at a single landlord is unusual. It implies gigawatt-class ambitions over the life of the contract, which in turn implies transmission upgrades and generation additions that neither party controls alone.

    Winners, Losers, and the Miner-to-AI Trade

    The clearest winner is any miner sitting on energized capacity in a utility territory friendly to large loads. TeraWulf’s deal will be used as a comparable by peers negotiating their own AI conversions, and it validates the equity story that has driven the miner-to-AI rerating. The clearest pressure point is on traditional wholesale data center developers, who now face a well-funded competitor class that already owns the power.

    For Anthropic, the strategic read is independence. Locking in dedicated capacity outside the big three clouds gives the company optionality on where its next generation of models trains and serves, and reduces the risk that compute becomes a chokepoint controlled by a strategic investor or competitor.

    The Grid Question Behind the Deal

    Every large AI lease today is really a bet on the interconnection queue. Utilities in the regions where miners cluster — parts of Appalachia, Texas, and the upper Midwest — are already signaling multi-year waits for new large-load connections. A lease of this scale will draw scrutiny from regulators, ratepayer advocates, and neighboring loads who compete for the same megawatts.

    None of that is a criticism of either party; it is the operating reality of the market. But it means execution risk on a deal of this size sits less with the tenant or the landlord than with transmission planners and permitting timelines that neither company can accelerate on its own.

    Background

    Anthropic, founded in 2021, has grown into one of a small group of frontier AI labs whose compute needs now rival those of the largest cloud tenants. Like its peers, it has relied on hyperscaler partners for training capacity while seeking to diversify its infrastructure footprint.

    TeraWulf emerged from the last bitcoin cycle with a portfolio of power-anchored sites in the eastern United States. As mining economics compressed and AI compute demand surged, the company — along with several listed peers — began marketing its energized capacity to high-performance computing and AI tenants, a pivot investors have tracked closely under the miner-to-AI-landlord thesis.

    Source: Anthropic inks $19B AI data center lease with TeraWulf – SiliconANGLE — report on Anthropic’s multi-billion-dollar hosting agreement with the Nasdaq-listed bitcoin miner.

  • Texas Bets on 765 kV Lines to Power the Next Wave of AI Data Centers

    Texas Bets on 765 kV Lines to Power the Next Wave of AI Data Centers

    Texas has committed to building out its grid with 765 kilovolt (kV) transmission lines — the highest-capacity class of overhead power line used in North America — in a strategy Data Center Knowledge summarized on July 5, 2026 as “build the wires, the AI will follow.” Rather than waiting for AI data center projects to sign up first, the state’s approach is to construct extra-high-voltage backbone capacity in anticipation of that demand arriving on the ERCOT grid.

    Executive Summary

    The decision reported here is less about a single project than about a planning philosophy. Historically, most U.S. transmission has been built reactively: a large customer or generator commits, studies are run, and wires follow years later. Texas is inverting that sequence at the 765 kV level — the class of line capable of moving several times the power of the 345 kV circuits that have long formed the backbone of ERCOT, the grid operator serving most of Texas.

    Why it matters: access to power has become the single biggest constraint on AI data center siting. A state that can credibly promise deliverable gigawatts on a known timeline gains a decisive edge in attracting capital-intensive AI campuses. But anticipatory building also shifts risk — if the forecast load arrives late, smaller than expected, or somewhere else, the cost of underused infrastructure lands on someone, and that someone is usually the ratepayer.

    Why 765 kV Is a Statement, Not Just a Specification

    Voltage class is the freeway-versus-farm-road question of the power grid. A 765 kV line can carry far more power than a 345 kV line over the same corridor, with proportionally lower electrical losses, which means fewer parallel lines, fewer towers, and less land consumed per delivered gigawatt. For a grid staring at data center campuses that each want hundreds of megawatts — sometimes a gigawatt or more — 765 kV is the only overhead technology that comfortably matches the scale of the ask.

    Choosing it is also a signal. 765 kV projects take longer to permit and build, require specialized transformers with notoriously long lead times, and cost more up front than incremental 345 kV additions. A jurisdiction that standardizes on 765 kV is telling the market it expects load growth measured in tens of gigawatts, not incremental upticks — and that it intends to be structurally ready rather than perpetually catching up.

    The Economics of Building Ahead of Demand

    The core bet is that transmission, not land or fiber, is now the scarce input for AI infrastructure. Interconnection timelines — the queue a new large customer or generator waits in before it can plug into the grid — have stretched to years across much of the country. Every month of waiting is a month of idle capital for an AI developer whose chips depreciate quickly. If Texas can compress that wait by having backbone capacity already energized, it converts grid readiness directly into economic development.

    The counterargument is forecast risk. AI load projections are among the most volatile numbers in the utility industry right now: they depend on chip supply, model efficiency gains, corporate capital cycles, and siting decisions that can pivot on a single tax incentive. Building wires for demand that hasn’t signed contracts means the state is, in effect, underwriting a demand forecast. If the forecast is right, the infrastructure looks prescient. If it’s wrong, Texas will have built expensive capacity whose carrying costs must still be recovered.

