CoreWeave's 240MW India Bet: Megawatts as the New Unit of Account

CoreWeave just signed a 240-megawatt deal with AdaniConneX, its first entry into India. The headline number is the megawatts. The number that actually matters is that CoreWeave reports capacity in megawatts at all. No GPU count. No H100-equivalent. Just power. That is not a marketing choice — it is an accounting one, and it tells you exactly what the binding constraint in AI compute has become. Follow the gas, not the hype. Three years ago the same logic ran in reverse: Ethereum miners measured their fleets in hashrate and kilowatt-hours, because the scarce input was electricity, not silicon. CoreWeave is the company that learned that lesson first, and it never unlearned it. CoreWeave did not start as an AI company. It started as Atlantic Crypto, an Ethereum mining operation that accumulated GPUs and cheap power contracts. When Ethereum's Merge killed proof-of-work mining in September 2022, the firm did not liquidate. It repointed the same hardware, the same data-center leases, and the same power relationships at AI training workloads. The rebrand to CoreWeave was a reclassification, not a reinvention. That history explains the current transaction. AdaniConneX is a joint venture between the Adani Group and EdgeConneX, one of India's largest data-center platforms. Adani brings land, grid interconnection, and government relationships; CoreWeave brings AI-specific operations and hyperscale customer contracts. It is the standard local-partner, foreign-operator structure. CoreWeave listed in March 2025, which matters: public companies need a capacity-growth story every quarter, and India is a new chapter in that story. The evidence chain starts with the unit of account. CoreWeave quotes megawatts because its true constraint is power availability, not chip procurement. Traditional clouds quote vCPUs and GPU SKUs. CoreWeave quotes the thing it actually competes for. When I built the standardized schema for the 2024 ETF data framework — mapping over 10,000 blockchain addresses to KYC-verified entities — the lesson was identical: the unit you choose to measure reveals what you believe is scarce. CoreWeave believes power is scarce. The market should price that belief. Then there is the disclosure gap. The announcement confirms three things and omits five. Confirmed: a counterparty (AdaniConneX), a capacity figure (240MW), and a market (India). Omitted: transaction structure, timeline, anchor customer, capital expenditure, and GPU sourcing. Each omission is material. Structure first. Is this a lease, a capacity reservation, or an equity joint venture? A lease is off-balance-sheet; a JV drags capital expenditure onto CoreWeave's books. The two have opposite implications for a company already carrying heavy debt. Timeline second. A 240MW commitment with no delivery schedule is financially inert. Data-center builds run 12 to 24 months from signing to energized capacity, and cross-border builds run longer. Customer third. CoreWeave's revenue is highly concentrated among a handful of hyperscale clients on multi-year, take-or-pay contracts. Its expansion model is customer-pulled, not capacity-pushed. Without an anchor tenant, a 240MW site risks becoming stranded capacity. Scale fourth. Against a contracted portfolio reportedly in the gigawatt range, 240MW is roughly 10 to 20 percent depending on the denominator — meaningful, not decisive. Sourcing fifth. India's grid stability, green-power availability, and cross-border bandwidth are known friction points. And US export controls on advanced AI accelerators to India add a policy risk that no amount of real estate can hedge. A site with power but no chips is a very expensive substation. Quantify the manipulation here means quantifying the narrative. The press framing is global expansion. The mechanical reading is defensive: multinational clients subject to India's DPDP Act and EU data-localization rules need in-country compute, and CoreWeave is following them. That is a rational move. It is not the same as an offensive land grab. The tempting inference is that more megawatts equal more revenue equal more value. That is a linear extrapolation across three links that do not hold. Megawatts are not usable compute — power, cooling, networking, and GPU supply must all clear before a watt becomes a token. And a capacity announcement is not a signed revenue contract. The market reliably bids up capacity news because CoreWeave's valuation rests on a discounted-cash-flow story of future capacity times unit price times margin. Announcements feed the story; they do not feed the cash flow. DeFi efficiency is math, not marketing — and so is data-center economics. The second blind spot is competitive. India already hosts AWS, Azure, Google Cloud, and Alibaba, plus domestic players like Reliance Jio, Tata, Yotta, and CtrlS. AdaniConneX serves other tenants too; the 240MW may well be non-exclusive. CoreWeave's historic edge came from priority access to NVIDIA supply. Export licensing to India could neutralize that edge for everyone, which is a strange kind of competitive parity. Data does not care about the press release. Correlation between a capacity headline and durable earnings is not causation. Watch the next earnings call. If CoreWeave raises capital-expenditure guidance and discloses an India delivery schedule and an anchor tenant, the deal is real. If it stays a narrative line item, it is story maintenance. The deeper signal is whether India becomes a template for non-US capacity — Southeast Asia, the Gulf — or a one-off. Follow the gas, not the hype. The megawatt is now the ticker.