SpaceX’s $500 Billion AI Revenue Bet: The Math Investors Are Missing

SpaceX is asking investors to believe in one of the biggest business transformations in corporate history: a rocket company becoming one of the world’s largest sellers of artificial intelligence compute in roughly 18 months. 

Elon Musk says 10 gigawatts, or GW, of AI capacity could produce between $300 billion and $500 billion of annual revenue. The multiplication is correct. The investment case is not yet proven because three important variables are hidden inside that simple equation: how much capacity actually becomes operational, how much customers use and how long premium AI pricing survives.

Let’s break down what SpaceX has already delivered, what the $500 billion claim really assumes and what its current valuation appears to be pricing in.

More importantly, let us build a simple framework investors can update every quarter instead of relying on one ambitious headline.

What Did Elon Musk Say About SpaceX AI Revenue?

During an August 11 company address, Musk said SpaceX’s AI revenue could exceed the rest of the company’s revenue as early as September 2026 and significantly exceed it during the fourth quarter.

He also said SpaceX plans to increase its AI compute capacity from 1.4 GW currently to 10 GW by the end of 2027. Musk estimated that every watt of capacity could produce between $30 and $50 of annual revenue.

That gives us the headline calculation:

10 billion watts × $30 per watt = $300 billion

10 billion watts × $50 per watt = $500 billion

Musk also suggested AI could eventually represent 99% of SpaceX’s value. This would make rockets, spacecraft and even Starlink look more like the infrastructure supporting a much larger AI platform.

However, the most important part of the statement is not the $500 billion number. It is the phrase “by the end of next year.”

Will SpaceX Make $500 Billion in AI Revenue in 2027?

No. At least not based on the information currently available.

SpaceX wants to reach 10 GW of capacity by the end of 2027. Therefore, $300 billion to $500 billion is closer to an annualised revenue run-rate entering 2028.

That is different from recognising $500 billion of revenue during calendar year 2027.

Consider a data centre that becomes operational on December 31, 2027. It may have the capacity to produce billions of dollars over the following 12 months, but it contributed almost nothing to 2027 revenue.

This distinction matters because revenue run-rate measures the speed of the business at a specific point in time. It does not show how much revenue was actually earned during the year.

A more accurate description of Musk’s target would be:

SpaceX wants enough AI capacity operating by the end of 2027 to support an annual revenue run-rate of $300 billion to $500 billion during 2028.

That is still an extraordinary goal, but it is not the same as SpaceX earning $500 billion next year.

Can AI Become SpaceX’s Largest Revenue Business by September 2026?

SpaceX generated $2.56 billion of AI revenue during the second quarter of 2026. Its connectivity and space businesses together generated $5.25 billion.

Here is the approximate monthly comparison:

If Musk’s statement means AI revenue will exceed all other businesses combined, monthly AI revenue must rise from approximately $854 million to more than $1.75 billion.

That requires a jump of around 105% from the Q2 monthly average.

If he only meant AI will become SpaceX’s largest individual segment, it needs to move above connectivity’s monthly average of approximately $1.43 billion. That still requires growth of around 68%.

Either interpretation represents a large increase within a few months. However, the target is not impossible. SpaceX has signed major cloud-compute agreements that are beginning to contribute revenue, while its pending Cursor acquisition could add another fast-growing software business.

The company reported $14.1 billion of contracted cloud sales during Q2. These agreements contributed approximately $1.6 billion of incremental AI infrastructure revenue during the quarter. SpaceX’s Q2 earnings release defines contracted sales as enforceable, non-cancellable revenue, which makes it stronger than a general pipeline estimate.

The September claim is aggressive, but unlike SpaceX’s lunar factories or orbital data centres, investors will not have to wait years to test it. The Q3 and Q4 results should provide a clear answer.

How Does SpaceX’s AI Business Actually Make Money?

Calling SpaceX an “AI company” can be confusing because the group now contains several different businesses.

The simplest way to understand it is through what we call the SpaceX 4R Stack.

The rocket layer allows SpaceX to launch satellites at a lower internal cost. The router layer uses those satellites to sell internet connectivity. The rack layer sells computing power to AI companies. The reasoning layer sells models, coding tools, agents and other AI products.

