Valuing the Unprofitable Giants: OpenAI, SpaceX and Anthropic Through the Lens of Excess Earnings

Introduction

June 2026 will be remembered as the moment three of the most anticipated IPOs in history converged. OpenAI has filed for a Q4 listing targeting a $1 trillion valuation. SpaceX confirmed its mid-2026 float after 24 years as a private company, with recent tender offers valuing it at $1.75 trillion. Anthropic filed confidential IPO papers on June 1 and is now valued at over $1 trillion-surpassing OpenAI on some measures.

None of these companies is profitable in the traditional sense. Not one.

OpenAI reported $12 billion in annualised revenue for July 2025 but spent billions more on compute. SpaceX's Starship development alone has consumed capital that would bankrupt a public company. Anthropic went from a $61.5 billion valuation to over $1 trillion in just over a year while burning through cash for model training and data centre capacity.

This poses a fundamental challenge for business valuation. The capitalised excess earnings method-a staple of Australian SME and professional practice valuation-simply breaks down when applied to companies where current capex exceeds current revenue. How do you value a business where the thesis is that today's massive spending creates tomorrow's infrastructure monopoly?

This article explores the excess earnings and revenue multiple methods through the lens of OpenAI, SpaceX, and Anthropic, and asks what their valuations mean for how we think about value in the age of AI.

Part 1: The Three Giants

OpenAI

• IPO target: Q4 2026, targeting $1 trillion valuation
• Latest funding: $40 billion in April 2025 (SoftBank-led) at $300 billion post-money
• Revenue: $3.7 billion (2024) to $12 billion annualised (July 2025)
• Revenue growth: 3.2x in one year. 20 million ChatGPT paid subscribers, 5 million business users
• Capex: The Stargate Project-a $500 billion joint venture with Oracle, SoftBank and MGX for AI data centre infrastructure. Plus $250 billion in committed Azure services from Microsoft
• Structure: Converted to a Public Benefit Corporation (PBC) in 2025. Nonprofit retains 26% ownership. Microsoft holds 27%

SpaceX

• IPO: Confirmed by Elon Musk in December 2025. Expected mid-2026 (H1). Four investment banks selected in January 2026. Widely described as the largest IPO in history
• Valuation: ~$1.75 trillion in recent tender offers. Trajectory: $1.3 billion (2012) to $12 billion (2015) to $46 billion (2020) to $74 billion (2021) to $350 billion to $1.75 trillion
• Revenue: Not publicly disclosed. Starlink is the primary revenue driver, with an estimated $8-10 billion in 2025 revenue. The launch business holds 45% of the global commercial launch market
• Capex: Starship development (the largest rocket ever built), Starlink constellation expansion, and SpaceXAI ($2 billion invested July 2025)
• Key insight: SpaceX has been private for 24 years. Musk repeatedly rejected going public because he didn't want quarterly earnings pressure on long-term capital projects

Anthropic (Claude)

• IPO: Confidential papers filed June 1, 2026. Target debut fall 2026
• Valuation: $61.5 billion (March 2025 Series E) to $183 billion (September 2025) to $380 billion (February 2026 Series G) to over $1 trillion by June 2026, surpassing OpenAI on some measures
• Revenue: Not publicly disclosed. Industry estimates suggest a $1-2 billion run rate
• Funding: Amazon ($8 billion-plus total commitment), Google ($2 billion-plus), Lightspeed Venture Partners
• Capex: $1.8 billion cloud deal with Akamai. Access deal with xAI's Colossus 1 data centre. Google TPU access (up to one million chips). Data centre expansion commitments exceeding $5 billion

Part 2: Why Traditional Methods Break Down

The Capitalised Excess Earnings Method

This method, commonly used for Australian SME valuations, works as follows:

1. Calculate normalised maintainable earnings (typically adjusted EBITDA)
2. Apply a capitalisation rate (the inverse of a PE multiple) based on risk
3. Add the net asset backing

For a construction business with $700K normalised EBITDA and a 25% capitalisation rate, the calculation is straightforward: $700K / 25% = $2.8 million.

The problem when applied to AI giants:

If you attempt a DCF on OpenAI using $12 billion in revenue, a 20% operating margin assumption, and a 10% discount rate, approximately 75% of the implied value comes from the terminal value beyond year 10. The calculation becomes a bet on what happens after 2036, not a valuation of the business as it stands.

The Revenue Multiple Method

This is the method the market is actually using. It's simple, transparent, and entirely conviction-based.

To put these in context:

These multiples are not justified by current earnings. They are justified by a specific thesis: that massive current capital expenditure creates future infrastructure monopolies with returns that justify today's spending.

Part 3: The Infrastructure Monopoly Thesis

This is the intellectual foundation for these valuations. It deserves careful examination.

The AWS Analogy

Amazon Web Services was launched in 2006 after years of internal infrastructure investment. For the first seven years, AWS was a cost centre. Amazon was spending billions on data centres with uncertain returns. By 2013, AWS was generating $3.1 billion in revenue. By 2024, it was generating over $100 billion with margins approaching 30%.

The AWS investment thesis was simple: build infrastructure at a scale competitors cannot match, then sell access to it at a premium. The early years of negative returns were the price of building the moat.

OpenAI and Anthropic are making the same argument. The Stargate Project's $500 billion price tag creates AI compute infrastructure at a scale no single competitor can replicate. If AI models become a commodity (like cloud compute), the companies that own the infrastructure earn infrastructure margins.

The Starlink Monopoly

SpaceX's Starlink already has over 5 million subscribers globally and is the dominant low-earth orbit satellite internet provider. There is no meaningful competition for global satellite broadband. OneWeb is a distant second.