    Winners, Losers, and Who Carries the Risk

    The clearest winners are large-load customers — AI and cloud data center developers — who gain siting certainty, and the transmission utilities and equipment suppliers who get a multi-year construction pipeline. Landowners along new corridors face the familiar friction of routing and easement disputes, which 765 kV’s larger towers can intensify even as its higher capacity reduces the total number of corridors needed.

    The pivotal question is cost allocation. In ERCOT, transmission costs have traditionally been spread across consumers, which works when new load broadly benefits everyone but becomes contentious when the driver is a handful of very large private customers. Whether Texas requires AI-scale loads to shoulder a larger, more direct share of the wires built substantially for them — through contribution requirements, minimum-take commitments, or special rate classes — will determine whether this build-out is remembered as smart industrial strategy or as a subsidy from households to hyperscalers. The source piece frames the bet; it does not settle who holds the downside.

    What It Means Beyond Texas

    Other states and grid operators are watching, because Texas is running the experiment they have avoided: proactive, speculative, extra-high-voltage expansion in a market famous for moving faster and regulating lighter than its peers. If the wires fill up with AI load on schedule, expect copycat programs and renewed pressure on slower-moving regional planning processes elsewhere. If they don’t, the episode will become the cautionary tale cited in every future transmission docket.

    For the data center industry itself, the message is immediate: power-first siting is now official policy in at least one major market. Developers comparing regions will increasingly weigh not just today’s available megawatts but a grid’s demonstrated willingness to build ahead of them — and Texas has just bid aggressively on that dimension.

    Background

    Texas operates most of its grid through ERCOT, a system largely separate from the rest of the U.S., which allows the state to plan and permit infrastructure faster than regions governed by multi-state processes. That autonomy, combined with abundant land and energy resources, has already made Texas one of the country’s fastest-growing data center markets. The backbone of the ERCOT grid has long been built at 345 kV; standardizing new backbone corridors at 765 kV represents a step-change in the scale of power the state is preparing to move.

    The backdrop is the AI infrastructure boom: since the early 2020s, demand from AI training and cloud computing has transformed electricity access from a routine utility matter into the decisive factor in where billions of dollars of data center capital lands. Grid operators nationwide have struggled with long interconnection queues — the waiting line for new large loads and generators — and Texas’s 765 kV program is a direct attempt to turn that bottleneck into a competitive advantage.

    Source: Texas’ 765 kV Decision: Build the Wires, the AI Will Follow — Data Center Knowledge’s July 5, 2026 report on Texas’s anticipatory extra-high-voltage transmission strategy for AI data center growth.

  • Galaxy’s Helios Phase I Delivers 133 MW of AI Capacity to CoreWeave

    Galaxy’s Helios Phase I Delivers 133 MW of AI Capacity to CoreWeave

    Galaxy announced on July 5, 2026 that it has completed Phase I of its Helios data center campus in West Texas, delivering 133 megawatts (MW) of critical IT load to CoreWeave, the AI-focused cloud provider. Critical IT load refers to the power available to the computing equipment itself — servers and GPUs — as distinct from the total power a facility draws for cooling and other overhead.

    The completion converts a site that began life as a Bitcoin mining campus into dedicated AI infrastructure under Galaxy’s long-term lease arrangement with CoreWeave, one of the most prominent examples of the crypto-to-AI conversion trend reshaping the data center market.

    Executive Summary

    Galaxy, the digital assets and data center infrastructure firm, has finished the first phase of its Helios campus buildout and handed over 133 MW of critical IT load to its anchor tenant CoreWeave. Phase I completion moves the project from promise to delivery: Helios is now an operating revenue-generating AI data center rather than a conversion story on a slide deck.

    The milestone matters beyond Galaxy. Helios is the flagship test case for whether former cryptocurrency mining sites — which come with grid interconnections and power contracts already in place — can be economically retrofitted to the far more demanding standards of AI training and inference infrastructure. Delivering a first phase at this scale suggests the model can work, at least for sites with strong power positions.

    For CoreWeave, the delivery adds substantial contracted capacity at a time when access to powered land and energized shells — not GPUs — is widely seen as the binding constraint on AI cloud growth.

    Why Crypto Sites Became AI Real Estate

    The most valuable asset in data center development today is not land or buildings but secured power: a grid interconnection agreement and the megawatts behind it. Bitcoin mining operators spent the late 2010s and early 2020s locking up exactly that, often in low-cost power markets like West Texas. When AI demand exploded, those interconnections became worth far more serving GPUs than mining rigs, because AI tenants sign long-term leases at data center economics rather than riding volatile crypto margins.

    Galaxy’s Helios campus, acquired from a Bitcoin mining operator, is the highest-profile execution of that arbitrage. The conversion is not trivial — AI facilities require far denser power delivery, liquid or advanced air cooling, and enterprise-grade redundancy that mining sites never needed — but the timeline still beats greenfield development, where new grid interconnection requests can queue for years.