Each layer can strengthen the next one.

But there is an important warning. Vertical integration creates value only when every layer is efficient. Otherwise, profitable businesses may simply fund weaker ones while the combined reporting makes it difficult to see where returns are being generated.

That is why investors need to study each segment separately.

What Do SpaceX’s Latest Financial Results Reveal?

SpaceX’s Q2 2026 results show a company with three very different financial profiles.

Source: SpaceX Q2 2026 earnings release

Revenue grew 92% year-on-year, while SpaceX’s net loss narrowed to $541 million. The company also had approximately $100 billion in cash and marketable securities at the end of June, largely supported by its record IPO.

But the segment data tells a more interesting story.

Starlink Is Funding the Experiment

The connectivity business generated a 38.6% operating margin during Q2. It had 12 million Starlink subscribers, up from 6 million one year earlier, and generated $1.66 billion of operating profit.

This is SpaceX’s current economic foundation. Starlink is helping fund Starship development and the company’s AI expansion.

AI Revenue Is Growing, but the Spending Is Bigger

AI revenue increased 247% year-on-year to $2.56 billion. That is impressive growth.

However, SpaceX spent $15.83 billion on AI capital expenditure during the same quarter. In simple terms, it invested more than six dollars in AI infrastructure for every dollar of AI revenue recognised.

The AI segment reported adjusted EBITDA of $1.15 billion, giving it an attractive-looking 44.7% adjusted EBITDA margin. But it still recorded a $1.26 billion operating loss.

The difference largely comes from depreciation and share-based compensation. For a capital-heavy data-centre company, depreciation is not an imaginary cost. Servers, chips and supporting equipment can become economically outdated quickly.

This is why investors should not value SpaceX’s AI business using adjusted EBITDA alone.

Free Cash Flow Remains the Main Financial Pressure

During the first half of 2026, SpaceX generated $3.47 billion of operating cash flow but spent $28.48 billion on capital expenditure.

That results in approximately $25 billion of negative free cash flow before adjusting for acquisitions or other investments.

SpaceX has a strong balance sheet, but the present AI expansion is being financed by IPO proceeds, debt and Starlink cash flow rather than self-funded AI profits.

What Does SpaceX’s 10 GW AI Compute Target Mean?

A gigawatt measures power. One GW equals one billion watts.

SpaceX’s 10 GW target therefore represents enough maximum power capacity to support an enormous number of AI processors and related equipment.

But nameplate capacity is not the same as productive capacity.

Think of it like an airline. The number of seats in its aircraft represents capacity. But revenue depends on how many seats are occupied and how much each passenger pays.

Similarly, AI revenue depends on three variables:

Annual AI revenue = Capacity × Utilisation × Revenue per active watt

SpaceX’s headline calculation effectively assumes that the full 10 GW becomes operational, customers use that capacity and the company consistently earns between $30 and $50 per watt.

Here is how the result changes under different assumptions:

The valuation column uses SpaceX’s current enterprise value of approximately $1.87 trillion and gives no separate value to Starlink or the space business. It is meant to show sensitivity, not provide a price target.

The real stretch becomes visible when we compare Musk’s target with current performance.

SpaceX generated $2.19 billion of AI solutions and infrastructure revenue during Q2. Annualising that figure gives $8.78 billion. Dividing it by the quarter-end capacity of 1.4 GW produces roughly $6.27 of annual infrastructure revenue per nameplate watt.

Musk’s target of $30 to $50 per watt requires revenue per watt to improve by approximately 4.8 to 8 times.

This comparison is imperfect because capacity increased during the quarter and recently signed contracts were still ramping. Still, it shows that SpaceX must achieve more than a sevenfold capacity increase. It also needs sharply better monetisation of every watt.

How Could SpaceX Increase AI Revenue per Watt?

There are four possible reasons.

• First, new capacity could be based on more powerful processors, allowing SpaceX to charge more for every unit of electricity consumed.
• Second, utilisation may improve as large cloud contracts with Anthropic, Alphabet and other customers ramp up.
• Third, SpaceX could move beyond leasing compute and sell higher-value services such as Grok models, APIs, enterprise agents and Cursor subscriptions.
• Fourth, greater vertical integration through its Terafab semiconductor project could eventually reduce chip and infrastructure costs.

