It's 2028, and the most likely outcome for Solowin's 100 megawatts is that no one is counting them anymore. That is the pre-mortem. Before the bull case, before the market sizing, before the inevitable CoreWeave comparisons: the base case for an infrastructure target announced four years ahead of delivery, with no site, no interconnection agreement, no anchor tenant, and no disclosed capital partner, is that it quietly becomes a slide in a deck a company stopped updating.
I have run this drill before. During the 2022 Terra collapse I published within 48 hours because the incentive math was already broken in public; the only thing missing was someone willing to say it out loud. Here the situation is the inverse. The information is thin, not wrong. But thin information married to a long timeline is its own category of risk, and the Crypto Briefing item that surfaced Solowin's target carries roughly five usable facts. Hunting for the story that defines the next cycle requires recognizing when a story has a headline but no manuscript. This one does.
The arithmetic is the first honest witness.
100MW of IT load at current AI rack densities implies roughly 67,000 H100-class accelerators running at 700W each. Add cooling, distribution losses, and networking, and a facility at an industry-typical PUE of 1.3 draws closer to 130MW gross from the utility. Shell-and-core construction for that footprint typically runs $150M to $300M depending on whether you retrofit a mining site or pour a greenfield slab, with the silicon sitting on top as a separate and usually larger line item.
None of those numbers disqualify Solowin. They define the entry ticket. The problem is that the entry ticket is the easy part, and everything after it is where projects die.
Solowin's announcement, as reported by Crypto Briefing, couples a 2028 delivery target with the unremarkable observation that demand for scalable compute is rising, and it lands beside the well-worn note that the United States Congress remains stalled on crypto legislation. Four facts, one adjective, zero disclosures. No site location. No utility partner. No GPU vendor commitment. No customer. No financing. No token โ which means no token economics, no unlock schedule, no supply structure to stress test, and no way to run the incentive analysis that has correctly flagged almost every structurally broken model I have written about since 2020.
That absence is informative rather than merely inconvenient. DePIN and decentralized compute projects almost always lead with token design because the token is the product. When a company distributing through crypto media does not mention one, you are looking at either a traditional infrastructure developer borrowing an audience, or a pre-token entity still deciding what it wants to sell. Both are legitimate. Neither is analyzable from the outside.
Historical context matters. The 2021 cycle monetized hash rate and power contracts. The 2022 cycle monetized the promise of decentralized physical infrastructure before most of it shipped. The 2023โ2024 cycle monetized compute, and the crucial difference is that compute demand, unlike the prior two, was and remains real. NVIDIA's data center revenue trajectory was not a narrative; it was a queue. That is exactly why compute became the preferred wrapper for projects that needed a credible story โ and exactly why the wrapper now carries a premium the underlying asset doesn't always deserve.
The regulatory moat question runs opposite to where most readers expect.
When I built the compliance-first disclosure template with legal partners in Singapore and Vancouver in 2025, the pattern across roughly 30 early-stage projects was consistent: teams that cleared regulatory ambiguity fastest were the teams with revenue, because revenue forces you to choose a jurisdiction and a disclosure posture. A 2028 target with no customers has no reason to resolve that ambiguity, and therefore no regulatory moat โ only regulatory fog.
There is strategic logic to the fog, though, and I suspect it is deliberate. Positioning as "AI compute infrastructure" rather than "crypto mining" changes how state utility commissions, interconnection authorities, and lenders treat you. Mining carries an established regulatory and ESG posture, some of it openly hostile. AI compute currently carries a favorable one. If Solowin is converting an energy-intensive crypto-adjacent footprint into AI hosting, that reframing is not marketing. It is a permitting strategy with a legal department behind it.
The binding constraint on AI compute in 2026 is not GPUs. It is the interconnection queue.
In most US power markets outside ERCOT, the queue for large-load interconnection is measured in years rather than quarters. Northern Virginia, the densest data center market on earth, is effectively capacity-constrained. I have watched projects with signed offtake agreements stall for 30-plus months waiting on a single substation upgrade, burning opex the entire time and renegotiating delivery dates with customers who had already built their launch plans around them. A 2028 target implies two to four years of construction and commissioning, which leaves almost no slack between today's press release and the first energized rack.
Then the capital question. Microsoft, Google, and Meta collectively deployed well over $200 billion in capital expenditure in 2024 alone. CoreWeave crossed the gigawatt threshold with disclosed contracted backlog. Lambda, Foundry, and a dozen regional operators are competing for the same power, the same EPC contractors, and the same transformers โ hardware with lead times that would embarrass a semiconductor fab. A $150Mโ$300M facility is real money and simultaneously a rounding error in that field.
Capacity without a tenant is not a compute business. It is a real estate option.
Neocloud economics depend on multi-year offtake contracts signed before construction, because no lender will finance speculative AI capacity at these rack densities. The single most important undisclosed fact in the entire Solowin story is whether an anchor tenant exists. If one does, this becomes a project finance exercise with a visible revenue line and a bankable model. If one does not, the announcement is an option on future demand, purchased with shareholders' patience.
Second-order effects are modest but worth naming. 67,000 accelerators is meaningful to a regional grid and negligible to NVIDIA's order book โ Apple or Google can absorb more than a hundred times that volume in a single year. The real transmission runs into regional power markets, into data center REIT competition, and into the residual pool of crypto mining sites whose most valuable asset has quietly become the substation, not the machines bolted to the racks.
Here is the contrarian read, and it cuts against both sides.
The consensus bear case is that Solowin cannot execute. The consensus bull case is that demand will paper over execution risk. Both miss the mechanism. What kills these projects is not technical incompetence โ it is the order in which commitments must arrive. Signing an offtake contract requires a credible energization date. A credible energization date requires an interconnection agreement. An interconnection agreement requires site control and deposits. Each step depends on the previous one, and none of them can be borrowed from a press release. The sequence is the moat, not the megawatts.
The second contrarian point is about the narrative itself. "Compute scarcity" is being sold the same way "liquidity fragmentation" was sold in DeFi โ as a structural crisis only a new generation of products can solve. In DeFi, the fragmentation argument was manufactured by funds holding positions in the products that resolved it. Compute scarcity is a genuine phenomenon, but the framing around it has the same vendors, the same conference circuit, and the same fundraising cadence.
And there is a recycling pattern I have flagged before. When capital rotates into a hot sector, labels get reapplied faster than infrastructure gets built. That is precisely what happened to Bitcoin Layer 2s, where the overwhelming majority of what markets under that name is Ethereum infrastructure wearing a new logo. Compute is now attracting identical behavior from operators who understand the vocabulary better than they understand the busway. Hunting for the story that defines the next cycle means separating the label from the load-bearing structure beneath it.
Three signals are worth tracking, and everything else is noise. An interconnection agreement or utility filing is the first unforgeable proof of intent โ it costs real deposits and cannot be fabricated in a blog post. A named anchor tenant with disclosed contract duration is the second, because it converts capacity into revenue and revenue into a model that can be stress-tested. The eventual emergence of a token would turn a press release into something with an unlock schedule, a supply structure, and an incentive analysis worth running. Until one of those lands, Solowin is not a compute company. It is a hypothesis with a power figure attached, and the next four years will decide whether the megawatts were built or merely announced. Hunting for the story that defines the next cycle, the answer will arrive in filings, not press releases.
