Data Centers: where's the money now?
In our data center market deck, you will find everything you need to understand the market
SUMMARY
Data Centers: where's the money now? The money is in the bottlenecks that make AI capacity real: GPU clouds, power, electrical infrastructure, cooling, powered campuses, and cluster connectivity.
The old data center lens is now too narrow. The market is no longer just about land, buildings, leases, and uptime, but about who can assemble GPUs, power, transformers, cooling, financing, permits, and customers fast enough.
The strongest money signal is not generic AI excitement. It is repeated capital movement into the physical constraints that decide whether AI workloads can actually run.
AI cloud and neocloud operators look less like software companies now and more like compute utilities. CoreWeave’s backlog, active power, GPU-backed financing, and large customer agreements show that controlled compute access is becoming infrastructure collateral.
Power may be the deepest shift in the whole market. Nuclear, fuel cells, gas access, long-term PPAs, and utility relationships now matter because a renewable credit does not keep a GPU cluster running at night.
Electrical infrastructure is one of the clearest money zones because the order books are already showing it. Eaton, Vertiv, and GE Vernova are seeing demand, backlog, and pricing power move in the same direction.
Cooling has been re-rated from technical subcategory to strategic capacity unlock. The CoolIT and Boyd Thermal deals show that liquid cooling is no longer a future topic but a near-term hyperscale requirement.
Hyperscale developers and colocation platforms are attracting major capital, but only when they have serious execution proof. The market is paying for powered campuses with tenants, financing, equipment access, and credible delivery teams.
Chips still dominate the narrative, but the money around chips is becoming more interesting. Capital is moving into custom silicon, optical interconnect, switches, servers, and integration because the real problem is making thousands of chips behave like one machine.
Construction is investable only when the chain is already assembled. Land alone is not enough anymore; power, permits, debt, equipment slots, hyperscale demand, and delivery speed are what create value.
Software is real, but still smaller than the physical bottlenecks. It becomes much more important only when vendors prove they can unlock stranded megawatts or cooling headroom, rather than just offer better dashboards.
So the ranking is fairly clear now: neoclouds, power, electrical gear, cooling, powered campuses, cluster infrastructure, campus delivery, and then software. The pattern is simple: the closer a company is to a scarce input, the more money seems to be flowing toward it.
This market map, featured in our data center market deck, highlights top companies and startups in the data center market
What are the main company categories in data centers right now?
Data centers used to be easy to describe: land, buildings, leases, uptime, and customers. That version is now too small.
These days, the data center market looks much more like an AI infrastructure supply chain, where the scarce pieces are GPUs, power, transformers, substations, cooling systems, financing, and teams that can bring everything online fast enough.
That’s why we need to distinguish the categories first.
| Category | Simple description | Example companies |
|---|---|---|
| AI cloud / neocloud operators | Companies that rent GPU compute and increasingly control the data centers, power, and financing behind that compute. | CoreWeave, Lambda, Crusoe, Nebius, Nscale, Fluidstack, TensorWave |
| Hyperscale data center developers and colocation platforms | Owners and developers of large campuses leased to hyperscalers, AI labs, cloud providers, and large enterprises. | Aligned, Vantage, QTS, DataBank, EdgeCore, Digital Realty, Equinix, STACK |
| Power origination and energy-backed infrastructure | Utilities, nuclear operators, gas developers, and power platforms securing electricity for AI campuses. | Constellation, Talen, Vistra, Oklo, TerraPower, Kairos Power, X-energy, GE Vernova, NextEra |
| Electrical infrastructure and grid equipment | Transformers, switchgear, UPS systems, substations, and power distribution equipment that turn a site into usable megawatts. | Eaton, Schneider Electric, Vertiv, GE Vernova, Siemens Energy, ABB, Legrand |
| Cooling and thermal management | Liquid cooling, cold plates, CDUs, immersion systems, chillers, and thermal hardware for high-density AI racks. | CoolIT, Boyd Thermal, Motivair, Accelsius, Submer, Vertiv, Schneider Electric, Eaton |
| AI chips, networking, and server infrastructure | GPUs, custom accelerators, optical interconnects, switches, networking silicon, and server integration. | Nvidia, Broadcom, Marvell, Celestica, Astera Labs, Arista, Supermicro, Celestial AI |
| Data center construction and campus delivery | Builders, EPCs, modular providers, and developers that turn land, power, permits, and customers into live capacity. | DataBank, EdgeCore, Compass, DPR Construction, Turner, Jacobs, Quanta |
| Data center software and autonomous operations | Software that optimizes cooling, power, carbon, workload placement, and facility operations. | Phaidra, EkkoSense, nZero, Schneider Electric, Siemens, Johnson Controls |
Is money flowing into AI cloud / neocloud operators right now?