This is where the distinction between selling compute and selling intelligence becomes important.

Selling compute is similar to renting out a furnished office. The customer pays for access to the infrastructure.

Selling intelligence is more like operating the business that works inside that office. The supplier owns the software, customer relationship and intellectual property, creating the possibility of higher margins and stronger customer loyalty.

The best version of SpaceX is not simply a larger CoreWeave. It is a vertically integrated platform that controls chips, electricity, computing infrastructure, connectivity, AI models and software distribution.

The question is whether it can build that platform before competitors reduce AI prices or customers build more of their own capacity.

Is Morgan Stanley Right That Investors Are Undervaluing SpaceX AI?

Morgan Stanley argues that the market assigns only around $12 per SpaceX share to the AI business. Based on approximately 13.18 billion shares outstanding, that implies an AI valuation of roughly $158 billion.

The analysts believe this is too conservative because investors are focusing mainly on SpaceX’s neocloud contracts and giving limited value to Grok and Cursor. Morgan Stanley has a $300 price target on SpaceX, compared with its August 12 closing price of $146.15.

We partly agree with the strategic argument, but not with the idea that $158 billion is obviously cheap.

Consider the closest public comparison. CoreWeave is used here only as a comparable for SpaceX’s neocloud operations because both companies sell GPU-based AI computing capacity through large cloud contracts. It does not capture the potential value of Grok, Cursor, Starlink or SpaceX’s other businesses, but it provides a useful baseline for valuing the AI infrastructure already generating revenue. 

SpaceX AI and CoreWeave currently generate almost identical quarterly revenue. Yet Morgan Stanley’s implied SpaceX AI valuation is approximately 50% higher than CoreWeave’s enterprise value.

SpaceX deserves some premium because it is growing faster, has a stronger balance sheet and owns consumer AI products. But the comparison shows that the market is not valuing its AI business as if it were worthless.

Morgan Stanley’s argument becomes convincing only if Grok and Cursor develop into major software platforms.

How the $60 Billion Cursor Acquisition Could Help SpaceX AI

SpaceX has agreed to acquire Cursor, the AI coding platform developed by Anysphere, for $60 billion. The deal is expected to close during Q3 2026. Associated Press reported that Cursor will become a wholly owned SpaceX subsidiary.

Morgan Stanley estimates that Cursor’s annualised revenue could rise from $4 billion in June 2026 to:

At its current $4 billion revenue run-rate, SpaceX is paying around 15 times revenue. That is not a cheap acquisition.

However, the deal gives SpaceX something its compute contracts do not: direct access to enterprise users and software developers.

Cursor can distribute Grok models to paying customers. Grok can reduce Cursor’s reliance on competing models. SpaceX’s infrastructure can provide the computing capacity required by both products.

If this integration works, SpaceX can earn money from the infrastructure, model and application layer. If it fails, the company may have paid $60 billion for a fast-growing application in a market with rapidly changing technology and low switching costs.

Can SpaceX’s Grok Compete With OpenAI, Anthropic and Google?

The early evidence has improved.

Grok 4.6 scored 61 on the Artificial Analysis Intelligence Index, matching OpenAI’s GPT-5.6 Sol and sitting slightly below Anthropic’s leading models. It also showed strong performance on long-running agent tasks.

Pricing starts at $2 per million input tokens and $6 per million output tokens. That is over 60% cheaper than some similarly capable frontier models, according to Artificial Analysis. SpaceX’s official Grok 4.6 announcement also shows strong results across coding and agent benchmarks.

This is real progress. But one strong model release is not a durable moat.

AI benchmarks can change every few weeks. OpenAI, Anthropic, Google and Chinese developers continue to release new models. A sustainable advantage requires more than leading a temporary leaderboard. It requires reliable models, low inference costs, enterprise security, customer distribution and the ability to keep improving without destroying margins.

The best signal will not be Grok’s benchmark rank. It will be how much revenue Grok and Cursor generate independently of SpaceX’s infrastructure contracts.

The SpaceX Evidence Ladder: What Is Proven vs Speculative?

