OpenAI Weighs $1.2T Pre-IPO Round, SpaceX Signs $1.1B/Month AI Compute Deal, FT Exposes $300B Off-Balance-Sheet AI Debt

Four stories, one week, and the AI capital cycle is accelerating faster than anyone modeled.

1. OpenAI Weighs a $1.2T Pre-IPO Round

Per Reuters (Sept 15), citing the Financial Times, OpenAI is in early-stage, investor-initiated discussions about a new funding round at a valuation near $1.2 trillion — roughly 41% above the $852B post-money valuation set in March 2026.

The round would be a pre-IPO move: Sam Altman has stated OpenAI will not go public in 2026, citing safety considerations. An IPO is expected in 2027. At $1.2T, this would leapfrog Anthropic ($965B, May 2026) and make OpenAI the most valuable private company in history.

Why it matters: A $1.2T pre-IPO round is not a capital raise in the traditional sense. It is a valuation event that re-prices the entire AI sector. Every downstream model, datacenter, and chip vendor gets repriced against this anchor. The 2027 IPO, when it comes, will be the largest in history — larger than SpaceX's $85.7B raise in June 2026.

2. SpaceX Signs $1.1B/Month AI Computing Deal

At a Goldman Sachs conference, SpaceX CFO Bret Johnsen disclosed a new AI computing contract: $1.1 billion per month, or $13B annualized. Payments begin in December. The target is a $100B annualized run-rate by end of 2026.

This is not a datacenter lease. This is SpaceX selling compute capacity — the same way a hyperscaler sells cloud. SpaceX is now both a buyer of AI infrastructure (for Starlink's AI workloads) and a seller of it. The $13B/year contract is roughly 15% of SpaceX's estimated $80B+ annual revenue at current pace.

Why it matters: A $13B/year AI compute contract from a company that just completed the largest IPO in history ($85.7B, June 2026) signals that AI infrastructure demand is not just from hyperscalers. It is now from rocket companies. The compute economy is fragmenting across more buyers, which means more capex, more datacenter buildouts, and more power demand.

3. FT: Big Tech Hides ~$300B AI Infrastructure Debt Off Balance Sheets

An FT investigation (Sept 20) reveals that Big Tech is using residual value guarantees to move roughly $300 billion in AI infrastructure debt off their balance sheets.

The mechanism: a hyperscaler signs a datacenter build, then uses a special purpose vehicle (SPV) or financial intermediary to hold the debt. The hyperscaler provides a residual value guarantee — a promise to buy back the asset at a set price at the end of the lease term. The debt sits on the SPV's books, not the hyperscaler's.

Key findings:

• Alphabet guarantees $16.9B that balloons to $43.8B in six months — less than 2% on its balance sheet
• Meta uses a Delaware SPV for its $50B Louisiana Hyperion datacenter (Blue Owl 80% / Meta 20%), with ~$28B in guarantees backing $27B in Pimco/BlackRock/Apollo debt
• Broadcom, Nvidia, Oracle, Amazon all use variants of this structure

Why it matters: The AI capex cycle is being financed through structures that look like operating leases on the surface but are essentially long-term debt obligations. When you normalize for residual value guarantees, the true leverage of these companies is significantly higher than reported. This is the same playbook that fueled the 2007 commercial real estate cycle — and it means the AI infrastructure bubble, if it comes, will pop through the SPV channel, not the hyperscaler balance sheets.

4. Huawei: AI Chip Demand Outstrips Supply

At a conference on Sept 17, Huawei chairman Eric Xu said the company "cannot produce enough AI computing equipment to meet demand in China." Huawei is actively limiting overseas sales to prioritize domestic demand.

Two new AI chips are confirmed:

• Ascend 960DT — Q1 2027 (three quarters ahead of original schedule)
• Ascend 960PR — Q3 2027 (one quarter ahead)

Why it matters: When a chip maker says demand exceeds supply, the implication is that the AI infrastructure buildout is not slowing. It is accelerating to the point where supply cannot keep up. The fact that Huawei is pulling chips from overseas markets to protect domestic supply is a direct signal that Chinese AI demand is not a tailwind — it is a constraint on global AI infrastructure availability.

The Position

These four stories are not four separate news items. They are four data points on the same curve:

1. Valuation — OpenAI at $1.2T pre-IPO sets the ceiling for AI asset pricing
2. Demand — SpaceX at $13B/year in AI compute shows the buyer base is widening
3. Financing — $300B in off-balance-sheet debt shows the capital is leveraged, not organic
4. Supply — Huawei saying demand exceeds supply shows the physical constraint is real

The AI capital cycle is in its infrastructure buildout phase. The question is not whether it will peak — it will — but whether the off-balance-sheet financing structures will hold under stress. When they don't, the correction will look less like a tech stock selloff and more like a 2007 commercial real estate unwind: slow, leveraged, and distributed across SPVs that no one on the balance sheet will show you.

Position before you predict.