Hunterbrook披露Modine数据中心管线疑似内部库，指向Google/AWS等约230亿美元建模需求

Based on Hunterbrook Media’s reporting, at the time of publication Hunterbrook Capital is long $MOD and short a basket of comparable securities. Positions may change. This article is not investment advice or a recommendation to buy, sell, or hold any security. See full disclosures below.
“This is a heavy lift.”
That’s how CEO Neil Brinker described the work ahead for Modine — which makes industrial chillers, air-handling systems, and other cooling gear — as it scales up its suddenly enormous data center business.
The lift: ramping factories across North America fast enough to deliver on long-term data center customer commitments, including a $4 billion agreement with a customer Modine did not name.
Hunterbrook Media uncovered a remarkably detailed view of what that work actually entails, and who it is for, as the fast-growing cooling company enters what seems to be the most pivotal moment in its 100-plus-year history.
In a rare and unusually granular look into corporate planning records, Hunterbrook found what appears to be the contents of Modine’s internal database for its data center business, uploaded on GitHub about a month ago during an apparent database migration to Google Cloud. Modine did not respond to repeated requests for comment on the authenticity of the information, but a wide range of verifiable details matches public statements made by the company, as well as independent data.1
The repository records where its data came from: URLs on modine.sharepoint.com, the company's SharePoint drive. The codebase also includes Modine email addresses tied to apparent Google employees, based on a review of LinkedIn profiles.
The GitHub repository’s tables detail factory planning, customer projects, supplier constraints, demand forecasts, production schedules, prices, internal costs, and engineering problems. It even captures what appear to be unit-level economics, including gross margin tied to specific customers, and tracks projects that ran late and ones that Modine lost.
Hunterbrook found the database, which did not require a password or payment to access, after asking a large language model basic questions about Modine’s customers — then had a team of reporters and engineers spend a week exhaustively vetting the findings.
One of the files seems to answer perhaps the biggest question hanging over the company: The unnamed $4 billion customer is Google.
The repository also shows a separate set of Google cooling opportunities in various stages totaling $4.5 billion worth of demand, with about $433 million booked so far on orders dating to as early as April of last year.
Beyond Google, the files also detail about $3 billion in firm or likely commitments to Modine from Amazon Web Services, Crusoe, and more than a dozen other data-center customers.
The $3 billion appears to be what Modine considers orders with more than an 80% conversion rate. Beyond that are billions more in what Modine labels as early-stage demand for its cooling products across the same top-tier data center customers.
The total: nearly $23 billion of modeled demand. AWS alone is the customer for over half of that total.
Modine hasn’t publicly named AWS as a customer either. But the database suggests the Amazon opportunity, even after discounting it by Modine’s own conversion odds, is worth more than the recent Google deal, and it mostly involves a different product. Firm demand already accounts for over $700 million of that opportunity. The data also details ongoing collaboration between AWS and Modine to jointly design and test custom builds, depicting Amazon as more of a long-term partner than a spot buyer.
For Modine’s fiscal quarter ending in September, the internal data center pipeline implied in the data2 — discounted by Modine’s internal conversion rating — is $540 million, about 20% above the company’s stated guide.3 For the entire fiscal year 2027, the probability-adjusted data points to $2.3 billion, which would amount to over 100% growth year-over-year; and beat the high-end of the company’s guided 60-80% growth.
Across the database, it’s clear the century-old HVAC company’s challenge is no longer finding customers. It is, like so many sleepy companies that suddenly became critical to the data center build-out, dealing with the privileged problem of perhaps too much demand.
And the GitHub repository shows a company attempting to deal with a wide range of challenges, from sold-out suppliers to understaffed engineering teams, in an effort to seize a historic multi-billion-dollar revenue opportunity.
In a couple of cases, Modine appears to have brought on entirely new suppliers, which is also confirmed by publicly available trade data.
The codebase also shows Modine pursuing ambitious projects that would take the company closer to silicon, not unlike its bigger rival Vertiv ($VRT).
One example: Modine appears to be trying to qualify a 1-megawatt coolant distribution unit against Nvidia’s liquid-cooling requirements. The project has no dollar value but offers an early glimpse of Modine’s entry into the fast-growing liquid-cooling market it has so far mostly missed.