SpaceX contains everything from recognised revenue to ideas involving lunar factories. Investors should not assign the same confidence to every part of the story.

The following Evidence Ladder can help:

SpaceX’s IPO presentation estimated a $26.5 trillion long-term AI opportunity, but this is a total addressable market, not a revenue forecast. The same SEC-filed investor material says orbital AI compute deployment may begin in 2028 and highlights significant technical, capital, power, chip, regulatory and integration risks.

The common investment mistake is jumping directly from Level 1 to Level 5.

A better approach is to move one level at a time. First verify revenue. Then verify contracts. Then verify the revenue run-rate. Capacity forecasts and long-term markets should receive meaningful value only as supporting evidence appears.

SpaceX Stock Valuation: What Is SPCX Already Pricing In?

SpaceX closed at $146.15 on August 12, giving it a market capitalization of approximately $1.93 trillion and enterprise value of around $1.87 trillion. The SPCX stock remains above its $135 IPO price, but well below its post-listing peak.

Annualising Q2 results gives:

These multiples show that investors are already valuing SpaceX on future results rather than current earnings.

A simple reverse valuation makes the expectations clearer.

Suppose a mature SpaceX deserves an enterprise value equal to 25 times operating profit and eventually achieves a 30% operating margin.

At its current $1.87 trillion enterprise value, the company would need roughly:

$1.87 trillion ÷ 25 ÷ 30% = $249 billion of annual revenue

At Morgan Stanley’s $300 price target, SpaceX’s market value would approach $3.95 trillion. Using the same assumptions, the company would need approximately $519 billion of annual revenue.

These are not forecasts. They show what must eventually become true under one reasonable valuation framework.

The result is revealing. SpaceX’s current valuation already requires something close to the lower end of Musk’s AI ambition, unless investors assign substantial separate value to Starlink and the space business.

Morgan Stanley’s $300 target requires either something close to the $500 billion outcome, unusually high long-term margins, or a premium valuation multiple that remains elevated for years.

What Could Break the SpaceX AI Story?

The largest risk is not that AI demand disappears. It is that SpaceX builds capacity faster than profitable demand develops.

SpaceX has the money to absorb mistakes today. It does not have unlimited time.

If revenue per watt rises, AI margins improve and free cash flow begins recovering, the capital expenditure will look like a well-timed investment.

If capacity rises while revenue per watt falls, SpaceX risks becoming an expensive infrastructure provider in a market where customers hold most of the bargaining power.

Our View: Is SpaceX’s $500 Billion AI Target Realistic?

The $500 billion figure is mathematically possible but financially unproven.

The capacity target alone is not enough. SpaceX must increase operational capacity more than sevenfold, improve infrastructure revenue per watt by as much as eight times and keep most of that capacity occupied at premium pricing.

We believe the more realistic near-term question is not whether SpaceX can generate $500 billion. It is whether the company can cross three smaller milestones:

1. AI revenue genuinely becomes larger than the rest of SpaceX.
2. The December 2026 revenue run-rate approaches $100 billion.
3. AI revenue grows faster than AI capital expenditure after 2027.

Morgan Stanley is right that investors may be underestimating SpaceX’s ability to combine infrastructure, models and software distribution.

But Morgan Stanley’s $12-per-share implied AI valuation is not clearly cheap when compared only with neocloud businesses. It becomes attractive only if Grok and Cursor develop into large, independent software platforms.

Our current view of SpaceX is therefore straightforward:

SpaceX is a highly profitable satellite-connectivity business funding an enormous AI infrastructure build, with a valuable but still unproven software option attached.

It is no longer just a rocket company. But it is also not yet a proven full-stack AI platform. The next few quarters should reveal which description becomes more accurate.

What Should SpaceX Investors Track Every Quarter?

Instead of focusing only on launches or AI model announcements, investors should maintain this six-metric dashboard:

If operational capacity, revenue per watt and free cash flow improve together, SpaceX’s AI strategy is working.

If capacity rises but revenue per watt and free cash flow deteriorate, the business may be scaling spending rather than economic value.

That is the difference between building the world’s most valuable AI platform and merely building the world’s most expensive collection of data centres.