Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Omdia analyst Vlad Galabov says global data-center capital expenditure could surpass $1 trillion by 2030, and that estimate may still be conservative. The financial challenge is not simply finding that much money: operators must secure power, remove heat from increasingly dense AI hardware, build supply chains, and hire people who can run the systems. Galabov’s conclusion is practical rather than fatalistic: “most of the problems are solvable.”
What Galabov’s trillion-dollar forecast actually says
Data Center Knowledge reported Galabov’s forecast after speaking with him at Data Center World 2025 on April 24, 2025. The report says global data-center capex could surpass $1 trillion by 2030, with Galabov warning that the figure could be conservative.
That is a capital-expenditure projection for the worldwide data-center industry, not a quoted construction budget for one facility, a guaranteed annual spending total, or a price for an AI data center. The interview does not specify whether the figure is cumulative or an annual run rate, so it should not be converted into a per-site or per-rack cost.
For personal-finance readers, the useful interpretation is that AI infrastructure is becoming a very large capital-allocation problem. The forecast identifies the scale of money that may be deployed; it does not establish what consumers will pay for cloud services, electricity or hardware.
#1 Best Overall
Omdia identifies Galabov as its Senior Research Director for Enterprise Infrastructure. His remit includes cloud and data-center research, and he developed Omdia’s data-center capex and capacity model, according to Omdia’s analyst profile.
Why spending alone cannot remove the bottlenecks
Power distribution is a physical constraint
AI clusters concentrate far more computing in each pod than many traditional enterprise workloads. That raises the amount of electricity that must reach a site, but also the engineering challenge of distributing it reliably inside the facility. Galabov names power distribution as one of the industry’s biggest hurdles.
More capital can fund substations, switchgear, backup systems and new generation, but those projects still depend on equipment availability, permitting, interconnection studies and skilled crews. The interview points to self-generated energy as one response, including on-site natural-gas generation for facilities that cannot wait for a conventional grid connection.
Cooling must keep pace with rack density
Higher-performance AI racks turn more electricity into heat in a smaller physical footprint. Air cooling alone may not be practical at the highest densities, so operators are moving toward liquid-cooling designs. The transition affects the entire heat-removal chain rather than just the server chassis.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Rank #2
Talent is an infrastructure input
Galabov also identifies the talent crunch as a major hurdle. Designing, commissioning and operating high-density power and cooling systems requires electrical, mechanical, controls, networking and data-center operations expertise. A facility can be funded and physically built yet remain constrained if it cannot recruit or retain the people needed to run it safely.
Tariffs, supply shocks and geopolitics add volatility
Tariffs, supply-chain disruptions and geopolitical uncertainty can change the timing and cost of equipment. Those risks affect transformers, power electronics, cooling components, servers and other specialized inputs. They make a long-term capex forecast less like a single invoice and more like a sequence of interdependent projects exposed to changing prices and lead times.
How AI facilities may change by 2030
On-site generation and microgrids
A facility that cannot obtain enough grid capacity may combine utility power with generation on the property. The forward-looking discussion in “How Will Data Centers Be Different in 2030?” describes on-site natural-gas generation as one possible approach.
Omdia’s discussion of the AI build-out also emphasizes microgrids, AI megaclusters and efficiency. Its AI and data-center presentation frames local power systems as part of the response to the boom, while the Data Center Asia 2025 summit material says AI adoption requires tailored power and cooling solutions.
Free tools Windows power users keep installed
One-click scans. No signup required.
A microgrid can improve control over generation and resilience, but it does not make fuel, permitting, maintenance or emissions considerations disappear. It is a power strategy, not proof that a site has unlimited capacity.
Liquid cooling as a system, not a single product
The 2030 discussion describes several parts of a liquid-cooling architecture:
- Cold plates that transfer heat directly from processors or other high-power components.
- Connectors and manifolds that route coolant through racks and rows.
- Cooling-distribution units (CDUs) that manage the interface between facility water systems and IT equipment loops.
- Two-phase fluids that absorb heat through phase change in some direct-chip designs.
Standardized connectors and cold plates could make equipment easier to deploy across vendors. However, the same discussion notes supply-chain limits for direct-chip two-phase cooling and says the market needs more vendors and greater manufacturing scale. Liquid cooling therefore addresses heat transfer, but it does not by itself solve component shortages, installation capacity or operations staffing.