Asked for comment, Nvidia, Google, and Modine did not respond, despite repeated requests. Amazon declined.
Someone Just … Posted It
The repository on GitHub shows that Modine was seemingly migrating its enterprise data from Microsoft Power BI dashboards to Google Cloud Cortex. The upload came from a single personal GitHub account owned by a software engineer who appears to have played a role in that migration.
Beyond the main user, the only other contributor to the repository — “Claude” — suggests that this whole thing may have been an effort to vibe-code the data migration.
Hunterbrook has chosen not to identify the individual, given the scale of the data in the repository.
We have not independently established whether the GitHub account owner works for Modine directly, through a contractor, or in some other capacity. We did, however, find that the account owner had a Modine email address.
The three Microsoft Power BI files contain over 100,000 rows of raw data across 41 tables covering monthly planning from April 2025 out into the early 2030s, daily factory production, individual purchase orders, and week-by-week parts shortages, among other operational details.
The database names more than a dozen customers, among them AWS, Google, and Crusoe, and ties them to specific factories, per-unit prices, standard costs, and margins.
Across all the files, one customer name is hard to miss: Google.
Google’s $4 Billion Cooling Platform
On May 26, Modine announced a deal large enough to reshape the company.
An unnamed “strategic data center customer” had entered a long-term capacity agreement, or LTA, under which Modine would reserve capacity to supply more than $4 billion of Modine’s flagship Airedale cooling products from 2027 through 2029. The customer paid Modine $165 million up front.
Modine’s same-day earnings release described it more specifically as an agreement with a major hyperscaler for chiller sales. Modine said the customer was an existing one, but kept the name confidential.
At the time, some were skeptical. Herb Greenberg, the legendary journalist popular in the short-selling world, wrote:
“Think the AI bubble is over? Hardly - $MOD up 15-17% pre-market on what amounts to a headline, with little in the way of details - not even an 8-K, just a press release. Just a big number and a claim that the deal ‘solidifies’ its position, blah, blah, blah. The beat goes on…”
The repository files provide some of the details Greenberg may have been looking for.
A table called Portfolio Review lists a project codenamed gCUB4.3. The project details a planned delivery of 3,512 total units of Modine’s next-generation 3-megawatt chillers, at $1.2 million apiece, to Google, with an estimated total sale price of $4,214,400,000, matching the “over $4 billion” ticket of the original press release. A field in the table sets first-site delivery in January 2027, consistent with the announced timeline.
The project-level data is also corroborated by another table called Monthly Demand that breaks down all demand by month. The numbers add up closely: $4.2 billion worth of Google demand for over 3,000 3-megawatt chillers, spread from January 2027 to November 2029. The table also shows, however, that about $500 million of that is marked as closed and lost, leaving just $3.7 billion in live demand. 80% of that is marked as firm.
And gCUB4.3 does not appear to be Modine’s first Google project of this kind.
The table also points to ongoing opportunities to ship a total of 5,983 units of the current-generation 2-megawatt chillers to Google, with orders placed as early as July last year. The total value of the 2-megawatt chiller projects is $3.55 billion,4 with a project codenamed gCUB4.1.2 shipping since July and a later-numbered version, gCUB4.1.4, marked “closed and won” in the files.
The majority, or $2.02 billion, is marked as early-stage, about $688 million as firm, and another $837 million as likely.
That incumbent relationship has a separate public echo. Modine’s fiscal-2026 annual report said one unnamed global technology customer accounted for approximately 11% of total company sales, about $350 million. It also said Modine had confidentiality agreements with three significant data center customers, although it didn’t name Google as one of them.
In total, there appears to be about $8.16 billion in live gross demand from Google across various stages spanning 2025 to 2030. While the bulk of Google demand is 2-megawatt and 3-megawatt chillers, it also includes related gear, such as computer-room air handlers, which makes up another $918 million.
But not all projected demand will convert to actual orders. Even on the recent LTA, Brinker has told investors that orders are taken four to six months before delivery, and Modine’s annual report warned the customer was not obligated to purchase every dollar of the anticipated volume.
The data show exactly how Modine plans around that.