Yes, money is clearly flowing into AI cloud and neocloud operators right now. At this point, the evidence is strong enough to call this one of the hottest parts of the whole data center market.
CoreWeave is the cleanest example because its numbers now look more like infrastructure than software. In Q1 2026, the company reported nearly $100 billion of revenue backlog and more than 1 GW of active power. That changes the way we should read the category. A neocloud with that much contracted demand is not just renting GPUs by the hour but actually becoming a balance sheet for AI capacity.
The financing confirms that shift. CoreWeave has used GPU-backed debt, note offerings, and large customer agreements with companies like Meta, Anthropic, and Jane Street. That mix matters because it means both lenders and customers are starting to treat GPU capacity as something close to infrastructure collateral. The asset here is controlled access to compute.
The private market is saying the same thing. Lambda raised more than $1.5 billion in late 2025, only months after a smaller Series D. Crusoe raised $1.375 billion at a valuation above $10 billion, after being valued around $2.8 billion roughly ten months earlier. Nscale raised $2 billion at a $14.6 billion valuation, with Nvidia involved and senior Meta figures joining its board. These are not cute AI rounds but capital-heavy bets on who can combine chips, power, data centers, and customers before the window closes.
It is also becoming broader than the basic Nvidia GPU trade. TensorWave raised $350 million in June 2026 at a $1.55 billion valuation to expand AMD-based AI infrastructure. That detail is important because it shows investors are funding any credible path to scarce AI compute.
So the conclusion is pretty direct: the best neoclouds are becoming compute utilities. The risk is real, because the capex is brutal and the financing stack is complex. But right now, money is flowing there because these companies sit exactly where the market is tight: GPUs, power, data center access, and signed customer demand.
If you want more recent data on this point, please see our latest data center market report.
As this chart shows, and as featured in our data center market deck, search interest in data centers has increased significantly
Is money flowing into hyperscale data center developers and colocation platforms right now?
Yes, money is flowing into hyperscale data center developers and colocation platforms, but only the serious platforms are getting the real money.
The Aligned deal says a lot. A consortium including BlackRock’s Global Infrastructure Partners, MGX, and AIP agreed to buy Aligned at about a $40 billion enterprise value. That is not just a big real estate transaction but the market putting, actually, a very high price on a platform that already has scale, customers, and AI-ready development capacity.
The same pattern shows up in the platforms still raising capital. Vantage raised $9.2 billion of equity in 2024, after upsizing the round by $2.8 billion, and said the money would support tens of billions of dollars of data center development. EdgeCore entered 2026 with 1.8 GW delivered or under development, backed by $5.9 billion of equity and a much larger planned investment pipeline. DataBank secured $2 billion of construction financing for the first three buildings at its Red Oak campus in Texas, delivering 180 MW in the first phase.
What makes these deals more convincing is that they are not just based on “AI demand will be big.” DataBank’s first Red Oak facilities are tied to hyperscale demand, including Oracle-linked capacity. That is exactly what lenders want to see now: a tenant, a power plan, a construction path, and a site that can actually become live capacity.
Oracle’s capex pressure makes the logic even clearer. The company spent about $55.7 billion on capex in fiscal 2026 and guided to roughly $70 billion for the following year. Investors hated the cash intensity, but for data center platforms, it proves something useful: even the largest cloud companies need external partners, financing structures, and third-party campuses to absorb the physical buildout.
Is money flowing into power origination and energy-backed infrastructure right now?
Yes, money is flowing into power origination and energy-backed infrastructure, and this may be the deepest shift in the market.
The nuclear wave is the clearest proof. By 2026, every major hyperscaler had signed at least one nuclear-related power deal, and disclosed projects added up to roughly 9.8 GW of committed nuclear capacity. Meta alone has lined up up to about 6.6 GW across Vistra, Oklo, TerraPower, and Constellation. Microsoft’s 20-year agreement with Constellation helps restart Three Mile Island Unit 1, adding roughly 835 MW. Amazon’s Talen agreement can supply up to 1.92 GW from the Susquehanna nuclear plant.
These are not just nice climate announcements. That is the important part. A renewable credit does not keep a GPU cluster running at night. Firm power does. That is why the market has moved so quickly toward nuclear, gas, on-site generation, and long-term PPAs. Hyperscalers are basically admitting that AI growth is now constrained by electricity.
