Johnson Controls raised its full-year adjusted profit forecast to $4.85 per share from $4.70, beating analyst expectations of $4.76, as demand for its data center cooling systems accelerates faster than anticipated amid surging AI infrastructure buildouts that are driving rack densities to 100 kW and making thermal management the binding constraint alongside power and chips.
The forecast revision reflects a structural shift in data center economics. Cooling systems already account for up to 40 percent of a data center's total energy consumption, and AI workloads are pushing rack power densities from 20 kW to 100 kW and beyond. That increase is not linear. It requires entirely new thermal management architectures, including liquid cooling loops, direct-to-chip cooling, and hybrid air-liquid systems that Johnson Controls specialises in. The company has seen backlog growth in high-demand sectors, with management explicitly calling out AI-driven expansion as the primary driver. Partnerships like the one with NVIDIA for next-generation computing infrastructure underscore the connection between GPU clusters and thermal infrastructure.
Johnson Controls is not a traditional HVAC story. It divested its residential HVAC business to Bosch earlier this year to focus on faster-growing commercial segments, particularly data centers, building automation, and digital services. The AI data center exposure is material: the company introduced energy-efficient chiller systems tailored for AI workloads and reports a growing order book from hyperscalers and neocloud operators. Management commentary emphasises that clients are prioritising durable, energy-efficient solutions capable of handling variable high-density loads. That demand profile aligns with the shift from air cooling to liquid systems, where Johnson Controls has deep expertise from decades of industrial applications.
The cooling economics are straightforward and punitive for operators who get them wrong. A 100 kW rack generates heat loads that overwhelm air cooling, requiring liquid systems that cost $1 million to $2 million per megawatt of IT capacity. Poorly designed cooling leads to hotspots, reduced GPU lifespan, and throttled performance. Effective cooling unlocks the full capacity of the compute investment. Hyperscalers like Microsoft, Google, and AWS are already specifying liquid-ready facilities in RFPs. Neoclouds like CoreWeave and Lambda face even tighter margins, where cooling efficiency directly determines whether they can offer competitive pricing. Johnson Controls benefits from multi-year service contracts that lock in revenue as facilities scale.
For SF readers, the story is that AI infrastructure value is dispersing into public supply-chain layers where industrial incumbents hold structural advantages. Cooling joins power, chips, memory, fiber optics, and land as scarce inputs that command premium pricing. Startups building AI applications compete on model performance and distribution. Infrastructure startups compete on cost and reliability in these commodity layers. Investors who understand the dispersion pattern can identify the next Johnson Controls: Vertiv for power, nVent for electrical infrastructure, nLIGHT for fiber lasers. The scarcity premium accrues to the suppliers who solve the physics problems at scale.
Whether cooling becomes a defensible AI infrastructure moat depends on the pace of commoditisation. Liquid cooling is not new; it has been used in supercomputers for decades. AI data centers are applying it at unprecedented scale, which creates short-term scarcity as supply chains ramp. Johnson Controls' moat is execution: global manufacturing footprint, service networks, and integration with building management systems. Competitors like Trane Technologies and Carrier Global are in the same position. The real question is whether Chinese suppliers like Gree or Midea can replicate the EV playbook and flood the market with low-cost chillers. For now, Western incumbents hold the high-end market where reliability trumps price.
Rising thermal management costs are already reshaping data center decisions. Location choices favour colder climates like the Nordics, Quebec, or the US Midwest where ambient temperatures reduce cooling loads. Facility design is shifting to liquid-ready from day one, with raised floors giving way to in-row cooling and direct-to-chip manifolds. Financing models are incorporating cooling efficiency as a key metric, with lenders and equity investors demanding PUE below 1.2 for new builds. Operators who prioritise compute density over cooling efficiency face stranded assets as rack power continues to climb.
Also read: Qwen3.6 Heretic v2 shows the local AI community is now engineering refusal-free frontier models • Gen Z treats subscriptions as event tickets, and platform loyalty is officially dead • Christian creators are outsourcing AI-generated devotionals to Fiverr, and the model works for any niche media category