Alongside each opportunity's gross unit quantity, the Monthly Demand table carries a projected unit quantity that appears to apply one of three probability weights tied to the opportunity's conversion probability: 100% for orders marked firm, 80% for those deemed likely, and 40% for early-stage prospects.5 In other words, the projected quantity is the gross quantity marked down by how likely Modine believes the deal is to convert.
Probability ratings appear largely associated with the opportunity stage: for example, projects in the “closed and won” or sometimes the “specification review” stage get marked as firm and given a 100% conversion rate, and projects in the “quotation & negotiation” stage generally get marked as likely and are given an implied 80% conversion rate. The rest, “Project in Planning,” are marked as early-stage and get a more speculative 40% probability of conversion.
An example of how we treated this analysis: Discounting the total gross demand value of $8.16 billion by the data’s own probability rating, you get about $6.07 billion in estimated demand from Google that Modine appears to be planning around.6
Beyond the dollars, the data points to the breadth and stickiness of Modine’s relationship with Google.
These gCUB project names — 4.1.2, 4.1.4, 4.3, and 4.4.0 — suggest Modine has been participating in successive generations of a Google-specific cooling architecture. That’s what Brinker told investors in May: “We’ve partnered with one hyperscaler, and we are now working on the second and soon to be third generation of that modular data center.”
And “gCUB” is not merely a suggestive internal codename. Google itself defines a gCUB as a modular “global central utilities building” that cools Google’s data center campuses.
Google’s public photo gallery shows gCUB infrastructure at campuses in New Albany, Ohio; Midlothian, Texas; and Storey County, Nevada. At Midlothian, Google says 10 modular cooling plants make up a complete gCUB and that the number is expected to grow; at New Albany, it says pipes carry cooling from a gCUB to two separate data-center buildings.
In other words, a gCUB is a repeatable campus utility architecture assembled from modular cooling plants and capable of serving multiple buildings — not the name of one chiller or one isolated data center.
That’s what the Modine data shows, too. Internal notes on the gCUB4.3 project indicate that Google’s “first site” order represents 38 units, suggesting more sites are expected to follow. And Google's orders name about 30 distinct location codes across multiple product lines.
Read together with Google’s own description of gCUBs operating across multiple campuses and feeding multiple buildings, those site codes support a larger interpretation: Modine appears to have been designed into a standardized Google cooling platform that can be repeated across sites.
Of course, the files still do not prove that Modine is Google’s exclusive supplier, that all units will be purchased, or that every Google campus uses Modine chillers.
Google did not respond to repeated requests for comment.
Amazon: Big Ambitions, With Real Bookings Already
Beyond Google, the databases uploaded to GitHub estimate potential demand from Amazon Web Services for Modine’s cooling products stretching out to 2033.
The total value of estimated gross demand? Roughly $12.38 billion.7
While Google is buying chillers from Modine, which produce the cold water or refrigerant, most of Amazon’s demand, about $11.42 billion, is for air handlers, which cool, filter, and circulate the conditioned air through the room. The air handlers consist of two types: 93,500 cubic-feet-per-minute DAHUs and the much larger 400,000 cfm TAHUs, which make up over two-thirds of the total air handler demand. The total estimated AWS demand also includes about $1 billion for chillers.
That indicates Amazon and Google aren’t necessarily competing for capacity from Modine; they’re mostly buying different things.
Modine substantially expanded its air-handler offerings more recently when it bought Calgary-based Scott Springfield Manufacturing for about $184 million in March 2024. In July of last year, the company announced a $100 million North American capacity expansion, citing "unprecedented demand from U.S. hyperscale and colocation data center customers."
About $741 million of total Amazon demand has already been closed and won, with orders placed as early as April last year. Most of the rest of the demand is marked as early stage, with the lowest conversion probability rate, 40%.
But even at that 40% rate, that’s roughly $5.6 billion in probability-weighted demand from Amazon, surpassing Google gCUB4.3’s total value.
Besides the dollar figures, the file notes provide rich color on a working engineering partnership.
For example, notes on one air handler program with AWS describe the two companies jointly uprating an existing model, adding that testing of one component “with AWS has been completed” and that AWS is scheduled to be on site at Modine’s factory in December to run additional testing. Notes on a separate air handler program with AWS track progress on designing a prototype for a customer variant of an existing design.
Together, the notes point to an ongoing, hands-on engineering relationship, with Amazon involved in testing and custom variants still in development, rather than a one-time purchase.