This also changes geography. Data centers used to follow fiber, tax incentives, and cheap land. Today, the best sites increasingly start with power: nuclear adjacency, gas access, interconnection queues, utility relationships, and grid flexibility. A slightly worse location with real power can beat a better location stuck waiting for the grid.
Oracle’s expanded partnership with Bloom Energy fits the same pattern. Oracle and Bloom are supporting up to 2.8 GW of fuel-cell capacity, with an initial 1.2 GW deploying across U.S. Oracle projects. That is not a small side experiment but a sign that some buyers no longer trust normal grid timelines enough to wait patiently.
At this point, power is not just an input for data centers. It is rather becoming the gatekeeper of the whole AI infrastructure market. If you control power, you control whether AI capacity can actually exist.
If you want more recent data on this point, please see our latest data center market report.
This chart, featured in our data center market deck, illustrates yearly venture capital funding for data center startups
Is money flowing into electrical infrastructure and grid equipment right now?
Yes, money is flowing into electrical infrastructure and grid equipment, and this is one of the least debatable parts of the whole market.
Eaton’s numbers make the case without much interpretation needed. In Q1 2026, its Electrical Americas rolling 12-month orders were up 42%, driven by data center momentum, and Electrical-sector backlog was up 48% year over year. The more revealing detail is that Eaton’s data center orders were reportedly up about 240% versus Q1 2025, while data center revenue was up around 50%.
Vertiv tells the same story from the critical power and cooling side. Q1 2026 net sales were up 30%, Americas organic sales were up 44%, and management raised guidance. For a company selling the guts of data center infrastructure, that is a very direct read-through from AI capex into real revenue.
GE Vernova may be the strongest example because the order book looks almost absurd. In Q1 2026, its Electrification segment booked $2.4 billion of data center equipment orders, more than it booked in all of 2025. Total Electrification orders nearly doubled year over year, and backlog rose by more than $13 billion quarter over quarter. When a company books more data center electrical demand in one quarter than in the entire previous year, we do not need to be cautious about the read.
There is also pricing power, which is often the real proof of a bottleneck. GE Vernova said power equipment orders early in Q2 had already exceeded Q1’s value, with pricing expected to be 10 to 20 points higher than late 2025 on a dollar-per-kW basis. Suppliers are not only selling more units. They are selling into scarcity.
Is money flowing into cooling and thermal management right now?
Yes, money is flowing into cooling and thermal management right now, and the category has clearly been re-rated.
The CoolIT deal is the kind of signal that changes how we should read the market. Ecolab agreed to buy CoolIT for $4.75 billion, and CoolIT is expected to generate about $550 million of sales over the next 12 months. That already looks rich for an infrastructure hardware company. But the valuation jump is the real story: KKR reportedly bought control of CoolIT in 2023 when the company was valued around $270 million, and the Ecolab sale gives KKR roughly a 15x return.
That kind of jump happens when a niche suddenly becomes strategic. CoolIT’s revenue reportedly quadrupled under KKR, its workforce more than doubled, and the company won more large cloud customers. In less than three years, liquid cooling moved from “future topic” to real hyperscale demand.
The industrial buyers are acting like they know this too. Eaton agreed to buy Boyd Thermal for $9.5 billion, with Boyd expected to generate about $1.7 billion of revenue in 2026 and most of it coming from data center customers. Schneider Electric bought Motivair and built out a fuller liquid-cooling portfolio. Accelsius raised $65 million in early 2026, led by Johnson Controls, with Legrand participating.
The reason is simple: AI racks are becoming too dense for traditional air cooling to carry the whole load. Better cooling does not just lower the electricity bill. It can also let a site run more compute inside the same power envelope, which makes cooling a capacity unlock.
So yes, cooling is now a real money zone.
If you want more recent data on this point, please see our latest data center market report.
This chart, featured in our data center market deck, shows how Equinix is capturing share in data centers
Is money flowing into AI chips, networking, and server infrastructure right now?
Yes, money is flowing into AI chips, networking, and server infrastructure, but the more interesting money is now around the cluster, not only the GPU.
Nvidia is still the center of gravity. In Q1 fiscal 2027, Nvidia reported $75.2 billion of data center revenue, up 92% year over year. That number is so large that it almost becomes the baseline. We do not need to argue that Nvidia is winning. The market already knows that.
The better question is where money flows around Nvidia. Broadcom’s AI semiconductor business is growing at triple-digit rates, driven by custom accelerators and networking. That matters because hyperscalers want custom silicon, better cluster fabrics, and ways to scale models without being trapped inside one cost structure.
