I spent a morning last week trying to price a deal that has no price.
The headline was clean. CoreWeave โ the GPU cloud company that listed on NASDAQ in March 2025 โ signed a 240-megawatt capacity agreement with AdaniConneX, the data center joint venture between India's Adani Group and EdgeConneX. First CoreWeave footprint in India. The coverage framed it as a "strategic expansion" into "one of the world's fastest-growing AI markets."
What the announcement did not contain: a transaction value. A structure โ lease, capacity reservation, or equity joint venture. A delivery timeline. A named anchor customer. A power source.
I audited ICO whitepapers by hand in 2017, from a Telegram group in Bangkok, and flagged eight of fifteen as hollow. The tell was never in what the projects said. It was in the numbers they left out. Code doesn't lie, but narratives do โ and a narrative with no price tag is a narrative you cannot stress-test. That instinct has not aged. It applies just as cleanly to a NASDAQ-listed infrastructure company as it did to a no-name token during the mania.
So let me do what I do. Let me read the numbers they left out.
CoreWeave is not a general-purpose cloud. It is an AI compute wholesaler, and that distinction is the entire lens through which this deal should be read. AWS and Azure sell you a menu: vCPUs, storage tiers, databases, managed services, priced per unit of compute. CoreWeave sells you power. Its capacity is denominated in megawatts, not in H100-equivalents. That is not a marketing quirk. It is an admission of what actually constrains the business.
The binding constraint in AI infrastructure is no longer silicon procurement. It is power availability โ the ability to source, deliver, and cool hundreds of megawatts at a single site, on a schedule, at a defensible price. Once you internalize that, the 240MW figure stops looking like a GPU count and starts looking like a utility contract wearing a tech costume.
AdaniConneX matters here for the same reason. It is a joint venture between Adani Group โ India's infrastructure and power conglomerate โ and EdgeConneX, a US data center developer. Adani brings land, grid relationships, and political capital. EdgeConneX brings the operating playbook. This is the classic local-partner, foreign-operator template, and it exists precisely because foreign hyperscalers cannot replicate Indian land-and-power acquisition on their own timelines.
India, meanwhile, has become the frontier everyone wants to claim. AWS operates in Mumbai and Hyderabad. Azure in Pune and Chennai. Google Cloud in Mumbai and Delhi. Reliance Jio is building AI data centers with NVIDIA. Tata, Yotta, and CtrlS run domestic capacity. Into this crowded field walks CoreWeave, announcing 240 megawatts it has not yet built.
Now โ the analysis. And I want to be forensic about it, because this announcement is engineered to resist exactly that.
240 megawatts is not a size. It is a schedule, and the schedule is missing.
Here is how to think about scale. CoreWeave's US campuses run individual sites in the 100โ300MW range, and its contracted portfolio with major customers is reported in the gigawatt range. Against a gigawatt-scale backlog, 240MW is a single offshore flagship โ important, but not company-defining. It lands somewhere around ten to twenty percent of total contracted capacity, depending on how you count.
That is the first thing the headline hides. A number that sounds enormous in isolation becomes an increment in context. Alpha hidden in the noise, but also noise hidden in the alpha โ a big number can be a small number depending on the denominator you never see.
The second thing: megawatts are a promise about the future, not a measurement of the present. A signed 240MW commitment is a commitment to deliver power and cooling to racks that will hold GPUs that have not been purchased, on a timeline that has not been disclosed. From signature to production, cross-border data center builds run twelve to twenty-four months on a good day. On a bad day, they run longer, cost more, and slip the revenue they were supposed to generate.
I learned this lesson the expensive way in DeFi Summer 2020. I tested liquidity mining strategies on my own capital and lost fifteen percent to impermanent loss before I understood that the yield I was chasing was a schedule, not a return โ it depended on conditions that could change before I captured it. Capacity announcements work the same way. The megawatts are the yield. The timeline is the impermanent loss you do not model until it has already happened.
A lease, a capacity reservation, and an equity joint venture are three completely different balance sheets wearing the same headline.
If this is a lease, CoreWeave carries an operating commitment โ a future obligation, not a present asset. If it is a capacity reservation, CoreWeave has secured optionality it may or may not exercise. If it is a joint venture, CoreWeave has committed capital expenditure and shares construction risk with Adani. Each structure produces a different cash flow profile, a different CapEx burden, and a different story for investors.
The announcement used none of those words. That is not an accident. It is a deliberate choice to keep the financial exposure ambiguous while the narrative does its work.
And the CapEx math is not small. At a rough industry benchmark of several million dollars per megawatt of fully built AI capacity, 240MW implies hundreds of millions to low billions in total investment โ split, financed, or committed in ways we cannot see. For a company that is still proving it can fund its existing buildout, that is material.
Here is the part that decides whether this deal is real.
CoreWeave's business model is customer-pulled, not capacity-pushed. Its revenue is concentrated in a handful of large, multi-year, take-or-pay contracts. The company builds where its anchor customers tell it to build. This is not a weakness in itself โ it is how the wholesale compute business works โ but it means the India site lives or dies on whether a Microsoft-scale or Anthropic-scale customer has committed to buy there.
The announcement named no customer. That absence is louder than any of the words that were present.
My read: the India site is most likely built to serve multinational AI customers who need to process data inside India for regulatory reasons โ the Digital Personal Data Protection Act of 2023 imposes localization constraints that make in-country processing attractive or mandatory for certain workloads. It is almost certainly not built for Indian AI startups. The domestic demand base is not yet large enough to fill 240 megawatts at the prices CoreWeave needs to charge. If you are an Indian AI founder reading this and hoping this deal means cheap local compute, temper that hope. The economics point the other way.
