February 24, 2026

The sharpest tell in today’s tape is how quickly the power-and-gear supply chain is being pulled into the data centre boom. In one place, the numbers are almost comical: Vertiv says Q4 orders jumped 252%, with data centres now 80–85% of revenue and backlog covering more than 100% of its 2026 sales guidance. When Vertiv, Eaton, GE Vernova and Legrand all talk about swelling backlogs while hyperscalers plan $650bn of data-centre and power-equipment spend this year, “AI buildout” stops being a narrative and starts being an industrial cycle.

The Big Stories

The clearest near-term winners are the firms that touch every MW of new compute. Electrical equipment makers Vertiv, Eaton, GE Vernova and Legrand reported large order increases and expanding backlogs as hyperscalers plan $650bn in data-centre and power-equipment spending this year. Vertiv’s 252% Q4 order growth and a backlog that already covers more than all of 2026 guidance is a blunt indicator: demand is outrunning delivery capacity, and the constraint is increasingly “how fast can you ship and install,” not “is there demand.” The investor implication is straightforward—this cycle is leaking out of the REIT/operator universe and into electrification and thermal management balance sheets.

In the Nordics, the scale just keeps climbing. atNorth will develop a 300MW campus on a 50-hectare plot at Hamre Industrial Park in Långsele, Sollefteå, Sweden, targeting operations in H1 2028. The company is leaning on modular design, renewable power, and heat-reuse partnerships—pretty much the Nordic playbook—but 300MW is a reminder that “green compute” regions aren’t staying boutique. Watch the timing: H1 2028 puts this in the next wave of capacity that arrives after today’s grid queues and supply-chain bottlenecks have (hopefully) been forced to adapt.

Grid interconnection workarounds are becoming policy, not just engineering. Enchanted Rock is pushing flexible, dispatchable onsite generation co-located with hyperscale data centres to accelerate interconnection, arguing a 500MW site could operate three to five years sooner and cut grid costs by $78m per GW. The notable detail is regulatory: FERC Order EL25-49 tells PJM to establish non-firm pathways—effectively “connect while building”—with PJM first to implement and other regions circling similar ideas. If this sticks, it’s a shift in bargaining power: the fastest projects won’t necessarily be those with the best substation, but those that can show up with credible, dispatchable onsite capacity and operational discipline.

Cybersecurity is increasingly being productised around “AI factories,” and the OT layer is next. NVIDIA is teaming with Akamai, Forescout, Palo Alto Networks, Xage Security and Siemens to bring AI-driven zero-trust to OT environments using BlueField DPUs, with demos at S4x26 in Miami. The pitch—agentless segmentation, local inspection/enforcement at the edge, and centralized AI threat analysis—reads like an attempt to make OT security scale the way hyperscale networking does. Why it matters: as more power gear, cooling, and facility controls become remotely managed and instrumented for efficiency, the OT attack surface becomes a board-level risk, not an ops footnote.

Cooling is heading toward a less intuitive destination: hotter water. Nvidia says its Vera Rubin processor can be cooled with 45°C water, eliminating the need for chillers and improving AI data-centre efficiency. Vendors across the stack are lining up behind liquid cooling, with the market pegged near $3bn in 2025 and projected to reach $7bn by 2029. The practical takeaway is that “cooling” is becoming a system design decision (water temps, heat rejection strategy, facility layout), not a bolt-on—yet another reason the capex mix is tilting toward mechanical/electrical complexity rather than just shells and racks.

In Brief

• 1623 Farnam partnered with Bridged Broadband to expand carrier-diverse routes from its Omaha carrier hotel, including reach into rural Missouri via an 800G DWDM backbone. It also disclosed a second Omaha interconnection facility planned for mid-2028 with at least 5MW.

• Light Source Communications (LSC) will build a 240-mile dark fibre network in Indianapolis, due by Q3 2027, aimed at GPU-driven AI workloads and hyperscaler/neocloud demand. LSC is also building in Las Vegas, Phoenix, and Tulsa—suggesting a coordinated push around AI corridor connectivity rather than one-off metro plays.

• Logix Fiber Networks is adding high-capacity routes in South Dallas and along Austin-to-Bastrop to link growth zones back to Dallas–Fort Worth carrier hubs. The piece frames this against big local data centre projects and spends (including Stack’s 220MW campus and cited Google/QTS commitments).

• Braemont Capital, Hamilton Lane and Delta-v Capital invested in VFN Holdings (Vero Networks) to accelerate FTTP and wholesale fibre expansion and pursue US-market M&A. It’s another sign that fibre platforms still see a financing window—especially where wholesale demand is being pulled by data centre clustering.

• India’s “domestic compute” message is getting more concrete. The India AI Impact Summit 2026 showcased domain-specific deployments and infrastructure commitments, including Sarvam AI’s 105B LLM hosted on India-based data centres and NPCI’s FiMI for national payment rails. In parallel, Deloitte argues India can become a global data centre hub but flags power availability, grid upgrades, and renewable integration as gating factors, while L&T says it will expand its data centre capacity to 32MW by March.

• The water-and-environment narrative is getting noisier—and more politicised. OpenAI’s Sam Altman rejected the viral “one ChatGPT query equals a bottle of water” claim while acknowledging AI’s energy use is a real concern, and Meta argues its Richland Parish data centre water use is roughly comparable to the farmland it replaced. Separately in the UK, campaigners urged Thurrock Council to reject approvals tied to Tilbury 3 and a quarry/landfill decision, citing a 30,700+ signature petition opposing a Google-led Thurrock data centre on Arena Essex LoWS.

• OT and building controls remain a soft underbelly. Claroty warned that legacy LonTalk is still embedded in many BMS deployments, with internet-accessible controllers exposing LonTalk-over-IP and about 75% of organisations managing BMS devices with known exploited vulnerabilities. If you’re underwriting or operating facilities, this is the kind of “old protocol, new outage” risk that quietly destroys resilience assumptions.

• Public funding continues to shape rural network modernization. West Virginia rural hospitals are using the federal Healthcare Connect Fund to secure discounted broadband and network upgrades for telehealth, EMR, and secure data exchange, with local partners advising on cybersecurity and resilient architectures.

• Grid investment isn’t just a US/EU story. An Aboitiz Renewables unit filed for approval of a P564.6m dedicated interconnection to link a 187MW-peak Ilocos solar project with a 277MWh battery to the Luzon grid. It’s a reminder that “more renewables + more load” often collapses into the same bottleneck: who pays for the wire.

• Ageing thermal fleets are being pulled back into the spotlight by AI demand. Integrated Global Services argues asset-integrity work (citing a Philippines boiler-tube case) will matter more as AI-driven data-centre growth increases reliance on older thermal assets for grid stability.

• The cultural backlash to “casual AI usage” is also taking shape. An opinion piece argues routine AI use in college settings increases energy demand via a digital rebound effect and calls for digital responsibility rather than bans.