Amazon did not provide a comment to Hunterbrook as of the time of this publication.
Other Data Center Customers
Beyond Google and AWS, the repository also includes over 20 other data center customers covering over $2.3 billion in pipeline demand. Nearly $1 billion of that is classified as firm. About $640 million is marked as likely.
The notes in the database add color to the on-the-ground work behind the numbers, suggesting they’re not mere paper accounting.
For example, notes on a project titled TurboChill (Abilene) track progress through Buildings 7 and 8 and say 28 units had already been installed in Building 8. They also describe rework, late deliveries and commissioning-schedule pressure. Abilene almost certainly refers to Crusoe’s flagship 1.2-gigawatt Stargate AI data center in Abilene, Texas, built for OpenAI and Oracle.
There are other public signs of Modine’s presence in Abilene, too.
At least five LinkedIn profiles show Airedale/Modine chiller and HVAC roles in Abilene, one describing the job as "Critical Facilities Technician HVAC at Stargate Data Center as a chiller SME." Modine has also repeatedly recruited data-center cooling technicians specifically in Abilene.
The hardware matches, too. Modine's Abilene orders list the Crusoe chiller as part 80021 ("TCC14R28K-XX CRUSOE"), which Airedale identifies as a roughly 2-megawatt TurboChill. Its product images, 24 top-mounted fans in two rows of 12 over a V-frame coil, closely match the chillers seen in drone footage of the campus.
Crusoe did not respond to Hunterbrook’s request for comment as of the time of this publication.
Modine Is Trying to Get Into Nvidia’s Liquid-Cooling Stack
Another internal project has no estimated revenue at all. But it may be one of the most strategically important entries in the database.
The customer field says: “NVIDIA.”
And the project is named: “1 MW NVIDIA Certification.”
And the product is a CDU — a coolant distribution unit, the equipment that circulates cooling fluid between a data center facility’s water system and the liquid-cooled servers themselves.
According to Modine’s internal project notes, the company had already assembled its first prototype, was completing a second, and was preparing to discuss the testing process with engineers and the University of Texas at Arlington.
The records do not show that Nvidia has certified the CDU. They do not show a commercial purchase order. And they do not establish Nvidia as a revenue-producing Modine customer.
What they show is Modine trying to qualify a megawatt-scale liquid-cooling product around Nvidia’s requirements.
That is a potentially meaningful expansion beyond the products driving most of Modine’s current data-center growth.
Much of the company’s disclosed business sits on the facility side: enormous chillers, air handlers, FanWalls and CRAHs that cool buildings and data halls.
A CDU sits much closer to the silicon.
That distinction is becoming increasingly important as Nvidia pushes data-center architecture toward dense, liquid-cooled rack-scale systems. Nvidia’s GB200 and GB300 NVL72 systems are liquid-cooled, while the company says future AI-factory architectures will move toward megawatt-scale racks and beyond.
The University of Texas at Arlington connection is also unusually specific. UTA researchers have worked with Nvidia on liquid-cooled data-center systems before, including research involving cold plates and coolant distribution units.
So the zero in Modine’s revenue field may not reflect the scale of the opportunity.
If the company can get a CDU qualified against Nvidia’s increasingly dominant liquid-cooled architectures, it would give Modine another way to participate in the AI buildout — not just through facility-scale cooling equipment, but by moving coolant toward the racks themselves.
Nvidia did not respond to Hunterbrook’s request for comment.
Demand Isn’t the Problem. Supply Might Be
The customer pipeline is enormous, and a separate table shows how Modine plans to meet it — and where it might fall short.
A table, Build Plan, spells out the speed and magnitude of the ramp-up Modine is planning across its factories for its high-ticket items. For 3-megawatt chillers, the next-generation product that the recent $4 billion Google agreement covers, Modine appears to be planning its first production in October, then roughly doubling production each quarter until reaching full run-rate by early 2028, with the steepest jump in the earlier quarters.
Even then, it’s unclear whether capacity will be able to catch up to demand. Comparing the planned production figures in the Build Plan table with the monthly unit demand in the Monthly Demand table shows that, even with scheduled ramp-ups across Modine’s U.S. factories, modeled production will fall short of projected demand in most months for some products.
