How to read an infrastructure IPO whose income statement hasn't caught up with its contracts: time-bucket the backlog and put its capital bill beside it, strip non-cash charges without forgetting what they cost, underwrite the yield-vs-financing spread, and map the circular relationships between tenant, supplier and sponsor. Not whether to buy, but which numbers carry the case.
1. Time-bucket the backlog and put the CapEx bill next to it
The repeatable method
- Find the remaining-performance-obligation table in the filing: split contracted revenue into near-term (e.g. years 1–8) and long-dated buckets, and note the weighted-average remaining contract term.
- Find the disclosed future capital expenditure needed to deliver that backlog; divide backlog by CapEx and compare against the timing — long-dated rent against front-loaded spend is a financing problem, not a demand proof.
- Compare with how backlog behaves at a software company or chipmaker (converts within a few years at high incremental margin) before letting the headline number set a multiple.
Here: SB Energy's
$439B backlog is
$82B in years 1–8 and $357B (81%) after year 8, ~20-year weighted term, against
~$178B of CapEx ($174B data centers) — on $139M of H1 revenue. "SB Energy has contracted decades of future rent, not hundreds of billions of revenue that will arrive anytime soon."
read ↗
Watch for
- First rent commencement (Cosmos phase 1, expected Q4 FY26) and the near-term bucket growing; any CapEx estimate revisions per contracted GW.
2. Strip the non-cash loss — then price what the non-cash items actually cost
The repeatable method
- Decompose the GAAP net loss into cash operations vs non-cash items (warrant/derivative revaluation, stock-based comp); check cash from operations as the reality test.
- Identify who holds the revalued instrument — warrants issued to a customer are a customer-acquisition cost paid in equity.
- Keep SBC in the dilution math and the warrant value in the cost of the key contract, rather than adding them back and forgetting them.
Here: the
$3.2B H1 loss is
$2.573B of OpenAI-warrant revaluation +
$590M SBC, with only ~$56M of operating cash used — yet "OpenAI's warrants represent a very real economic cost of securing its most important customer."
read ↗
Watch for
- Warrant liability swinging with the IPO price; fully diluted share count including OpenAI's warrants and NVIDIA's discounted forward shares.
3. Underwrite a landlord on the spread, not the margin
The repeatable method
- Find the lease structure (triple-net? opex pass-through? escalators? term) and the pricing basis (yield on cost).
- Compare the unlevered yield on cost with the marginal cost of debt; the spread, times leverage, is the equity return.
- Stress it: raise construction cost ~15% or borrowing cost 100 bps and see how much of the spread survives; note any contractual penalties for late delivery.
- Don't confuse high property-level margins with free cash flow — depreciation and interest sit between them.
Here: data-center developers target
~8–11% unlevered yield on cost; borrowing at
6–7% makes equity returns "look great. But that spread is unforgiving" — a 15% cost jump or worse loan terms "shrinks fast," and some leases carry abatement or termination rights if capacity is late.
read ↗
Watch for
- Disclosed project-financing rates and yields once campuses close financing; cost overruns or equity injections on PORTS-Pike / Milam.
4. Map the closed loop — who is tenant, supplier, lender and shareholder at once
The repeatable method
- List every counterparty with more than one role (customer + warrant holder, supplier + investor + credit guarantor, sponsor + customer + controller).
- For each, ask which way its incentives point if the cycle slows — and whether a single party's retrenchment hits revenue, financing and governance simultaneously.
- Separate fresh capital from secondary exposure in strategic investments (a prepaid forward with the parent is not money into the company).
- Check control: a sponsor above 50% of votes means controlled-company exemptions and limited minority influence.
Here: OpenAI = anchor tenant + warrant holder + board designee;
NVDA = compute supplier +
$1.5B Class N buyer +
$1.5B forward with parent Energy Global (only half fresh capital) + credit support on OpenAI's leases;
SFTBY = founder + major customer +
>50% of votes. "SB Energy sits at the center of a circular AI infrastructure ecosystem."
read ↗
Watch for
- New customers outside the OpenAI/SoftBank loop (other hyperscalers, sovereigns, enterprises); any change in frontier-lab spending plans.
The repeatable method
- For any gigawatt-scale project, identify the input with the longest lead time (generation, turbines, transmission, interconnection) from the risk factors.
- Compare that lead time with contracted delivery dates; a finished building without power earns nothing.
- Credit developers who own the scarce input (power development in-house) — the same factor is both moat and bottleneck.
Here: PORTS-Pike needs
9.2 GW of new gas generation and the filing warns
gas-turbine lead times can stretch to seven years, against first phases in 2028 — while SB Energy's renewable-development heritage is its edge in securing power.
read ↗
Watch for
- Turbine orders and interconnection approvals for Pike; phase timing slipping past 2028.
6. IPO rule: "It's Probably Overpriced" — wait for proof of the unit economics
The repeatable method
- When current revenue can't anchor a multiple, name the single question the valuation depends on (here: how much project economics accrue to equity).
- Write the bull and bear cases as conditions that will be observable (first operating assets, realized yields, financing terms).
- Defer until the first of those observations arrives rather than paying for the backlog at the IPO price.
Here: at a reported
>$50B, "investors are not buying the backlog itself. They are buying the returns SB Energy can earn on the capital required to serve it" — the author wants "the first major campuses operating and the project-level economics proven."
read ↗
Watch for
- The S-1 price range; post-IPO trading vs the offer; first operating-campus disclosures.
Methods distilled from the Premium App Economy Insights newsletter (article text in transcript.txt) for personal study. Not investment advice. © App Economy Insights for source material.