Can data centers get enough power for AI?
Sometimes, but not through a universal formula. The answer depends on the site’s grid connection, the timing of utility upgrades, the power profile of the compute cluster, backup requirements and whether the operator adds generation or a microgrid.
Rank #4
Galabov’s framework is to treat power availability or generation as a first-order design decision. A project that starts with a building and tries to add electricity later can be delayed by the grid. A project that plans generation on-site may gain schedule control but takes on fuel supply, equipment and operating responsibilities.
Neither approach guarantees that every proposed AI campus can be energized. The available evidence supports a range of strategies, not a single capacity number for all facilities.
What cooling does a 600 kW rack need?
There is no single cooling specification that can be derived from “600 kW rack” alone. The required design depends on the rack’s thermal profile, coolant type, supply and return temperatures, allowable component temperatures, redundancy target, room layout and the facility’s water and heat-rejection systems.
The relevant takeaway from Galabov’s discussion is that a rack at this density would be evaluated as part of a liquid-cooling system: cold plates or another direct-to-chip method, reliable quick-disconnects, manifolds, a CDU and a facility loop. Two-phase approaches are another option, but the interview highlights their current supply-chain and vendor-scale constraints. A site engineer must size pumps, heat exchangers, controls and redundancy from actual equipment specifications; a headline rack number is not enough.
Recommended Free Tools
A financial and operating decision framework
Operators comparing an AI data-center strategy can use four tests. The evidence below reflects the issues Galabov and the related Omdia material emphasize.
| Decision area | Questions to answer | What the available evidence indicates |
|---|---|---|
| Power | How much firm capacity is available, when will it arrive, and is on-site generation or a microgrid required? | Power distribution is a leading bottleneck; self-generation and microgrids are identified responses, not guarantees. |
| Cooling and density | What rack densities will be deployed, and can the facility support direct liquid cooling and heat rejection? | High-density AI hardware is driving liquid-cooling adoption, including cold plates, CDUs, manifolds and connectors. |
| Standardization and supply chain | Are components interchangeable, and are qualified suppliers available at the required volume? | Standardization can ease deployment, while two-phase cooling and other specialized parts face scale and supply constraints. |
| Workforce and operations | Who will design, commission, monitor and repair the power and cooling systems? | The talent crunch is one of the named industry hurdles; staffing is part of the infrastructure plan. |
| External exposure | How sensitive is the schedule to tariffs, geopolitical events and supplier interruptions? | These factors can disrupt equipment availability, delivery dates and project economics. |
Galabov’s workload advice still matters
In an earlier Data Center Knowledge interview, Galabov offered colocation providers a cost-control principle: “not every workload requires the latest technology, and not every workload requires a new server.” The full February 4, 2022 interview is available at Data Center Knowledge.
That advice is relevant to the trillion-dollar challenge because demand should be matched to hardware and facility design. Reserving the newest, densest systems for workloads that genuinely need them can reduce unnecessary pressure on power, cooling and scarce equipment. It does not eliminate AI growth, but it can prevent every application from being treated as an extreme-density deployment.
What the forecast means—and does not mean—for households
The $1 trillion figure signals the scale of infrastructure investment under consideration through 2030. It does not, by itself, predict a household’s electricity bill, a cloud subscription price, a particular company’s profit or a guaranteed return for investors. Those outcomes would depend on local utility rules, project financing, utilization, technology prices and many other facts not established in the interviews.
For anyone evaluating a data-center company or infrastructure fund, the practical questions are therefore operational: Does the business have secured power? Can it obtain and standardize cooling equipment? Has it planned for supply interruptions? Can it hire the required specialists? A large capex market can still produce poor returns when projects are delayed, underutilized or overbuilt.
Bottom line
Vlad Galabov’s forecast is best read as a warning about execution, not merely a headline about spending. Global data-center capex could exceed $1 trillion by 2030, but power distribution, liquid cooling, supply-chain scale and skilled labor determine how much of that capital can become usable AI capacity. Self-generation, microgrids, standardized liquid-cooling components and workload discipline are practical tools; none removes the need for site-specific engineering and disciplined financial decisions.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