    What 133 MW Actually Buys

    133 MW of critical IT load is a substantial block of capacity by any historical standard — a few years ago it would have ranked among the larger single-tenant deployments in the world. In the AI era it is best understood as a first tranche: large frontier training clusters are increasingly specified in the hundreds of megawatts, and operators including Galaxy have discussed multi-phase expansion at Helios well beyond Phase I.

    Because the load is contracted to a single tenant, the economics resemble a triple-net real estate deal more than a retail colocation business: predictable lease revenue over a long term, with Galaxy carrying development and delivery risk and CoreWeave carrying utilization risk. That structure has become the dominant template for AI data center finance because lenders can underwrite the lease.

    Winners, Losers, and the Competitive Field

    The clearest winners are holders of energized or near-energized power positions — converted mining sites, utilities with spare interconnection capacity, and developers who queued early. CoreWeave benefits by adding capacity faster than greenfield timelines would allow, supporting its competition with hyperscale clouds for AI workloads. The pressure lands on developers still waiting in interconnection queues, and on regions whose grids cannot absorb gigawatt-class requests.

    The open competitive question is durability. Conversion sites tend to sit in remote, power-rich locations, which suits training workloads that tolerate latency. If the market shifts toward inference — which favors proximity to users — the value of remote megawatts could be repriced. Phase I’s completion answers the execution question; it does not settle the location question.

    Background

    Helios began as one of the larger Bitcoin mining campuses in the United States before Galaxy acquired the site and redirected it toward AI and high-performance computing. Galaxy subsequently signed long-term lease agreements making CoreWeave the campus’s anchor tenant, with capacity to be delivered in phases — Phase I, now complete, being the first.

    The conversion sits inside a broader industry shift: as demand for AI compute outran the pace of new grid connections, sites with existing power infrastructure — many of them crypto mining facilities in Texas and the Mountain West — became prime targets for repurposing. Helios is widely watched as the leading proof point for whether that playbook delivers at scale.

    Source: Galaxy Completes Phase I of Its Helios Data Center Campus, Delivering 133 Megawatts of Critical IT Load to CoreWeave — PR Newswire press release, July 5, 2026, announcing Phase I completion at Galaxy’s West Texas AI campus.

  • JadePuffer: What the First Fully LLM-Driven Ransomware Attack Signals

    JadePuffer: What the First Fully LLM-Driven Ransomware Attack Signals

    Security publication Dark Reading has reported on JadePuffer, an incident it characterizes as the first complete ransomware attack driven end-to-end by a large language model (LLM) — the AI technology behind chatbots and coding assistants. The report, published July 5, 2026, frames JadePuffer as a milestone: not malware that merely used AI for one task, but a campaign in which the AI itself reportedly orchestrated the attack.

    Executive Summary

    According to the Dark Reading report, JadePuffer represents a threshold the security industry has warned about for several years: ransomware in which a large language model does not just assist a human operator but drives the attack itself. If the characterization holds up, the distinction matters enormously. AI-assisted crime scales with the number of human criminals; AI-driven crime scales with compute.

    Details available at publication remain limited to the report’s central claim, so the responsible reading is twofold. First, the trajectory it describes is consistent with what researchers have documented publicly — proof-of-concept AI-powered ransomware and confirmed criminal misuse of commercial AI tools both surfaced well before this report. Second, “first” and “fully LLM-driven” are strong claims that deserve independent technical corroboration before the industry treats them as settled fact. Either way, the operational lesson for enterprises and infrastructure operators is the same: plan for adversaries whose speed and volume are no longer bounded by human labor.

    From AI-Assisted to AI-Driven Is a Difference in Kind

    Criminals have used AI for years to write phishing emails, debug malicious code, and research targets — but a human stayed in the loop, making decisions at each step. What the JadePuffer report describes is categorically different: an LLM reportedly executing the ransomware kill chain — reconnaissance, intrusion, data theft, encryption, and extortion — as an autonomous agent. In practical terms, that is the criminal application of the same “agentic AI” pattern legitimate businesses now use to automate customer service and software development.

    The precedent did not appear from nowhere. Security researchers had previously demonstrated proof-of-concept ransomware that used an LLM to generate its attack logic on the fly, and AI vendors have publicly disclosed catching threat actors abusing their models for extortion operations. JadePuffer, as reported, would move that trajectory from lab demonstrations and AI-augmented crews to a fully automated operation in the wild.

    The Economics Shift in the Attacker’s Favor

    Ransomware has always been constrained by skilled labor. Ransomware-as-a-service — the criminal franchise model where developers rent tools to affiliates — was itself an answer to that constraint, and it still required capable humans to run intrusions. An LLM-driven attack removes that bottleneck. The marginal cost of one more victim falls toward the price of compute and API calls, and a single operator could in principle run campaigns that once required a team.

