FEATURED

The AI Thermal Crisis Nobody Is Pricing

Start with physics, not a ticker.

When you pack 700 watts of thermal output into a single GPU chip and then stack eight of them into one server chassis, you create a heat problem that air conditioning fundamentally cannot solve. A single NVIDIA H100 GPU consumes 700 watts. A 256-GPU AI training cluster requires more than 180 kilowatts of power and generates heat that would overwhelm any conventional cooling system. The next generation is worse: the B200 platform exceeds 1,000 watts per chip, and the GB200 Superchip, combining two B200s with a Grace CPU, consumes 2,700 watts per unit. The B300 (Blackwell Ultra), which began shipping in January 2026, draws up to 1,400 watts per chip in rack-scale configurations. Air cooling is not a suboptimal option at these densities. It is physically impossible.

This is where the obvious opportunity ends and the interesting one begins.

Most investors tracking the AI buildout are watching chip companies and power utilities. Those are real opportunities. But they are first-derivative plays on a trend the entire market already sees. The second and third-order effects have not been fully reflected in prices. And one of the clearest expressions of that gap sits inside a 100-year-old Wisconsin industrial company that CNBC has not put on TV.

The Thermal Inflection Nobody Modeled

The data center cooling market does not get the headlines that chips and transformers do. It probably should. According to Grand View Research, the global data center liquid cooling market was valued at $6.65 billion in 2025 and is projected to reach $29.46 billion by 2033, growing at a 20% compound annual rate driven almost entirely by the AI chip thermal problem getting worse with every product cycle.

Here is the chain of causality most investors skip past: AI spending drives GPU demand. GPU demand drives rack power density. Rack power density drives a thermal crisis. The thermal crisis drives a massive capital reallocation toward liquid cooling infrastructure. And liquid cooling infrastructure requires something every AI factory has to buy before a single inference query can run.

Chillers. Coolant distribution units. Heat exchangers. Industrial-grade thermal management systems at scale.

When engineers say liquid cooling is mandatory for GB200 NVL72 racks, they do not mean it is a nice-to-have upgrade. They mean the rack physically cannot operate without it. Facilities deploying Blackwell-generation hardware must install direct-to-chip liquid cooling infrastructure, including coolant distribution units, leak detection, and compatible rack designs. The GB300 NVL72 rack-scale system draws 132 to 140 kilowatts under typical load, with peaks reaching 155 kW. Air cooling is not an option. It is an architectural failure mode at this power density.

So every hyperscale AI data center being built right now needs industrial-grade thermal infrastructure. Lots of it. Urgently.

Who Actually Builds This Stuff

The obvious names in data center cooling, Vertiv and Schneider Electric, are well-covered and reasonably well-valued. What has not been written about nearly enough is the company several steps further down the industrial supply chain that just crossed $1.1 billion in data center revenue and locked in a $4 billion long-term customer agreement to supply a major hyperscale cloud operator from 2027 through 2029.

Modine Manufacturing. Ticker: MOD. Based in Racine, Wisconsin. Founded in 1916.

This is not a startup. It is not a pure-play AI story yet. It is a century-old industrial manufacturer that spent years building heat transfer and thermal management products for trucks, agricultural equipment, and commercial HVAC systems. What Wall Street is still working through is the speed at which Modine is transforming its revenue base, and how durable that transformation actually is.

In fiscal 2026, Modine’s data center business surged 73% to cross the $1.1 billion mark. That single segment drove the company’s Climate Solutions division to record revenues with 43% growth for the full year. Total company net sales hit $3.18 billion, up 23% from the prior year. Adjusted EBITDA climbed 20% to $471 million.

That is four consecutive years of record revenue and adjusted EBITDA. Most people in financial media have not noticed.

The $4 Billion Contract That Changes the Math

Here is the part that should be getting more attention. In May 2026, Modine announced a landmark long-term capacity agreement with a major hyperscale customer to supply more than $4 billion of data center cooling products, specifically Airedale chiller systems, during calendar years 2027 through 2029. The customer paid Modine an upfront $165 million payment to fund capacity expansion. That is not a letter of intent. That is a hyperscale operator writing a nine-figure check to ensure supply.

Think about what that signals. The customer does not trust the spot market to deliver. They are pre-purchasing capacity three years out, in advance, with capital commitment. That is the behavior of a buyer who believes cooling infrastructure will be constrained. And they are right to think so.

Modine’s data center business grew at a 93% compound annual rate over the past two fiscal years. Management guided fiscal 2027 data center revenue growth of 60% to 80%, which puts the segment on track to approach $2 billion in revenue within 12 months. Total company guidance for fiscal 2027 calls for net sales growth of 20% to 35% and adjusted EBITDA of $650 to $680 million, implying more than 38% EBITDA growth from the fiscal 2026 baseline.

Q4 fiscal 2026 results were telling. Net sales hit $954.4 million, up 47% year over year. Adjusted EPS came in at $1.71, beating the Zacks consensus estimate of $1.51 by 13.2%. Data Center revenues exceeded $400 million in the quarter, even after severe weather reduced production time across multiple facilities. That last detail is important. The company is already capacity-constrained on the upside.

Where MOD Sits in the AI Value Chain

Let’s be precise about where Modine sits, because the depth matters.

• First derivative: Chip companies (NVIDIA, AMD, Broadcom)
• Second derivative: Power companies, utilities, transformer manufacturers
• Third derivative: Data center operators, hyperscale real estate
• Fourth derivative: Thermal management suppliers, cooling infrastructure manufacturers

Modine is a fourth-derivative play on AI spending. That is precisely why it is underowned. It does not show up in any AI ETF. It does not appear on AI infrastructure stock screens. It registers in industrial sector databases as a diversified manufacturer, because that is what it was for the past 100 years. The re-rating has not finished yet.