Optical interconnect is another strong signal. Marvell agreed to buy Celestial AI for $3.25 billion upfront, with the deal potentially reaching $5.5 billion if milestones are hit. Celestial had already raised at a $2.5 billion valuation earlier. That quick jump tells us optical connectivity is no longer a side issue. It is becoming one of the ways the market tries to solve the “how do we wire a giant AI cluster?” problem.
Server integration is also getting paid. Celestica reported Q1 2026 revenue of $4.05 billion, up 53% year over year, with improved margins. That is less famous than Nvidia’s number, but it is useful because it shows AI infrastructure demand spreading into the companies that physically build and integrate the systems.
So the answer is clear: chips are still attracting huge money, but the stronger read today is broader. Capital is moving into everything that lets thousands of chips behave like one machine: custom silicon, switches, optical interconnect, servers, and cluster-level integration.
Is money flowing into data center construction and campus delivery right now?
Yes, money is flowing into data center construction and campus delivery, but the market is paying for execution, not just square footage.
DataBank’s Red Oak financing shows the point well. The company secured $2 billion of construction financing for the first three buildings at its South Dallas campus, delivering 180 MW of capacity. The broader campus is planned for eight buildings. What makes this important is that the first phase is tied to real hyperscale demand. The financing is following certainty, not just a glossy AI story.
The size of these projects also tells us the market has changed. EdgeCore entered 2026 with 1.8 GW delivered or under development. DataBank is building hundreds of megawatts in one location. Oracle says it is spending tens of billions a year on AI data centers. The physical unit of competition is no longer one building. It is the campus, the power plan, and the financing stack around it.
Construction has also become a balance-sheet problem. Oracle’s capex makes that obvious: roughly $55.7 billion in fiscal 2026, then about $70 billion guided for the following year. Investors did not love it because the cash burn is massive. But for the construction and development ecosystem, it confirms the scale of demand. The real question is who can deliver it without breaking the financing model.
Speed now has financial value too. A developer that already has land, permits, utility coordination, equipment slots, and tenant commitments can move before the market reprices again. That is why delivery capacity itself is becoming investable.
At the end of the day, construction money is flowing where the chain is already assembled: power, permits, debt, equipment, tenants, and a credible delivery team. Generic construction exposure is not enough.
If you want more recent data on this point, please see our latest data center market report.
This chart, featured in our data center market deck, illustrates yearly funding for data center startups
Is money flowing into data center software and autonomous operations right now?
Yes, money is flowing into data center software and autonomous operations, but this is still a smaller money pool than the physical bottlenecks.
Phaidra is the best recent startup example. It raised more than $50 million in Series B funding in October 2025, with investors including Collaborative Fund, Index Ventures, Nvidia, and Sony Innovation Fund. The round is not huge compared with neoclouds or cooling deals, but Nvidia’s participation matters. It hints that optimization software becomes more valuable when every wasted watt can mean lower GPU output.
The software problem is real. These are not just dashboards. Phaidra is building AI agents for “AI factories.” EkkoSense and nZero are focused on energy visibility, cooling optimization, and carbon or power reporting. As racks get denser and power gets tighter, operators need faster control loops than manual facilities teams can provide.
The upside could be meaningful if vendors prove they can free real capacity. In a constrained campus, even a few percentage points of recovered power or cooling headroom can delay an upgrade, reduce waste, or let more GPUs run safely. That is the version of the category that could become much more valuable.
But we should be honest: the proof is not as strong here as it is in cooling, power, electrical gear, or neoclouds. We do not yet see a CoolIT-style valuation jump, an Aligned-scale platform deal, or a GE Vernova-style order explosion. The category is real, but it has not become one of the main money pools yet.
So yes, money is flowing into data center software, but today it remains an enabling layer. It moves up the ranking only when vendors can prove they unlock stranded megawatts, not just provide better dashboards.
So, where is the money in data centers right now?
The evidence is now strong enough to be very direct: the money is in the bottlenecks that make AI capacity real.
That means GPU clouds, power, electrical infrastructure, cooling, powered campuses, and cluster connectivity.
The old “data center = real estate” lens misses the point.