    That reshapes the target landscape. Human-operated ransomware gravitates toward victims worth the effort — large enterprises, hospitals, critical infrastructure. Automation makes small and mid-sized organizations, historically protected partly by being unprofitable to attack individually, economically viable at scale. It also compresses time: an autonomous agent can move from initial access to encryption faster than human incident responders can convene a call.

    Defense Becomes a Machine-Speed Problem

    For defenders, the implication is uncomfortable but clarifying. Signature-based detection — recognizing known malicious files — was already fading; an LLM that generates or adapts its tooling per victim can present a novel artifact every time. The durable signals are behavioral: unusual data movement, anomalous credential use, encryption activity, and network patterns that no rewrite of the malware can fully disguise. Detection and response pipelines that depend on a human analyst approving each containment step will struggle against an adversary operating at machine speed.

    This is also an infrastructure story. Autonomous attacks still need identities to hijack, networks to traverse, and data to reach — so the fundamentals compound in value: segmented networks, phishing-resistant multifactor authentication, least-privilege access, and immutable, regularly tested backups kept isolated from production. Offline, verified backups remain the one control that converts a ransomware catastrophe into an outage. Providers of data center, connectivity, and security services should expect customer demand to tilt toward exactly these capabilities.

    Strong Claims Deserve Strong Evidence

    A dose of rigor is warranted on the report’s framing itself. “First” is notoriously hard to establish in security — earlier incidents may simply have gone undetected or unattributed — and “fully LLM-driven” needs a precise technical definition. Did a model plan and execute every stage autonomously, or did it automate most stages with humans supplying access, infrastructure, and the ransom negotiation? The available material does not yet answer that, and the security industry has an economic incentive to headline AI threats, which makes independent verification more important, not less.

    None of that skepticism blunts the strategic point. Whether JadePuffer proves to be the first fully autonomous ransomware attack or an important step short of it, the capability curve it sits on is real and publicly documented. Organizations that wait for a definitionally perfect “first” before adapting will be responding to the tenth.

    Background

    Ransomware grew over the past decade from opportunistic file-locking scams into a multibillion-dollar criminal economy, professionalized through ransomware-as-a-service — a franchise model in which developers lease attack tools to affiliates for a share of ransoms. Since the arrival of capable large language models, security researchers have tracked steadily deepening criminal adoption: first AI-polished phishing and malware development, then documented cases of AI models being misused across whole extortion operations, and lab proofs-of-concept for AI-generated ransomware. The JadePuffer report, as framed by Dark Reading, marks the point where that progression is claimed to have reached full automation in a real attack.

    Source: JadePuffer: The First Complete LLM-Driven Ransomware Attack — Dark Reading’s July 5, 2026 report on a ransomware campaign characterized as the first driven end-to-end by a large language model.

  • Crusoe’s Reported $30B Valuation Shows AI Investors Now Pay for Power Access

    Crusoe’s Reported $30B Valuation Shows AI Investors Now Pay for Power Access

    TL;DR · 30-second read

    The Short Version

    Crusoe, a company that builds the giant computer warehouses used to train and run artificial intelligence, is reportedly raising $3 billion from investors. The deal would value the company at about $30 billion.

    Crusoe’s approach is unusual. It lines up electricity first, then builds. It got its start using gas that oil wells would otherwise burn off as waste.

    Why it matters: anyone with enough money can buy computer chips. Enough electricity to run a small city is much harder to find. A price this high suggests investors see reliable power as the scarcest ingredient in the artificial intelligence boom.

    SiliconANGLE reported on July 4, 2026 that Crusoe, a developer of data centers built for artificial intelligence workloads, is raising about $3 billion in new funding at a valuation of roughly $30 billion.

    Crusoe is privately held. It develops large AI campuses, including the Abilene, Texas site built for Oracle to serve OpenAI, and it runs a cloud service that rents out computing capacity. The round would make it one of the most highly valued private companies in AI infrastructure.

    Executive Summary

    Crusoe is reportedly raising $3 billion at a $30 billion valuation. For a company whose products are data center campuses and rented computing power rather than chips or AI models, that is a very large figure. It is best understood as a price on what Crusoe has spent years assembling: sites where large amounts of electricity can be delivered on a usable schedule.

    The round matters beyond Crusoe. AI buildout is increasingly limited by how fast power can be connected to buildings, not by how fast servers can be bought. A valuation at this level shows at least one group of investors pricing that constraint directly. Utilities, equipment suppliers, rival developers and AI customers all compete for the same energized sites.

    The terms remain thin. The lead investors, the split between equity and other financing, and the use of the proceeds have not been set out publicly. Those details will determine how much the headline number actually tells us.

    What a $30 Billion Price Tag Is Actually Paying For

    Crusoe does not design its own chips or train frontier models. The graphics processors inside its buildings come from Nvidia and are available to any buyer with enough capital and patience, and so are the servers, networking gear and cooling equipment around them. What is not interchangeable is a site where hundreds of megawatts can be switched on within months. A megawatt is roughly enough electricity for several hundred homes. Many projects instead wait years in utility interconnection queues, the waiting lists for new connections to the grid.

