BusinessIssue #198

The Contract Google Can Exit in 90 Days

SpaceX's earnings were flawless—except for the fine print in its contracts.

Oswarld (Kwangseob Ahn)· Writer & editor of OZ Talking

The Contract Google Can Exit in 90 Days

Opening

Reader, on August 4, SpaceX released its first earnings report since going public.

Q2 revenue came in at $7.81 billion, up 92% year-over-year. Loss per share was $0.09, far better than the projected $0.26 loss, and adjusted EBITDA1 came in at $3.54 billion. The CFO declared a “$100 billion annualized run rate by December.” By every measure, it was a flawless quarter.

And the stock started falling that very day. On August 5 alone, shares dropped 13.6% to close at $108.27. That marked a fifth straight week of declines—20% below the $135 IPO price set in June, and roughly half its peak of $225.64.

Normally, a scene like this gets chalked up to “lock-up expiration selling” and left at that. Today, August 6, happens to be the day 911.5 million shares unlock, so that explanation isn’t wrong, either. But I think there’s a deeper issue here—one that will outlast this news cycle. Let me cut to the chase: what the market read this time wasn’t the earnings report—it was the contract. Buried in SpaceX’s AI compute leasing agreements—the engine driving its growth—is a clause that lets customers walk away with just 90 days’ notice.

The Rocket Company’s Revenue No Longer Comes From Rockets

Let me first lay out what kind of company this actually is right now. Since SpaceX absorbed xAI last February in a merger (combined enterprise value: $1.25 trillion), the rocket company we used to know and Grok/X have become one entity. So the segment breakdown on the earnings report now looks like this.

SegmentQ2 RevenueYoYOperating Income/Loss
Space (Launch)$962 million+29%-$542 million
Connectivity (Starlink)$4.29 billion+66%+$1.66 billion
AI (Grok/X ads, compute rental)$2.56 billion+247%-$1.26 billion

Rockets make up 12% of this company’s revenue. Starlink is the only thing making money, and the AI segment is the one whose growth rate is exploding. The reason the market has priced this company north of $1 trillion, and the reason it’s marking that price down today, both live in that third row.

Let me dig a little deeper into the AI segment’s numbers. Adjusted EBITDA is a positive $1.146 billion. That’s a swing to profitability in a single quarter, up from a $609 million loss the prior quarter (Q1). But operating income for the same segment is negative $1.257 billion. The roughly $2.4 billion gap between these two figures is mostly depreciation2.

Read these two lines together and here’s what they mean: the company is generating cash, but the assets it bought to generate that cash are melting away even faster.

We Spent $6 to Earn $1 in Revenue

Capex3 is what drove the burn rate.

SpaceX’s Q2 capital expenditure came to $18.37 billion. Of that, $15.8 billion went to AI infrastructure. Q1 AI capex was $7.7 billion, so it doubled in a single quarter. That works out to $170 million a day — roughly 240 billion won a day — poured into GPUs and data centers.

That same quarter, AI-segment revenue was $2.56 billion. In other words, the company spent $6.2 to generate $1 of revenue. Musk said he wants to push power-and-cooling capacity to 15 gigawatts (with a stretch target of 20 gigawatts) by the end of 2027. The company hasn’t issued separate capex guidance for 2027, but Piper Sandler estimates it at roughly $65 billion — $17 billion above prior market expectations.

For investors to accept these numbers, they only need to believe one thing: that this equipment will keep making money steadily for years to come. Management explains it the same way, arguing that AI compute investments have a payback period4 of under 1 year, making them closer to cost of revenue than capital expenditure.

This is where we need to open up the contracts.

The word Google itself used was “temporary bridge”

SpaceX signed $14.1 billion in new cloud contracts in Q2. 2 customers make up the core of that.

Anthropic will pay $1.25 billion a month through May 2029, in exchange for use of the entire Colossus 1 data center in Memphis. Google will pay $920 million a month, covering roughly 110,000 Nvidia GPUs, from October 2026 through June 2029 — that is ₩1.27 trillion a month in our currency. Combine the 2 deals and you get $26 billion a year in recurring revenue from compute leasing alone.

That is the headline. But look at the terms of the Google contract, and you find this clause.

After December 31, 2026, either party may terminate the agreement with 90 days’ notice. If SpaceX fails to deliver the committed GPUs by September 30 (after a 1-month grace period), Google may terminate immediately.

Let’s do the math. The contract starts in October 2026. Termination notice can be given only after December 31, and once given, it takes 90 days to take effect. So the period that is actually locked in runs from October 2026 to around March 2027 — about 6 months. In dollar terms, that is roughly $5.5 billion.