The problem is especially noticeable for the 2-megawatt chillers, which according to the data are the main items currently being shipped to Google.
For the 3-megawatt chillers, the supply-demand comparison looks more optimistic because the capacity numbers are more theoretical: No plant is actually making 3-megawatt chillers yet.
The existing Google 2-megawatt gCUB chillers, as detailed in the data, are currently moving through Modine’s manufacturing system, where their capacity is projected based on real factory stats. But neither of the two plants that are supposed to make the 3-megawatt chillers, Dallas and Jefferson City, appears to have made a single 3-megawatt chiller yet.
That seems a little behind Modine’s original plans — in line with what the company said in a recent earnings call. In July, Brinker said component shortages were forcing Modine to “dynamically” resequence its capacity rollout, though he said it would be “back on track and on schedule” by the end of the fiscal year.
The GitHub data confirms the delay concerns, pointing to lead times for capital equipment as one reason. The "Top Risks" field in the Portfolio Review table reads, verbatim: "Capacity and readiness concerns at a plant level. Capital equipment lead times for Dallas and Jeff City."
Factory capacity isn’t the only bottleneck. A manpower shortage appears to be forcing Modine to ration engineering hours among competing projects.
At least $149 million of projects were paused for some period of time because Modine had reassigned engineers to more urgent work, according to the data.
One is CyrusOne’s $60.6 million project for 168 chillers. Notes in the data say: “design work paused so engineering resources can focus on crusoe childress.” Another field says engineering would begin only after the Childress design was completed and was “still on pause” as of June 30.
Another $88.7 million Crusoe program was stuck behind the same project. “Design work paused due to engineering resources being allocated to crusoe childress job,” the note says, “unsure on new start date for design work.”
It appears Modine was triaging by redirecting engineering hours to put out the biggest fires first.
One former Modine engineer told Hunterbrook they were pulled from the automotive group and reassigned to data centers because that segment was “so starved for engineers.”
They also described Modine as having the “theoretical” capacity to execute on its explosive data center orders, but as struggling to deal with operational inefficiencies, including a lack of uniform systems across factories building the same parts or products.
The Suppliers That Can Stop the Line
Another bottleneck appears to sit outside Modine’s factories: its suppliers.
The Portfolio Review table on the $4.2 billion deal with Google spells out the issue: “Securing supplier capacity for production.”
The company has already told investors that shortages from a handful of critical suppliers constrained data-center production this year. Brinker said Modine had identified four suppliers that could create vulnerabilities and described some component shortages as a “hard stop” on production.
A table named stg_tracker appears to name the exact shortages and their dollar value.
One shortage involves flooded evaporators needed at Modine’s Jefferson City plant, a shortage responsible for $25.6 million of impacted value beginning July 8.8 The shortage tracker lists Modine’s Rockbridge plant as the supplier, while a separate capacity table identifies “Weiland” — almost certainly a misspelling of the heat-exchanger manufacturer Wieland — as Modine’s evaporator supplier.9
The other is an axial fan used to move air through cooling equipment, supplied by ebm-papst, a major German manufacturer of industrial fans and motors. That shortage was gating about $114 million worth of orders from being considered “clear to build” at Modine’s Rockbridge plant, with the first affected order dated July 15.
The axial fan’s listed action is simply: “Validate Due Dates with Buyer.”
Another table models how supply will continue to fall short of unit demand.
MP&L Detailed Capacity shows fan supplies from ebm-papst largely flatlining despite increasing demand.
The table also shows a wide component supply-demand gap for Amazon’s new TAHU air handler, with the capacity at three of the four named component suppliers remaining flat and far short of demand through March 2027.
The impact of component shortages is also visible in an Amazon order detailed in the data.
On one program calling for $204.4 million worth of air handlers, Amazon wanted production to begin in September, but notes entered in the database say Modine secured coil allocation from supplier Cancoil beginning only in December.
The project’s risk register states: “Coil supply for customer’s requested production start date not yet secured.”
A manager at Cancoil, which supplies cooling equipment to Modine, told Hunterbrook that the company is able to keep up with booked orders because the company “plans conservatively,” but admitted the demand is outpacing its capacity. They said to really scale, the company would need 18 months to three years to build new factories. They declined to say whether new factories are planned.