    Crusoe has built its business around that scarce asset. The company began by generating power on site from natural gas that oil producers would otherwise flare, and it still treats energy sourcing as the first step of development rather than the last. Its flagship campus in Abilene, Texas, developed for Oracle to serve OpenAI, is planned at roughly 1.2 gigawatts, about the output of a large power plant. Without published financials, nobody outside the company can break the $30 billion figure down line by line. Still, what separates Crusoe from any well-funded buyer of GPUs is its pipeline of powered land. A valuation at this level is therefore, in substance, a price on access to energized capacity.

    Several groups feel that directly. Rival developers compete with Crusoe for the same sites near generation and transmission. Utilities receive the requests for large new loads. AI labs and cloud customers increasingly choose where to place workloads based on when capacity can go live, and less on price per rack.

    The Arithmetic of a $3 Billion Round

    If $30 billion is the post-money valuation, meaning the value after the new cash arrives, new investors would own about 10 percent of Crusoe. If it is the pre-money figure, they would own just over 9 percent. Either way, a check this large buys a fairly modest stake. That indicates investors are paying for expected future growth, not the current asset base.

    $3 billion is also small next to the cost of the campuses Crusoe builds. Once buildings, electrical equipment, cooling and computing hardware are included, a gigawatt-scale site needs financing many times that size. Crusoe developed Abilene with capital partners including Blue Owl Capital. Equity rounds in this sector usually serve as the base layer, and project debt, lease financing and joint-venture money are stacked on top of it. The reported raise is best read as the foundation for a much larger pool of capital, not as the whole of it.

    Vertical Integration Cuts Both Ways

    Crusoe owns more of the chain than a traditional colocation landlord, which typically leases space and power to tenants who bring their own equipment. Crusoe sources the energy, develops the campus, and in some cases operates the computing itself through its cloud service. That means fewer handoffs, faster delivery, and a larger share of the margin at each layer.

    It also concentrates risk. Crusoe’s best-known project is tied to a single anchor arrangement, and gigawatt-scale construction is exposed to equipment lead times, labor shortages and permitting delays. An integrated builder that also owns the chips carries both the buildings and the hardware if AI demand slows or pricing for rented computing softens. A $30 billion valuation implicitly assumes that demand holds and that Crusoe keeps delivering capacity on schedule. Both are reasonable bets, but neither is guaranteed.

    What It Signals for the Rest of the Buildout

    One funding round is not an industry trend. It does show where a significant group of investors believes the bottleneck sits. The companies most exposed to that view are the suppliers of turbines, transformers and switchgear that power-first campuses depend on. So are the utilities and pipeline operators that serve them, and the traditional data center operators who now compete with energy-led developers for land near generation.

    For enterprise and AI buyers, a well-capitalized Crusoe means another source of large-scale capacity outside the biggest cloud providers. How much that helps depends on how much of Crusoe’s future capacity is already committed to a small number of very large customers.

    Background

    Crusoe was founded in 2018 by Chase Lochmiller and Cully Cavness. It built its first business around natural gas that oil producers would otherwise flare at the wellhead, using that gas to generate electricity for computing on site. The company later moved into cloud computing for artificial intelligence and sold its bitcoin mining operations to NYDIG in 2025 to concentrate on AI infrastructure.

    Crusoe is best known as the developer of the Abilene, Texas campus built for Oracle to serve OpenAI, planned at roughly 1.2 gigawatts and developed with capital partners including Blue Owl Capital. The broader market has shifted toward what the industry calls neoclouds, specialized providers that rent GPU capacity, and toward developers that secure power before building. The reason is that grid connection timelines have become one of the main constraints on AI expansion.

    Sources

    Source: AI data center builder Crusoe reportedly raising $3B at $30B valuation (SiliconANGLE): report on Crusoe’s new funding round and valuation.

  • Two Ransomware Crews Reportedly Team Up in Joint Campaign

    Two Ransomware Crews Reportedly Team Up in Joint Campaign

    On 4 July 2026, IT Pro reported that cybersecurity experts had issued an alert describing an ‘unprecedented’ threat campaign in which two ransomware groups appear to be collaborating rather than operating independently. The public summary characterises the activity as a coordinated effort but does not, in the material available to us, name the groups, victims, sectors, or geographies involved.

    Executive Summary

    Ransomware-as-a-service crews typically compete for affiliates, victims and press attention. A public alert describing two named groups jointly running a single campaign — if it holds up on closer inspection — would mark a shift in how the extortion ecosystem organises itself, with implications for attribution, negotiation and defensive playbooks.

    For infrastructure operators, the immediate takeaway is not a specific new indicator of compromise but a reminder that the threat model is evolving faster than many incident-response runbooks. If two crews share tooling, access brokers or leak sites, defenders can no longer assume that a given intrusion set maps cleanly to a single adversary with a single playbook.

