Executive summary

Compute demand: Jevons paradox playing out in real time

ROI and agentic multipliers have driven parabolic token growth

Supply unable to keep up, data center power a key bottleneck

US AI data center power gap: Up to 80 GW by 2030

In the US, public opinion compounds the shortage

Popular opinion is creating political headwinds to US capacity buildout

Structural supply & demand mismatch creating historic opportunity

Capture short-term rates and GPUs can pay back in 12-18 months

ROI on power is massive for AI workloads

The same megawatt earns far more in AI compute than in industry

Financial markets maturing unlocks turbocharged scaling

In three years, AI compute went from corporate capex to an institutional infrastructure asset class

Rapid scaling is already happening

Customer and capital move fast, the key gate is powered land

Existing non-US power assets becomes the next frontier

Special Economic Zones (SEZs)

(as proxy for industrial park density)

Industrial parks in developing markets are especially relevant

Our approach: Systematic sourcing as an origination platform

Origination Partners secure the powered landshells our token factories need

Platform incentives sets us up for long term success

Making partnerships attractive for asset holders drives speed and volume

Partner motivation enables us to “blitzscale” power

Platform enables power asset holders to capture compute premia

We combine sites with GPUs to lease compute wholesale

We rent the shell, finance the GPUs and sell firm wholesale compute

At signing the buyer prepays 15% of firm contract value and GPU debt covers 60% of GPU cost, together 92% of the first site's GPU capex including interest, reserve and fees. One month after each draw's customer acceptance the pre-negotiated GPU facility steps up to 70% of GPU value, capped at 1.25x debt service cover, and returns SHP AI's equity. From Site 2 the facility is 65% stepping up to 80%, so buyers and lenders fund 95% at signing. Conservative against the market: Nscale's 2026 GPU loan steps each cluster from 50% to 55% of capex up to 75% to 80% once revenue starts and CoreWeave's DDTL 4.0 lends 90% of cost during installation, so our 70% after acceptance is below both; our prepayment covers 36% of GPU cost against 50% to 60% of capex on 70% of Nebius deals.

GPU financing: Buyers and lenders help fund the chips

Our financial structuring maximizes capital efficiency

Month 0 is power secured. Contract signing and GPU financial close at month 3, GPUs paid over six months, customer acceptance and billing from month 10. The step-up is pre-negotiated with the GPU lender and executes one month after acceptance: New 60-month amortizing debt at 11%, sized at the lower of 70% of current GPU value and forward cash flow at 1.25x DSCR. It repays the initial debt, pays a 2% fee and funds one month of debt service reserve, and the rest returns to SHP AI. The prepayment is credited against invoices from service month 25. Each 10 MW draw follows the same clock, two months apart. From Site 2 the facility is 65% stepping up to 80%: Buyers and lenders fund 95% at signing and the step-up returns 3.0x SHP AI's equity. Benchmarks: Nscale's 2026 GPU loan steps each cluster from 50% to 55% of capex up to 75% to 80% once revenue starts; our prepayment covers 36% of GPU cost against 50% to 60% of capex on 70% of Nebius deals. Activation reserve lowers peak funding by US$64mm.

And create unit economics that is attractive for all parties

We earn 36x while creating tremendous value for our Partners

We already have 300+ MW pipeline and five Anchor Partners

Proven operators and partners enable us to deliver rapid energization at scale

Repeat founders combining financial structuring with full cycle data center experience

Financials: Rapid scaling to blue chip-like EBITDA

Compute revenue reaches US$33bn in 2034 on 2.3 GW of billing capacity

Appendix

Base case by year

Compute revenue reaches US$33bn in 2034 on 2.3 GW of billing capacity

Financials by segment

Compute drives HoldCo financials, with sites and development a small cash contributor

Cash flow by segment

External funding peaks at US$95.5mm because compute cash funds the buildout from 2029

Base case peak capital need is US$95.5mm

Capital need peaks as the first contracts fund GPUs, then billing turns cash positive

Financial structuring: US$95mm funds US$73bn in five years

Partners, buyers and lenders fund the build, and recycled cash funds our equity

Base case, 2028-2032. GPU capex US$52.1bn by year of spend (about 1,570 MW IT deployed) plus US$3.9bn of interest, reserve and fees. Site capex US$17.3bn for 3 GW grid (2,040 MW IT) by construction start, at US$5.25mm per MW grid plus 10% contingency, 70% construction debt. Gross equity is US$3.0bn into GPU draws and US$0.5bn of SHP AI site equity. Recycled cash is debt step-ups, operating cash and site cash, net of HoldCo costs. Net peak is the HoldCo peak funding need in Sep 2028, including a US$10mm cash buffer.

Neocloud economics

Firm wholesale contracts earn a 72% EBITDA margin after paying the site its rent

GPU financing: Sources & uses

Customer prepayments and GPU debt fund 92% of the first tranche

GPU financing: How a debt step-up works

Our GPU equity comes back at acceptance and recycles into the next draw

Month 0 is power secured. Contract signing and GPU financial close at month 3, GPUs paid over six months, customer acceptance and billing from month 10. The step-up is pre-negotiated with the GPU lender and executes one month after acceptance: New 60-month amortizing debt at 11%, sized at the lower of 70% of current GPU value and forward cash flow at 1.25x DSCR. It repays the initial debt, pays a 2% fee and funds one month of debt service reserve, and the rest returns to SHP AI. The prepayment is credited against invoices from service month 25. Each 10 MW draw follows the same clock, two months apart. From Site 2 the facility is 65% stepping up to 80%: Buyers and lenders fund 95% at signing and the step-up returns 3.0x SHP AI's equity. Benchmarks: Nscale's 2026 GPU loan steps each cluster from 50% to 55% of capex up to 75% to 80% once revenue starts; our prepayment covers 36% of GPU cost against 50% to 60% of capex on 70% of Nebius deals. Activation reserve lowers peak funding by US$64mm.

GPU financing: Monthly flows & peak

Prepayment, GPU debt, step-ups and activation reserve enable US$55mm of equity to support US$2bn of GPUs

Month 0 is power secured; contracts and GPU financing close at month 3. The site deploys in seven draws, six of 10 MW and one of 8 MW, each accepted two months after the last (months 10 to 22). Each draw is paid over the seven months before acceptance: Customer prepayment of 15% of firm contract value and GPU debt at 60% of GPU cost go straight to the supplier, and SHP AI funds the 4% gap plus interest, reserve and fees. On acceptance the pre-negotiated facility steps up to 70% of GPU value (60-month amortizing, 11%, 1.25x DSCR), repays the initial debt and returns more than that draw's equity, which funds the next draw. Cash positive from month 16. Off balance sheet GPU vehicle not assumed. Benchmarks: Nscale's 2026 GPU loan steps each cluster from 50% to 55% of capex up to 75% to 80% once revenue starts; our prepayment covers 36% of GPU cost against 50% to 60% of capex on 70% of Nebius deals. Activation reserve lowers peak funding by US$64mm.

Serving open weights: Up to US$100mm per MW a year

Token serving is upside reached through M&A, while the base case stays on contracted leases

Origination Partner cash back from one site

The development capital is repaid at month 3 and every later dollar is profit