Every analyst currently covering MOD has a Buy rating. Zero Holds. Zero Sells. According to Investing.com, the average 12-month price target across seven analysts sits at approximately $341, with KeyBanc’s sum-of-the-parts valuation pegging the stock at $370. UBS lifted its view to $355. DA Davidson maintained its Buy rating with a $341 target as recently as June 2026.

The bullish case from analysts is worth reading carefully: management’s own targets anticipate 50% to 70% annual growth over the next two years, putting the company well on track to approach $2 billion in data center revenue. EBITDA margins are expected to expand toward 17% in fiscal 2027 as capacity expansion costs normalize and revenue leverage kicks in.

What Wall Street Is Still Missing

The margin story is where the analysis gets complicated. Modine is currently spending aggressively to expand manufacturing capacity, including a new 155,000 square foot data center cooling facility in Franklin, Wisconsin, as part of a multi-year $100 million expansion. That spending is compressing near-term gross margins, which came in around 23% for fiscal 2026. The headline margin number is what trips up short-term screens.

But this is a deliberate investment in operating leverage. When the capacity expansion is complete and revenue from the $4 billion long-term agreement begins flowing in 2027, the fixed-cost absorption math changes materially. Management has targeted EBITDA margins above 17% for the Climate Solutions segment in fiscal 2027. If data center revenue grows 60% to 80% into a partially built-out cost structure, the operating leverage is significant.

The other thing worth flagging: Modine is simplifying its business through a planned spin-off of its Performance Technologies automotive segment, merging it with Gentherm in a Reverse Morris Trust transaction expected to close before year-end 2026. When that closes, what remains is a near-pure-play climate solutions company with data centers as its largest and fastest-growing segment. That is a different valuation conversation entirely.

Price Levels and Positioning

MOD has delivered total returns of more than 1,200% over the past three years, yet the stock has pulled back sharply from its post-earnings peak near $323 as margin compression headlines temporarily dominated earnings reads. The market is doing what it often does with high-operating-leverage industrial companies: penalizing investment spend before the revenue it purchases arrives.

The stock is currently trading in the $240 to $260 range, which has acted as intermediate support following the post-earnings consolidation. The longer-term moving average structure remains intact, and the fundamental trajectory, a $4 billion long-term agreement, 60% to 80% projected data center revenue growth, and a portfolio simplification event, provides multiple re-rating catalysts over the next 12 months.

Volume patterns following the Q4 earnings report in late May 2026 showed institutional accumulation on pullbacks. Large funds building positions into a thesis that is still clarifying rather than fully priced. That is a different kind of signal than a stock that is already consensus.

Scenario Modeling

Bull case: Data center revenue grows at the high end of guidance (80%), capacity expansion costs normalize faster than expected, EBITDA margins reach 17% or higher in fiscal 2027, and the Performance Technologies spin-off is received as a pure-play re-rating event. Analyst price targets of $355 to $370 become the floor, not the ceiling. The $4 billion LTA customer extends or expands the agreement.

Base case: Data center revenue grows 60% to 70% in fiscal 2027, adjusted EBITDA reaches the midpoint of $650 to $680 million guidance, and the stock steadily re-rates as industrial investors recognize the transformation is structural, not cyclical. Shares track toward the $320 to $340 consensus target range as the Climate Solutions pure-play thesis gains credibility post-spin.

Bear case: Hyperscale capex spending pauses or gets delayed, chip thermal requirements improve faster than expected via architectural changes, or supply chain issues limit Modine’s ability to ramp capacity fast enough to capture the long-term agreement revenue. Gross margin pressure persists longer than modeled. The stock remains range-bound or pulls back further while the market waits for cleaner operating results.

The bear case is real but requires multiple things to go wrong simultaneously, including a reversal of the structural GPU thermal trend that is, by physics, getting worse each chip generation, not better.

Active Trader Framework

For traders approaching this as a catalyst-driven position, the Performance Technologies spin-off closing is the most significant near-term inflection point. When it completes, Modine’s reported financials will reflect the Climate Solutions business alone. The market will see the data center growth rate without the noise of the automotive segment alongside it. That reclassification event tends to attract institutional attention that was previously deterred by the conglomerate structure.

Key levels to monitor: the $240 range as meaningful support, $285 to $300 as the zone that has produced selling pressure, and a break above $300 on volume as a potential signal that the larger institutional thesis is accelerating. Implied volatility on MOD options remains elevated relative to historical norms, reflecting genuine uncertainty around the capacity ramp timeline and margin recovery pace. Position sizing should reflect that volatility environment.

Risk management consideration: the $4 billion long-term agreement customer is undisclosed. Any publicly visible shift in hyperscale capex from that operator would create headline risk even if Modine’s contractual position is protected.

What CNBC Has Missed About This Story

The thermal problem is not going away. It is actually getting worse. The B300 (Blackwell Ultra), already shipping to hyperscalers as of early 2026, draws up to 1,400 watts per chip in rack-scale form, double the H100’s 700 watts just two years ago. The AI chip roadmap is a one-way escalator toward higher thermal density, and every point on that escalator requires more cooling infrastructure than the last.

Modine did not invent this trend. It did not cause it. It just spent a century building the engineering capability to address it, and then found itself sitting at the intersection of an industrial skill set and the most urgent infrastructure buildout in modern technology history.

The $4 billion long-term agreement is not a bet on Modine’s potential. It is a hyperscale operator acknowledging that the thermal infrastructure problem is real, urgent, and supply-constrained enough to warrant locking in capacity years in advance with a nine-figure upfront payment.

Wall Street is slowly catching up. The analyst consensus just got there. The institutional position has not fully built. That gap, between what the data says and what is reflected in the stock price, is the part worth paying attention to.