    What ‘Unprecedented’ Actually Means Here

    The word ‘unprecedented’ is doing heavy lifting in the headline. Ransomware groups have long shared infrastructure informally: affiliates rotate between programmes, initial-access brokers sell to whoever pays, and code from leaked builders (Conti, LockBit) circulates widely. What would be genuinely new is a formal, sustained partnership in which two branded operations run a single campaign end-to-end. On the public reporting available, it is not yet clear which of those descriptions best fits the activity being flagged.

    Readers should therefore treat the alert as a lead rather than a conclusion. The substantive question for defenders is whether investigators are seeing shared command-and-control, shared negotiation portals, or merely overlapping affiliates — each of which carries a different weight.

    Why Crews Would Cooperate — and Why They Usually Don’t

    Cooperation is economically rational when it lowers cost or raises the ransom take. Sharing a proven intrusion chain, splitting proceeds on high-value targets, or pooling leverage over a single victim (double-extortion with two leak sites) can all lift returns. Law-enforcement pressure since the 2021–2024 wave of takedowns has also thinned the affiliate pool, giving surviving operators an incentive to consolidate rather than compete.

    Against that, ransomware brands are jealous of reputation. A shared campaign dilutes the ‘we always decrypt’ signal that groups use to convince victims to pay, and it creates operational security risk: every extra participant is another potential informant. Historically, crews have preferred loose federation to formal alliance for exactly that reason.

    Implications for Infrastructure Buyers

    For data-centre customers, cloud tenants and connectivity buyers, the practical response does not change dramatically because two groups are named instead of one. The controls that matter — enforced multi-factor authentication, segmented backups tested for restore, privileged-access monitoring, and rehearsed incident-response contracts — apply regardless of which brand appears on the ransom note. What does change is negotiation posture: if two crews are jointly holding data, a victim cannot assume that paying one buys silence from the other.

    Insurers and legal counsel will want to understand this quickly. Cyber-insurance policies and sanctions-screening workflows are built around identifying a specific threat actor. A joint operation complicates both attribution and any regulatory obligation to check whether payment would breach sanctions.

    How to Read Alerts Like This

    Threat-intelligence alerts serve two audiences at once: defenders who need actionable indicators, and a wider readership that includes journalists, executives and — inevitably — the attackers themselves. Strong alerts publish indicators of compromise, TTPs mapped to MITRE ATT&CK, and a clear statement of confidence. Where those elements are absent from the public summary, the honest analytical response is to note the gap rather than fill it with speculation.

    Background

    Ransomware has been the dominant cyber-extortion model since roughly 2019, when double-extortion — encrypting data and threatening to leak it — became standard practice. The ecosystem is organised around branded ‘affiliate’ programmes such as LockBit, ALPHV/BlackCat, Cl0p and their successors, most of which run as ransomware-as-a-service.

    Law-enforcement operations against LockBit and ALPHV in 2023–2024, together with source-code leaks from earlier crews such as Conti, reshaped the market. Affiliates rotated between surviving programmes, new brands emerged, and researchers have periodically flagged overlaps in tooling and personnel. Against that backdrop, a claim of formal cooperation between two named crews is notable but consistent with the direction of travel.

    Source: Cyber experts issue alert after two ransomware groups team up on ‘unprecedented’ threat campaign — IT Pro report, 4 July 2026, describing a joint ransomware campaign flagged by security researchers.

  • Heat Wave and Data Center Demand Push PJM Grid to the Brink in Northern Virginia

    Heat Wave and Data Center Demand Push PJM Grid to the Brink in Northern Virginia

    The Prince William Times reported on July 4, 2026 that a summer heat wave, layered on top of the enormous electricity appetite of the region’s data centers, pushed the regional power grid “to the brink.” The grid in question is operated by PJM Interconnection, the regional transmission organization that coordinates electricity across all or parts of 13 states and the District of Columbia — including Northern Virginia, home to the largest concentration of data centers in the world.

    The report frames a collision that grid planners have warned about for years: weather-driven peak demand from air conditioning arriving at the same moment as a structural, around-the-clock load from data centers that has grown far faster than new generation and transmission have been built.

    Executive Summary

    According to the report, the stress event unfolded in Prince William County, Virginia and the surrounding region — the heart of “Data Center Alley,” where Prince William and neighboring Loudoun County host an unmatched density of hyperscale and colocation facilities. During a heat wave, residential and commercial air conditioning drives electricity demand to its annual peaks; data centers, unlike air conditioners, draw near-constant power day and night, so their load sits underneath the weather peak rather than replacing it.

    Why it matters: grid operators plan for the single worst hour of the year. When a fast-growing baseload (data centers) raises the floor and a heat wave raises the ceiling, the margin between available supply and peak demand — the buffer that prevents emergency measures like conservation appeals or rolling outages — shrinks. A “to the brink” event is a concrete, dated data point in a debate that is often conducted in abstractions about future AI load forecasts.

    A caveat on sourcing: this is a single local-newspaper account, and the headline-level material available does not specify which emergency procedures, if any, PJM invoked, what demand peaked at, or how close reserves actually came to exhaustion. Those specifics matter, and we flag them below.