Starlink's capex is front-loaded. Once the constellation is fully deployed, the marginal cost of adding a subscriber is low. If Starlink achieves the margins SpaceX projects, the satellite internet business alone could justify a significant portion of SpaceX's $1.75 trillion valuation-before considering Starship, point-to-point transport, or Mars.

The AGI Option

This is the most speculative element and the hardest to value. If OpenAI or Anthropic achieves artificial general intelligence, the value creation could be an order of magnitude beyond anything in the current valuation.

An AGI option is not something a traditional capitalised excess earnings model can capture. It's a binary bet: either AGI happens and these companies become the most valuable in history, or it doesn't and they revert to being highly competitive AI model providers with thin margins.

Part 4: What This Means for SME Valuation

You may be reading this and thinking the comparison is absurd. What does OpenAI's $1 trillion IPO have to do with valuing a construction business in Sydney or a medical practice in Melbourne?

More than you might think.

Lesson 1: Capex is a bet, not a cost

When you invest in a new piece of equipment, a new software system, or a new hire, you're making the same decision OpenAI is making with its Stargate Project. You're spending today for a return tomorrow. The difference is scale, not logic.

A well-prepared valuation should identify and separate growth capex from maintenance capex. If your business has invested heavily in automation, a new facility, or a new product line, that investment should be reflected in your valuation-not as a current cost that reduces EBITDA, but as a future earnings driver that a buyer is acquiring.

Lesson 2: Revenue growth justifies a multiple premium

The reason OpenAI trades at 83x revenue is because its revenue tripled in a year. If your business is growing at 20-30% year-on-year, that is a valuation driver, not a footnote. Most standard SME valuations use a single multiple based on industry averages. They miss the premium that growth deserves.

When preparing your business for sale, document your growth rate, your growth drivers, and the investments you've made to sustain them. A buyer who sees a 20% CAGR with a clear driver (new product line, new geography, new channel) will pay a higher multiple than one who sees a flat earnings stream.

Lesson 3: The terminal value always matters more than you think

Even in a standard SME DCF, 40-50% of the value often comes from the terminal value-the period beyond the explicit forecast. The OpenAI example (75% terminal value) is extreme, but the principle applies to every valuation.

This is why a valuer's assumptions about long-term growth rates matter more than their assumptions about next year's profit. A 1% change in the terminal growth rate can change a valuation by 10-15%. When reviewing a valuation, scrutinise the terminal value assumption more carefully than any other input.

Part 5: The Risks

The Commoditisation Risk

If AI models become indistinguishable and margins compress to cloud-computing levels, the revenue multiples contract dramatically. OpenAI at a 10x revenue multiple (still generous for infrastructure) would be worth $120 billion, not $1 trillion.

The Capital Market Risk

These companies rely on continuous access to capital markets. If interest rates remain elevated or equity markets turn risk-off, their ability to fund the Stargate-level capex is constrained. SpaceX's 24 years of private funding worked because Musk had access to capital. A public company with quarterly reporting obligations may not have the same luxury.

The Technology Risk

Another breakthrough-an entirely new architecture, quantum computing, or a paradigm shift in how AI is delivered-could render today's infrastructure investments obsolete. The AWS analogy assumes that today's cloud infrastructure is the permanent foundation. In AI, nothing is permanent.

For a practical look at how DCF models handle high-growth terminal value assumptions, the sensitivity analysis templates at ExcelWiz walk through the mechanics here.

Conclusion

The valuations being assigned to OpenAI, SpaceX, and Anthropic challenge every assumption in traditional business valuation. The capitalised excess earnings method assumes that earnings will stabilise, that capex is a cost rather than an investment, and that the terminal value is a small component of the total. None of these hold for the AI giants.

Are the valuations justified? The excess earnings method says no. The AWS analogy and the AGI option say maybe. The market is making a bet: that today's billions in capex create tomorrow's infrastructure monopolies, and that the long-term returns justify the short-term pain.

Whether you agree with that bet or not, the IPO wave of 2026 will be the most consequential test of valuation theory in a generation. For SME owners watching from Australia, the lesson is simpler: growth matters, capex is a signal, and the terminal value is always where the real money is made or lost.

Frequently Asked Questions

Why don't traditional valuation methods work for OpenAI and Anthropic?

The capitalised excess earnings method assumes earnings will stabilise at a sustainable level. AI companies are spending billions on infrastructure today that may generate returns for decades. Traditional DCF requires a terminal value assumption so far in the future that the calculation becomes speculative, with 70-80% of value coming from the terminal period.

What revenue multiple is reasonable for a pre-IPO AI company?

There is no historically grounded multiple. OpenAI at a $1 trillion valuation and $12B revenue implies an 83x revenue multiple-an order of magnitude above the SaaS average of 8x-12x. These multiples reflect a call option on AGI and monopoly infrastructure returns, not current earnings.

How does SpaceX's valuation compare to its public market peers?

SpaceX's approximately $1.75 trillion valuation would make it one of the five most valuable companies in the world if it were public today. It trades on the belief that Starlink will achieve monopoly-like margins on global satellite broadband, and that Starship will open entirely new revenue categories in point-to-point transport and off-world logistics.

What is the single biggest risk in these valuations?

The capital expenditure thesis. These companies are spending billions today on data centres, satellites, and rockets with the expectation of long-term returns. If AI models commoditise faster than expected, or if Starlink faces price competition from terrestrial fibre and LEO competitors, the terminal value collapses and so does the valuation.

Is there a historical parallel for these valuation levels?

The closest parallel is Amazon in the late 1990s and early 2000s. Amazon traded at extreme revenue multiples for years while it built its fulfilment infrastructure. The market was betting on infrastructure dominance, not current earnings. OpenAI, SpaceX, and Anthropic are making a similar bet on AI and space infrastructure.