Today, the market is rewarding the companies that control scarce inputs and remove deployment friction.
| Rank | Category | Signals proving money is flowing |
|---|---|---|
| 1 | AI cloud / neocloud operators | CoreWeave near-$100B backlog and 1 GW+ active power; GPU-backed financing; major Meta, Anthropic, and Jane Street agreements; Lambda $1.5B+ round; Crusoe $1.375B at $10B+ valuation; Nscale $2B at $14.6B; TensorWave $350M at $1.55B for AMD-based AI infrastructure |
| 2 | Power origination and energy-backed infrastructure | Roughly 9.8 GW of hyperscaler nuclear commitments; Meta up to 6.6 GW; Microsoft-Constellation Three Mile Island restart; Amazon-Talen up to 1.92 GW; Oracle-Bloom up to 2.8 GW of fuel-cell capacity |
| 3 | Electrical infrastructure and grid equipment | Eaton Electrical Americas orders up 42%; Eaton data center orders reportedly up about 240%; Vertiv Q1 sales up 30%; GE Vernova booked $2.4B of data center electrification orders in one quarter, more than all of 2025; pricing moving higher per kW |
| 4 | Cooling and thermal management | Ecolab buying CoolIT for $4.75B; CoolIT expected at about $550M of next-12-month sales; KKR reportedly making roughly 15x from its 2023 CoolIT investment; Eaton buying Boyd Thermal for $9.5B; Accelsius $65M Series B led by Johnson Controls |
| 5 | Hyperscale developers and colocation platforms | Aligned sale at about $40B enterprise value; Vantage $9.2B equity raise; DataBank $2B financing for 180 MW in Red Oak; EdgeCore 1.8 GW delivered or under development; platform value tied to power and tenant certainty |
| 6 | AI chips, networking, and server infrastructure | Nvidia data center revenue up 92%; Broadcom AI semiconductor revenue growing at triple-digit rates; Marvell buying Celestial AI for up to $5.5B; Celestica revenue up 53%; money moving from GPUs into cluster connectivity |
| 7 | Data center construction and campus delivery | DataBank Red Oak moving from financing to phased delivery; Oracle capex at about $55.7B in fiscal 2026 and guided near $70B next year; EdgeCore 1.8 GW pipeline; construction value increasingly tied to execution certainty |
| 8 | Data center software and autonomous operations | Phaidra $50M+ Series B with Nvidia participation; EkkoSense and nZero targeting thermal and energy optimization; real upside if software unlocks stranded power or cooling capacity; still smaller than the hardware-heavy bottleneck categories |
If you want more recent data on this point, please see our latest data center market report.
This chart, featured in our data center market deck, compares the main business model options for hyperscale data center operators
OUR METHODOLOGY
We approached this question this way because “where is the money in data centers?” is no longer obvious from intuition. The market now cuts across compute, power, electrical equipment, cooling, campuses, construction, and software, so a broad answer would miss where capital is actually concentrating.
We broke the market into analytical dimensions and looked at recent signals inside each one. We prioritized financings, acquisitions, order growth, backlog, capex, power commitments, and signed customer demand because these signals show where money is moving more clearly than narrative momentum alone.
The final ranking comes from aggregating those signals and comparing their strength across categories. We gave more weight to evidence that was recent, repeated, financially large, and directly tied to AI capacity constraints.
That is why the analysis emphasizes bottlenecks such as compute, power, electrical infrastructure, cooling, and powered campuses, while treating software as real but still less proven as a major money pool.
We used company releases, investor relations pages, direct transaction announcements, and specialist financial reporting where available. For private valuations or market-reported transaction details, we used tier-1 financial or specialist sources.
Key sources used for this analysis include: CoreWeave’s Q1 2026 results, Lambda’s $1.5B+ financing announcement, Crusoe’s Series E announcement, Nscale’s Series C announcement, and The Wall Street Journal on TensorWave’s funding round.
Other key infrastructure sources include: Global Infrastructure Partners on the Aligned acquisition, Vantage on its $9.2B equity investment, DataBank on Red Oak construction financing, and EdgeCore on its 1.8 GW delivered or under-development capacity.
Power and grid sources include: Constellation on the Three Mile Island restart, Meta on nuclear energy projects, Bloom Energy on the Oracle partnership, Eaton’s Q1 2026 results, Vertiv’s Q1 2026 results, and GE Vernova’s Q1 2026 results.
Cooling, chips, and software sources include: Ecolab on the CoolIT acquisition, KKR on the CoolIT sale, Eaton on the Boyd Thermal acquisition, Accelsius on its Series B, Nvidia’s Q1 FY2027 results, Broadcom’s Q1 FY2026 results, Marvell on the Celestial AI acquisition, and Phaidra’s Series B announcement.
This chart, featured in our data center market deck, shows the revenue mix across customer segments in the data center market
Related blog posts
- How strong is fundraising in the data center market right now?
- The startups that have raised the most funding in the data center market
- The most highly valued startups in the data center market