    The Peak Problem: Flat-Out Air Conditioning Meets Always-On Compute

    Electric grids are sized for their worst hour, not their average one. In PJM territory that worst hour almost always occurs on a hot summer weekday afternoon, when tens of millions of air conditioners run simultaneously. Data centers change the arithmetic because they are effectively a new floor under demand: a large AI training or cloud facility draws a high, steady load 24 hours a day, in fair weather and foul. When a heat wave arrives, that steady draw does not politely step aside — it stacks. The result is that the same heat wave that a decade ago would have been routine can now push a region toward its limits, which is precisely the dynamic the Prince William Times describes.

    For lay readers, “to the brink” typically means the grid operator is working through its escalation ladder — asking generators to defer maintenance, importing power from neighbors, calling on demand-response customers who are paid to curtail, and in the worst case shedding load (rolling blackouts). The available reporting does not tell us how far down that ladder PJM went in this event, and that distinction — between a tight day and a genuine emergency — is the difference between a warning sign and a crisis.

    Northern Virginia Is the Stress Test the Rest of the Country Is Watching

    Prince William County is not a random dateline. Northern Virginia is the world’s largest data center market, and the AI buildout has accelerated demand there just as it has become harder to site new transmission lines and generation. PJM’s own capacity auctions — the mechanism by which the operator procures commitments of future power supply — have cleared at sharply higher prices in recent cycles, a market signal that supply is not keeping pace with projected demand. A heat-wave near-miss in this region is therefore a preview: other fast-growing data center corridors in Texas, Georgia, Ohio, and Arizona face versions of the same squeeze.

    The economics cut in several directions. Utilities and independent power producers benefit from higher capacity prices and large, creditworthy new customers. Data center operators face rising power costs and, increasingly, multi-year waits for grid connections — which is pushing some toward on-site generation, long-term nuclear and renewable contracts, and demand-flexibility commitments. Residential ratepayers, meanwhile, worry about absorbing the cost of grid upgrades driven by industrial customers, a tension that is now a live political issue in Virginia and across PJM’s footprint.

    Who Bears the Risk — and Who Blinks First in the Next Heat Wave

    Events like this sharpen a policy question that regulators have so far answered only partially: when supply gets tight, whose power is interruptible? Data centers have historically demanded — and paid for — extreme reliability, backed by on-site diesel or battery backup. That backup capacity is mostly idle during grid emergencies. Proposals to enroll data centers in demand-response programs, require flexible-load commitments as a condition of interconnection, or price peak consumption more aggressively all gain momentum every time a grid operator has a bad afternoon.

    There is also a reputational dimension. The data center industry argues, with some justification, that it pays substantial sums into the grid and that load growth also comes from electrification of homes, vehicles, and factories. But headlines that pair “heat wave” with “data centers” and “brink” land hard with the public regardless of the precise load attribution. Operators that can document flexibility — shifting deferrable computing work away from peak hours, dispatching backup assets to support the grid — will have an easier time in siting battles than those that cannot.

    Background

    Northern Virginia became the world’s data center capital over two decades, thanks to early internet exchange points, cheap land, favorable tax treatment, and proximity to federal and enterprise customers. Loudoun County led the first wave; Prince William County became the frontier of the next one, with the AI boom driving proposals for ever-larger campuses. PJM Interconnection, formed from a power pool dating to 1927, operates the transmission grid across the Mid-Atlantic and parts of the Midwest and has repeatedly flagged accelerating load growth — led by data centers — as a central reliability challenge of the coming decade.

    The tension surfaced well before this heat wave: PJM’s recent capacity auctions cleared at dramatically higher prices, utilities in Virginia have proposed new rate structures for large loads, and local land-use fights over data center siting in Prince William County have become some of the most contentious in the country. A dated, weather-driven stress event adds an operational exclamation point to what had largely been a forecasting debate.

    Source: Heat wave, data centers’ huge demand push regional power grid to the brink — Prince William Times, July 4, 2026, reporting on grid strain in the PJM region amid a heat wave and data center load growth.

  • China Switches On the First Commercial Underwater Data Center

    China Switches On the First Commercial Underwater Data Center

    China has brought online what is being described as the world’s first commercial underwater data center, according to a report published July 4, 2026 by the Spanish outlet OkDiario. The facility submerges sealed server modules in the ocean and uses the surrounding seawater as its cooling medium, an approach the report says sharply reduces the energy the facility consumes.

    The report frames the launch as a template other coastal regions could adopt, naming Cartagena, Spain as the kind of Mediterranean port city where the model might be replicated. It does not disclose the operator, the facility’s capacity, or its precise location.

    Executive Summary

    The announcement matters because it moves underwater data centers from experiment to product. Submerging servers has been tested before — most famously by Microsoft — but a commercial deployment means paying customers are expected to run real workloads on seabed infrastructure, and that changes the questions from “does it work?” to “does it pencil out?”

