A Full Stack AI Infrastructure Platform

1The proposal

• There is a short window of opportunity to build a large full stack AI infrastructure company.
• The proposal platform combines Chinese speed, efficiency and costs, financial structuring and capital efficiency, as well as deep western capital markets, to build a full stack AI infra platform.
• The proposal combines two playbooks: DayOne, the western capital funded Singapore datacenter builder that is of Chinese origin, whose 2024 first raise of US$587mm had only 70 MW of active sites, and Volta, which went from 0 to US$2.4bn in valuation in eight months, starting from scratch and with only two financiers.

2The picture

Two stacks under one platform. The developer stack finds, owns and builds the sites, and comes first. The cloud stack owns the chips and sells the compute, and is validated in parallel while the first site is built.

3The play, in order

Running in parallel from step 2:

• Hiring. Power is the recruiting asset. A serious operator will not take the call without a named site.
• The two demand tracks. A tenant or a neocloud onto the lease, and an end compute buyer for our own cloud. Only one has to land by the time the building is online.

4Why now

• Compute demand has gone parabolic. Anthropic alone has just contracted about US$80bn of compute.
• The bottleneck is buildout, not demand. Everyone wants capacity and almost nobody can deliver it on time.
• The market is not paying for installed megawatts. It is paying for pipeline, structure and speed.

5Why us

• The build. Wendy, Terrence and the backers deliver capacity at a speed and a cost the market cannot match.
• The pipeline. Sites and power across several countries, already in hand and under marketed.
• The capital and the structuring. Alan brings the Western networks and the structuring that makes those sites financeable by investors who cannot reach them today.
• Up the stack is where the value is. You can always move up the chain and the chips are the most expensive part of it. A vertically integrated player is the most competitive one.

6What is in, and what is out

• Vietnam is not an asset in this proposal. Western capital does not clear it. Even with a platform that is not majority backed from China, Vietnam needs a licence per shipment and the rules are strict today. We are not betting the base case on that.
• Vietnam, separately: get the capital back quickly. Let the building, potentially to MiniMax, then an equity recapitalization or a strategic buyer. It should not tie up their capital for years. Colocation, not a cloud.
• Limited PRC voting control by design. The chip rules test who controls the company and where its ultimate parent sits, not a percentage of economics. The backers hold economics without votes, and control sits with a non PRC founder.
• In: the sites and the power outside Vietnam, the build capability, the Western capital, and the route to chips.

7Next steps

• Discuss this and settle the changes.
• Build the financial models so we can see how it lays out.
• Then Alan puts the deck together, takes it to the backers, and runs the financing conversations through to signing.
• Still open: the ask and the amounts. The backers' percentage at the platform. What sourcing earns. The founders' split and the option pool. Which infra investor takes the site majority.

Appendix 1. The two precedents

DayOne, the developer playbook

• At its first raise, March 2024: US$587mm of Series A convertible preferred. Over 70 MW already revenue-generating, over 200 MW committed or reserved by customers, 330 MW in service and under construction, and 340 MW held for future development, across Hong Kong, Singapore, Johor and Batam.
• By January 2025: renamed from GDS International to DayOne. About 480 MW in service and under construction, plus about 590 MW held for development.
• Today: over 1.5 GW of total bookings across Asia Pacific and Europe since inception in 2022.
• The money: Series C closed at US$4.5bn of gross proceeds on 5 June 2026, led by Coatue and Hillhouse, with the Indonesia Investment Authority and Achi Capital Partners among the new investors. The initial close in January 2026 implied about US$10bn. By the final close, reports put it near US$20bn, which is also the target for a US listing.
• What GDS got. GDS, a PRC-listed operator, was diluted to 35.6% at the Series B and now holds a minority it does not control. On 13 January 2026 DayOne repurchased US$385mm of GDS's shares at the Series C price. In GDS's own words that recycled about 95% of its invested principal at nearly 6.5 times money, and its remaining stake was implied at over US$2.2bn, or US$11.18 per American Depositary Share.

Volta, how to build an enterprise from nothing

• Incorporated in January 2026 with one pound sterling of share capital. Valued at US$2.4bn in August 2026 on a US$300mm raise, with NVIDIA and Dell among the backers and a reported US$10bn compute partnership. Eight months.
• It started with no assets, no employees and no operating history, and it still has no revenue. It earns nothing until the capacity comes online, targeted at the end of 2026 and the first quarter of 2027.
• It took 121 MW IT of capacity from a third party landlord, Bitdeer, at Tydal in Norway. Two contracts, a colocation lease and a services agreement, executed on or before 29 June 2026.
• The lease was conditional and stayed conditional. In the landlord's own words it "remains subject to the satisfaction of certain conditions precedent beyond Bitdeer's control, including the completion of certain external customer and supplier arrangements by the counterparty." The tenant had not yet landed its customer or its chips when it signed.
• 16 year base term plus an 8 year option. About US$202 per kW per month, with electricity passed through. About US$4.7bn of contracted rent over the base term.
• Instead of a balance sheet it put up about US$1.3bn of standby letters of credit arranged through bank affiliates, roughly 4.4 years of rent. No parent guarantee.
• The landlord funded the conversion, about US$500mm of remaining capital expenditure, roughly US$4.0mm per MW IT. The landlord may terminate if the tenant misses its credit milestones. The tenant may walk at year 10 with no fee.
• Everything it did, it did on other people's assets and other people's credit, inside eight months.
• That is as much a statement about the market as about Volta. Banks wrote US$1.3bn of credit and a landlord committed US$4.7bn of capacity to a company incorporated eight months earlier with one pound of share capital and no revenue. The market is paying for speed and structure right now. With the right tools in hand, the number at the end of this is very large.

Why we combine them

• DayOne is the proof that a PRC-listed operator can hold a large non-controlling stake in a Singapore platform, and that the platform raises Western capital at scale and gets NVIDIA equipment.
• Volta is the proof that the enterprise itself can be built from nothing at speed, on other people's assets and other people's credit.
• DayOne is a landlord that does not own the compute. Volta is a cloud that owns no buildings. Each captures one margin.
• We do both sides and own both. On the developer side we hold the building and the contracted rent. On the cloud side we take the capacity conditionally and finance the chips.
• Because both sides are ours, the credit support is negotiated with our own project lender and infra equity rather than with a stranger, and we keep the margin Volta paid away.

Appendix 2. The full structure

Every entity, with Site 1 as the worked example. Each country repeats the site layer.

Every entity, with Site 1 as the worked example. Each country repeats the site layer. Reference material, so it is wider than the page and scrolls sideways.