Ebm-papst is also racing to add capacity. In May, the company said booming data-center demand had prompted it to significantly expand production of large-format fans at its Mulfingen-Hollenbach facility in Germany and added more than 100 production workers.
Ebm-papst is also expanding its plant in Telford, TN. The Washington County regional planning commission approved “phase two” of its construction plan for the site in March. The company is planning to double its plant footprint by the end of 2027, according to MM Maschinenmarkt, a German trade magazine for the manufacturing sector.
“The U.S. market is a key growth market for ebm-papst, and we continue to invest in our capabilities and footprint to support future customer demand. Therefore, we are planning to expand our production capacities in Tennessee,” an ebm-papst spokesperson told Hunterbrook in a written statement. “Among others, we are also planning to produce large fans for data center applications here.”
The company is set to be acquired by Chicago-based ventilation and cooling equipment manufacturer Madison Air Solutions for $5.4 billion, the companies announced on Monday.
New Suppliers Come Online
Modine is also taking the matter into its own hands, including by finding supply substitutes and redesigning around components it can’t get fast enough.
Brinker has said Modine is negotiating specific allocations with critical vendors and working to qualify additional sources where possible. In July, he said the company believed it was on track to meet its fiscal-year plan.
The repository data supports that narrative.
On a $204 million Amazon program, the company plans to qualify Madok, now Canadian Coil Technologies, as a second coil supplier capable of supporting Amazon’s September start: “Approving an alternate coil supplier that can meet the September supply demand,” the project note says.
On a project Modine's records label "Anthropic (Tom Barrow)" and ties to an apparent "Anthropic" data center, one note reads: “Tom Barrow accepted delta fan and we de-prioritized Edge.” The word “delta” seemingly refers to Delta Electronics.
A Related Digital project carries almost the same note: “Related Digital accepted 630 fan and we de-prioritized Edge.”
Trade data reviewed by Hunterbrook suggests at least one of those alternative relationships is already moving material at scale.
Between July 21 and August 12, Modine received five shipments from Delta Electronics containing 75 tons of cargo described as “BACKWARD BLOWER.”
Each shipment carried shipping marks identifying the end destination: Modine’s plant in Grenada.
Hunterbrook repeatedly asked Modine about this — and provided the company with ample opportunity to deny the authenticity of the data. We received no reply.
If you have any other notable takeaways from the data — or disagree with any of our conclusions — email ideas@hntrbrk.com. We’re sure there’s stuff we missed.
Dhruv Patel is an investigative journalist based in Cambridge, Massachusetts, specializing in data-driven reporting. He helps spearhead investigative coverage at The Harvard Crimson, and his reporting has been cited and discussed by The New York Times, CNN, The Boston Globe, ABC News, and BBC, where he also frequently contributes commentary. A John Harvard Scholar at Harvard College, he studies computer science and economics to leverage machine learning for accountability journalism.
Jenny Ahn joined Hunterbrook after serving many years as a senior analyst in the US government. She is a seasoned geopolitical expert with a particular focus on the Asia-Pacific and has diverse overseas experience. She has an M.A. in International Affairs from Yale and a B.S. in International Relations from Stanford. Jenny is based in Virginia.
Vikas Kumar joined Hunterbrook from The Capitol Forum, where he led the corporate investigations team for a decade as a senior editor. He was previously an attorney at Gordon Feinblatt, a trial attorney for the Department of Justice, and a law clerk for a federal judge. He has a J.D. from University of Virginia School of Law and a bachelor's from Emory University. Vikas is based in Maryland.
Dan DeLorenzo is a creative director with 25 years reporting news through visuals. Since first joining a newsroom graphics department in 2001, he has built teams at Bloomberg News, Bridgewater Associates, and the United Nations, and published groundbreaking visual journalism at The Wall Street Journal, Associated Press, The New York Times, and Business Insider. A passion for the craft has landed him at the helm of newsroom teams, on the ground in humanitarian emergencies, and at the epicenter of the world’s largest hedge fund. He runs DGFX Studio, a creative agency serving top organizations in media, finance, and civil society with data visualization, cartography, and strategic visual intelligence. He moonlights as a professional sailor working toward a USCG captain’s license and is a certified Pilates instructor.