    The core appeal is cooling. Keeping servers from overheating is one of the largest energy costs in any data center, and the deep ocean offers a vast, stable heat sink at no mechanical-chilling cost. If seawater cooling delivers the efficiency the concept promises at commercial scale, it would arrive at a moment when AI-driven demand has made power and cooling the industry’s tightest constraints.

    That said, the source report is brief and light on specifics. It attributes no capacity figures, energy metrics, customer names, or operator details. The launch is a genuine milestone in cooling infrastructure if the commercial framing holds — but the evidence available in this report is a claim of a first, not a documented performance record.

    Why Put Servers on the Seabed?

    Data centers spend an enormous share of their electricity not on computing but on removing the heat that computing generates. The industry measures this with PUE — power usage effectiveness, the ratio of total facility power to the power that actually reaches IT equipment. Conventional air-cooled facilities need chillers, fans, and often large volumes of water to hold safe temperatures, and in hot climates that overhead climbs steeply.

    The ocean solves the problem passively. Below the surface, water temperature is low and remarkably stable year-round, and water conducts heat far better than air. A sealed capsule on the seabed can reject heat directly into an effectively unlimited sink, eliminating most mechanical cooling. Subsea deployment also removes evaporative water consumption — a growing point of friction between data centers and the communities that host them — and seabed real estate near dense coastal cities is not competing with housing or industry the way urban land is.

    From Microsoft’s Experiment to Chinese Commercialization

    The concept is not new; the commercial claim is. Microsoft’s Project Natick sank a sealed server vessel off Scotland’s Orkney Islands from 2018 to 2020 and reported that the submerged servers failed at a fraction of the rate of an equivalent land-based control group — likely because the nitrogen-filled, human-free capsule eliminated oxygen corrosion, humidity swings, and accidental knocks. Microsoft judged the experiment a technical success but never turned it into a product. China, meanwhile, has been running underwater data center pilots off its own coast for several years, so a progression from pilot to commercial service there is consistent with the trajectory — even though this report does not name the company involved.

    If the commercial characterization is accurate, China would be first to market with a technology a US hyperscaler proved and shelved. That is a familiar pattern in infrastructure: the economics that don’t fit one company’s portfolio can fit another market’s constraints, particularly where coastal land, grid capacity, and water for cooling are all scarce at once.

    The Hard Economics of Subsea Capacity

    The obstacles are as real as the appeal. A submerged module cannot be serviced by a technician; a failed component stays failed until the entire vessel is raised, which pushes operators toward redundant hardware and infrequent, expensive retrieval cycles. Marine engineering, corrosion-resistant housings, subsea power and fiber connections, and specialized deployment vessels all add capital cost that the cooling savings must repay. Insurance, uptime guarantees, and repair logistics for seabed assets are largely uncharted territory for enterprise customers used to walking their auditors through a facility.

    Environmental questions also need honest accounting. Rejecting heat into the ocean is thermodynamically unavoidable here, and while small-scale trials such as Natick reported minimal localized warming, the effect of dense clusters of commercial modules on marine ecosystems is site-specific and largely unstudied. Coastal permitting regimes — fisheries, shipping lanes, protected habitats — will shape where this model can actually go, and the report offers no detail on how the Chinese deployment cleared those hurdles.

    Could Cartagena Be Next?

    The report’s suggestion that coastal cities like Cartagena could follow is speculation, not an announced project, and it is worth being clear about that distinction. Still, the logic of the shortlist is sound: Mediterranean port cities combine dense populations that want low-latency services, constrained urban land and grids, warm climates that make conventional cooling expensive, and immediate deep water. Those are precisely the conditions under which subsea capacity is most competitive against land-based builds.

    For European adoption, the gating factors would be EU environmental review, marine-spatial-planning approvals, and — not least — the geopolitics of importing a Chinese-proven infrastructure model into European digital sovereignty debates. Any operator pursuing it would more likely license the concept or develop it independently than deploy Chinese-operated modules in EU waters.

    Background

    Underwater data centers trace to Microsoft’s Project Natick, which began with a proof-of-concept in 2015 and culminated in a sealed vessel of several hundred servers operating off Scotland from 2018 to 2020. The retrieved servers had failed at a small fraction of the rate of an identical land-based group, validating the reliability case — but Microsoft ended the program without a commercial product. China picked up the thread with coastal pilot deployments in the years that followed, pursuing subsea capacity as an answer to scarce coastal land, strained grids, and the water consumption of conventional cooling.

    The timing is not incidental. By 2026, explosive AI demand had made electricity and cooling the data center industry’s defining bottlenecks worldwide, pushing operators toward liquid cooling, novel sites, and any design that cuts overhead energy. A commercial subsea launch is China staking a claim to one of those frontiers first.

    Source: China just switched on the first underwater data center, cooling servers with the ocean to slash energy use, and coastal cities like Cartagena could be next — OkDiario report, July 4, 2026, on China’s launch of the first commercial seawater-cooled underwater data center.