Now the risk almost no one is pricing.
CoreWeave's competitive edge has historically rested on its priority access to NVIDIA GPUs โ a relationship close enough that the industry jokes about it. But India is not the United States. US export controls on advanced AI chips to India have tightened, and H100/H200-class exports require licensing. That constraint applies to CoreWeave and to every competitor equally, which sounds like a level playing field until you realize it erodes exactly the advantage CoreWeave was counting on.
A data center with power and no chips is a very expensive building. The geopolitical risk here is not that India says no. It is that Washington says not yet, and the megawatts sit idle while the depreciation clock runs.
This is where my cross-chain instincts kick in.
I have spent years watching Cosmos. IBC โ the Inter-Blockchain Communication protocol โ is technically elegant, arguably the cleanest interoperability design in the industry. And yet the application ecosystem fragmented, and ATOM captured almost none of the value flowing through the rails it built. Elegant infrastructure does not automatically capture the economics of what runs on top of it.
The same question haunts this deal. Who captures the value of CoreWeave's India footprint? CoreWeave brings the AI-specific architecture, the customer relationships, and the operating expertise. Adani brings the land, the power, and the government relationships. In India, the scarce inputs are the local ones โ power, permits, political cover. CoreWeave's uniqueness degrades into the role of an operator renting someone else's moat.
That is not fatal. Operators can be very profitable. But it reframes the deal from "CoreWeave conquers India" to "CoreWeave rents a position in India," and those are different investment theses. The lesson is that the party who owns the scarce layer takes the economics. In this build, Adani may own the scarce layer.
AdaniConneX serves other customers. It will keep serving them. Nothing in the announcement suggests CoreWeave's 240MW is exclusive, and there is no reason to assume it is.
That matters because the India GPU cloud market is filling up fast. Hyperscalers have multi-year head starts. Domestic players like Yotta and E2E Networks already operate. And CoreWeave's peers โ Lambda, Together, and others โ are all evaluating the same market with the same logic. If this deal is a signal, expect followers. If it is a first-mover position, expect it to be contested within a year.
The "strategic expansion" framing quietly omits the word "competitive." That is the genre. Announcements report the opportunity and leave the odds on the cutting-room floor.
240 megawatts is roughly the consumption of a mid-sized city. India's grid is coal-heavy, and its green-power procurement requirements for data centers are tightening. Whether this site runs on renewables, and at what cost, is an open question the announcement did not touch. For ESG-sensitive capital, that is not a footnote.
There is a sovereignty angle too. India's data localization regime is a genuine tailwind for in-country compute, and the IndiaAI Mission signals government appetite for exactly this kind of capacity. A foreign operator building local capacity, with a domestic power partner, is politically legible. I would not be surprised to learn there is quiet state support behind it.
But here is the crypto-native alarm bell, and I want to ring it clearly. Every deal like this concentrates AI compute into fewer hands. A handful of operators, a handful of power partners, a handful of chip suppliers. The industry that spent a decade preaching decentralization is watching the most consequential infrastructure buildout of the century centralize in real time โ and cheering, because the tickers go up.
Finally, the part that is most legible if you know what you are looking at.
CoreWeave went public in March 2025. Public companies need to keep telling a growth story, quarter after quarter, and capacity announcements are how compute companies feed that story. A 240MW India deal announced in a post-IPO window has a clear capital-markets function: it signals "the growth is real and it is global" to investors who are pricing a discounted cash flow model built on future capacity.
This is a narrative asset before it is a financial asset. If the deal has no disclosed value, no timeline, and no named customer, it is very unlikely to be generating cash flow yet. Its job is to support the story, not the P&L. That does not make it dishonest. It makes it a communication, and communications should be read as communications.
The trap for investors is linear extrapolation โ MW equals growth equals value. Compute is not a straight line. It is a schedule, a customer commitment, and a financing structure, and two of those three are missing from this announcement.
Here is where I will argue against the crowd โ including against my own instinct to be skeptical.
The reflexive read is that this deal is a defensive, story-driven hedge dressed as expansion. And that is probably true. But "defensive" is not the same as "bad." If CoreWeave's largest customers are increasingly required to process data inside specific jurisdictions, then a non-US node is not optional โ it is table stakes. Building it early, even expensively, may be the cheaper path than scrambling later. Hedges cost money by design. The question is whether you need the hedge at all.
The deeper contrarian point is aimed at my own tribe. The crypto industry keeps insisting that decentralized compute โ Render, Akash, io.net โ will democratize the AI infrastructure layer. Watch this deal closely and you see the opposite. A 240MW build cannot be decentralized. It requires one power contract, one land parcel, one operator, one grid interconnection. The physics of AI compute at scale is centralizing, and decentralized networks are competing for workloads this deal cannot even see.
Trust is the new currency โ and right now, trust in AI compute is being issued by a very small number of balance sheets. The decentralization thesis is not wrong. It is just running on a different track from the one the money is actually on.
So what should you watch, if you want to know whether this deal is real?
Watch the next earnings call for two things: an upward revision to CapEx guidance, and a disclosed production timeline for the India site. If neither appears, the deal is a narrative. If both appear, it is a buildout.
Watch whether CoreWeave templates this structure โ local partner, foreign operator, MW-denominated โ into Southeast Asia or the Middle East. One deal is an announcement. Three deals is a strategy. And watch whether any anchor customer puts its name on the India capacity. A signed customer turns this from a press release into a contract.
The megawatts are not the story. The numbers they left out are.