On paper, this is a 33-month, $30.4 billion contract. What is actually guaranteed is $5.5 billion. About 82% of the total is a number that could disappear at any time.

And when Google announced the deal, it described it, in its own words, as “short-term, timely bridge capacity” secured because demand for Gemini Enterprise grew faster than expected. Bridge — meaning a temporary bridge. Alphabet is pouring over $180 billion into its own infrastructure this year alone. A bridge is only needed until you have crossed the river.

This is today’s core asymmetry.

Costs cannot be cancelled. The $15.8 billion worth of GPUs have already been bought, the data centers have already been built, and depreciation will chew into the income statement for years, regardless of the contract.

Revenue can be cancelled. 90 days, and it is over.

Management’s claim of “under 1 year payback” is a calculation that only holds if the contract keeps running. If that premise wobbles, it is not just the conclusion that collapses — the equation itself falls apart.

One more thing worth flagging. The company disclosed total backlog of $47.5 billion. But just the nominal total of the 2 contracts above does not square neatly with that figure. It is possible the cancellable portions were not fully reflected in the backlog — though this is my own inference, so I would encourage you to check the backlog methodology directly in the original filing. Either way, the implication is the same. A “$100 billion annual run rate” and a “$47.5 billion backlog” are commitments of very different strength.

The corrected fragment is already accurate. Here it is unchanged:

It’s Not Just SpaceX’s Problem

This structure is now shared across the entire AI infrastructure industry.

The clearest comparison is Oracle. Oracle’s remaining performance obligations5 have swelled to $638 billion. A staggering number. But analysts estimate more than half of that comes from a single customer, OpenAI (this isn’t something Oracle disclosed—it’s an estimate based on the roughly $300 billion contract with OpenAI). In the same fiscal year, Oracle spent about $56 billion on equipment while posting a free cash flow deficit of $23.7 billion, and it’s now planning to raise about $40 billion to bridge that gap. Oracle even added warning language to its SEC filings stating that building AI infrastructure could pressure profitability.

spcxIt’s the exact same pattern. A backlog is a contract, and a contract isn’t cash. But the money being spent right now, against the collateral of that contract, is very real cash.

What makes termination clauses especially dangerous in this game is that most AI compute customers are companies busy building their own infrastructure. Google, Microsoft, Meta—they all use leased capacity as a bridge until their own facilities are finished. A large share of today’s compute-leasing demand isn’t structural demand at all—it’s timing demand. It only exists because supply is bottlenecked, and it disappears the moment the bottleneck clears.

Of course, there’s an opposing view too. Morgan Stanley’s Adam Jonas maintains an overweight rating with a $300 price target, arguing that the current stock price “effectively assigns almost zero value to the AI business.” Bank of America and JPMorgan have also put out targets of $235–$240. On the other side, Piper Sandler cut its target from $156 to $140 while keeping a neutral rating, and some investors call $30 the fair value. The fact that price targets range tenfold—from $30 to $300—is itself a signal that no one right now is confident about the durability of this company’s revenue.

Oswald’s Lens

When I’m building out a GTM strategy, I end up reviewing a lot of client pipelines, and there’s one thing I check before I even look at the contract value: the termination clauses and the minimum commitment period.

The total contract value that sales brings in and the number finance can actually put into the forecast are almost never the same. And that gap tends to be widest precisely when a company is doing well. When demand surges, customers sign flexible terms just to lock in a spot, and sellers accept those terms because they’re under pressure to show a headline number. Both sides are being rational in the moment. The problem is that the seller absorbs the entire cost of that flexibility. The customer bought an option; the supplier bought equipment.

The number that weighed on me most in this earnings report wasn’t the 92% revenue figure or the $18.3 billion in capex. It was seeing the AI segment’s adjusted EBITDA of +$1.1 billion sitting right next to an operating loss of -$1.26 billion. That doesn’t mean “still unprofitable” — it means this business’s P&L is being held hostage by a depreciation schedule. Depreciation isn’t negotiable. Revenue is.

So I see today’s lock-up expiration as a catalyst for the decline rather than its root cause. Even if the float ratio jumps from 4.9% to 11.8% and expands to roughly half by summer 2027, supply and demand is an event that eventually ends. By contrast, a structure where “cancellable revenue is funding non-cancellable assets” isn’t an event — it’s a business model.

