Why power matters
Understand electricity as the ceiling.
← Back to Compute College
Compute College
What is a Megawatt of AI Compute?
A megawatt of AI compute is a power-based way to describe possible data-center and accelerator capacity.
Plain-English definition
A megawatt of AI compute describes one million watts of electrical capacity allocated to operating AI infrastructure, including servers, accelerators, networking and, depending on how a statement is framed, facility overhead. It is not a fixed GPU count, but it helps readers size a buildout.
Memory trick: Megawatts are the fuel tank of the AI compute market: they show potential range, but not whether the vehicle is assembled and moving.
Why it matters
AI infrastructure growth increasingly depends on large, reliable blocks of electricity. Companies often describe campuses in megawatts or gigawatts because power can determine how much equipment may be energized, how quickly deployment occurs, and which regions can support additional supply.
- A power commitment can indicate infrastructure ambition before exact accelerator deployment is known.
- Grid, cooling, construction, and equipment constraints decide whether reserved power becomes useful compute.
- Megawatts provide a common scale language for comparing facilities, projects, and regional constraints.
Simple example
Assume a site has 10 megawatts of available total power and that, under an illustrative facility-overhead assumption, 80% is available for IT load. That leaves 8 megawatts for servers and networking. If a specified AI rack averages 100 kilowatts, the simple arithmetic suggests about 80 such racks before considering redundancy, layout, equipment, cooling, or operational reserve.
- Power-to-rack conversions require explicit assumptions about IT share and rack draw.
- Different GPU generations and system configurations can change power consumed per rack.
- The estimate describes possible scale, not confirmed operating capacity or public buyer access.
Example figures are illustrative calculations, not current quoted market prices.
Market signal
How to read the market signal
New power commitments can point to future AI capacity, while interconnection delays, substation work, utility limits, or behind-the-meter plans reveal how soon that capacity may become usable. Power news should be read as a timeline signal as well as a size signal.
- An energized megawatt with deployed hardware is a stronger near-term supply signal than a planned allocation.
- Large projects can affect buyer expectations about future supply and infrastructure competition.
- Power scarcity may support premiums for capacity already running in constrained regions.
Market read: megawatts measure opportunity, not completed compute. Combine power scale with timing, equipment, cooling, and access before translating an announcement into available GPU supply. Figures here are illustrative unless explicitly sourced and dated — see our methodology.
Common mistake
Do not convert megawatts directly into an exact number of GPUs without showing assumptions. Accelerator generation, server arrangement, rack density, PUE, cooling, network equipment, redundancy, and reserved operating margin all affect how much useful compute a power block supports.
Practical takeaway
What you can do with this
Use megawatt announcements to size infrastructure ambition and ask the next questions required for a supply read. Buyers should focus on capacity available during their workload window; analysts and journalists should state whether a project is proposed, connected, energized, equipped, or operating.
- Analysts: record capacity size, energization date, facility status, and hardware disclosure separately.
- Buyers: ask what operating cluster capacity can actually be reserved, not merely what site power is planned.
- Investors: connect power scale to construction, cooling, interconnection, and customer demand evidence.
- Operators: keep assumptions explicit when translating power into supported racks or services.
Decision check: never present an estimated GPU count from power alone without the IT-load, rack-density, overhead, and operating-readiness assumptions behind it.
Compute College
Turn the lesson into a number
Use the GPU-Hour Cost Calculator, AI Training Cost Calculator, or Model Serving Cost Calculator.
Compute College track
Power & Data Centers
Step 14 of 17: What is a megawatt of AI compute