That said, I don’t want to look at this from only one angle. Starlink is a genuine cash generator, with 12 million subscribers and $1.66 billion in quarterly operating income, and roughly $100 billion in liquidity gives this experiment the stamina to keep running for years. My verdict isn’t “this company is at risk” — it’s that what the stock price should be pricing in right now isn’t the growth rate, but the quality of the contracts.

Closing

To sum up, three things.

First, SpaceX’s Q2 results themselves were excellent. 92% revenue growth, beating consensus. The problem isn’t the earnings sheet — it’s the contract behind that revenue.

Second, the contract with Google, its key customer, can be terminated with 90 days’ notice any time after the end of 2026, and Google itself described this as a “temporary bridge.” Of the nominal $30.4 billion, the confirmed portion is only about $5.5 billion — six months’ worth.

Third, this asymmetry — where costs are non-cancellable but revenue is cancellable — isn’t unique to SpaceX. It’s a structural feature of the entire AI infrastructure industry right now. Oracle’s $638 billion in remaining performance obligations faces the same question.

Going forward, there’s really one thing to watch: whether Google issues a termination notice after December 31, 2026. Whatever disclosure comes out around that date will be the real first stress test of this company’s valuation. If you check the backlog accounting standards and AI-segment depreciation expenses separately in the Q3 results, you’ll be able to read far more than the headline numbers reveal.

For the record, this piece is not investment advice, nor is it grounds for any investment decision on a specific stock. All figures cited are based on public reporting and company disclosures, but if you’re interested, I’d recommend checking the original filings yourself.

Lastly, I want to ask Reader something. Have you ever had a forecast completely upended because of a renewal or termination clause — whether in SaaS, cloud, or a services contract? Tell me in the comments which clause was the problem and how you rewrote the contract afterward. I’ll gather these stories and write a follow-up issue on “how to measure the quality of a contract.”

💬 Share in the comments if a single contract clause once blew up your revenue forecast. I’ll factor it into a follow-up issue. 📨 If you have a colleague working in finance or sales pipelines, please pass this along to them.

References & Further Reading

Primary sources

Background

Related issues worth reading alongside this one

📝 Glossary

Kwangseob Ahn profile illustration

The author, Kwangseob Ahn, is a professor of business administration at Sejong University and lead consultant at OBF (Oswarld Boutique Consulting Firm). He teaches statistics and data analysis, including business data management and business analytics, while leading GTM and AI strategy consulting in the field, designing the seam between technology and business. He has published academic research on a memory architecture for AI dialogue systems (HEMA) and runs Daily Arxiv, a daily curation of global AI papers. He holds a master's from Korea University's Graduate School of Technology Management and a KMBA. He is the author of Homo Brainless: The People Who Outsource Their Thinking.

Footnotes

  1. Adjusted EBITDA: Earnings before interest, taxes, and depreciation, with one-time items also stripped out. Companies that buy a lot of equipment tend to look far better on this metric than on actual operating income, so you need to view the two side by side to see the real picture. ↩

  2. Depreciation: The accounting treatment that spreads the cost of a big up-front asset purchase across its useful life instead of expensing it all at once. If a GPU is assumed to last 6 years, one-sixth of its purchase price gets shaved off the income statement each year — and that expense keeps going even if the revenue stops. ↩

  3. CapEx (Capital Expenditure): Money spent on assets meant to be used over several years, like data centers, equipment, or GPUs. Because it isn’t expensed in full in the year it’s spent — it flows through depreciation instead — you won’t see it on the income statement alone; you have to look at the cash flow statement. ↩

  4. Payback period: The time it takes to recoup an investment through the cash it generates. A “sub-one-year” payback means the cost of a single GPU is recovered in lease payments within a year — but that assumes the lease contract stays in force for that entire period. ↩

  5. RPO (Remaining Performance Obligations): Contracted amounts that haven’t yet been recognized as revenue because the service hasn’t been delivered yet. Commonly called the “backlog.” It’s a preview of future revenue, but it isn’t cash — and if the contract has a termination clause, the preview may never play out as promised. ↩

Your take shapes the next issue

What resonated most in this issue, or where has your experience been different?

Any registered reader can comment for free.

Related issues

Issue #196Business

Why Midjourney Bought an Astrology App

Line up everything Midjourney acquired this year, and a pattern emerges.

Issue #194Business

What OpenAI's Astra Threw Away to Solve 10 Hard Problems

OpenAI's Astra finished a valid proof — then discarded it for a simpler one.

Issue #193Business

The Goal Wasn't Victory. It Was Division.

The same accounts attacked both Moon Jae-in and Yoon Suk Yeol.