The Two Sides of SpaceX: Starlink Earns 18 Million RMB Daily, AI Burns 170 Million RMB Per Day

The first quarterly report since SpaceX went public reveals an extreme sense of division.

The financial report shows SpaceX achieved revenue of $7.8 billion in the second quarter, up 92% from $4.1 billion a year ago; adjusted EBITDA reached $3.5 billion, far above $1.2 billion last year. However, SpaceX's capital expenditure in the second quarter soared to $18.4 billion, 2.4 times its revenue during the period. Of this, AI-related capex alone amounted to $15.8 billion, which means the company spent over $170 million on average every day.

On the earnings call, Elon Musk and CFO Bret Johnsen spent more than an hour trying to reassure the market, giving detailed explanations for the destination and potential returns on the massive $18.4 billion in capital spending in a single quarter. Although management emphasized that the huge current investments in AI computing power could recoup costs in a short period, the market was clearly unconvinced by these long-term promises, as the enormous spending has yet to translate into actual profits in the income statement. SpaceX shares dropped more than 7% after hours.

Another key factor intensifying risk aversion is the impending wave of massive share unlocking. On August 6, 910 million internal shares at SpaceX will be freed up for trading, an amount equivalent to 1.4x the current freely traded shares. Management failed to proactively address this liquidity shock during the call, only briefly responding in the Q&A session, which further deepened investor uncertainty.

Side A: Starlink, Earning $18 Million Net Profit Per Day

The Starlink business generated $4.29 billion in Q2 revenue, up 66% year-on-year. Operating profit was $1.66 billion, up 79%, with a profit margin of 38.6%. Starlink added 1.7 million net new users for a total of 12 million, covering 167 countries. Enterprise and government revenues reached $1.8 billion, more than double year-over-year.

Figure 1: SpaceX Q2 FY2026 Comparison of Revenue and Operating Profit from Three Business Units (Source: SpaceX SEC 8-K)

The US government awarded contracts worth more than $6 billion in Q2. American Airlines signed an in-flight Wi-Fi agreement. President Gwynne Shotwell stated on the call that Starlink aims to become the fourth largest US carrier—behind AT&T, Verizon, and T-Mobile. These three companies together have annual revenues of about $600 billion.

ARPU stabilized at $66. This figure was $85 in Q2 last year and also $66 in Q1 this year. The fastest user growth areas are in Latin America, Africa, and Southeast Asia, where pricing is much lower than North America, thus pulling down the average. However, a substantial portion of the 1.7 million net additions also come from aviation, maritime, and government clients, whose contract values per customer are much higher than individual subscribers.

Starlink's profit growth rate (79%) outpaced its revenue growth (66%), reflecting the effect of spreading fixed costs. There are now over 10,200 satellites in orbit, with each V3 broadband satellite having a downlink capacity of 1Tbps, 10 times that of V2. Shotwell noted that mass launch of V3 satellites will occur by mid-next year, which will give a dramatic boost to bandwidth.

If Starlink is separated and analyzed on its own, based on Q2 annualized figures, it would be a telco with $17.2 billion annual revenue and $6.6 billion in operating profit. T-Mobile, the fourth largest US carrier, had $81 billion in revenue and $13 billion in profit last year. Starlink is still far from that scale, but its growth rate is in a different league.

Side B: AI, Burning $170 Million a Day

The AI business reported $2.56 billion in revenue, up 247% year-on-year and 213% quarter-on-quarter. Cloud services contributed roughly $1.6 billion; AI solutions revenue jumped from $475 million in Q1 to $2.194 billion in Q2.

Adjusted EBITDA turned positive for the first time, flipping from –$609 million in Q1 to +$1.146 billion. However, operating loss still stood at $1.257 billion—before accounting for depreciation and amortization.

What really made the market nervous was the capital spending. The AI division spent $15.8 billion in Q2. For comparison, Nvidia's R&D budget for all last year was $12.9 billion and Meta's annual capex across all business lines was $27.2 billion. SpaceX's AI spending in one quarter surpassed Nvidia's full-year R&D, and amounted to about 60% of Meta's annual total capex.

Figure 2: SpaceX AI Segment Capital Expenditure vs. Revenue Trend (Source: SpaceX SEC filings)

This money is mainly poured into the Colossus data center in Memphis, Tennessee. At the end of Q2, nameplate computing capacity stood at 1.4 GW, with a year-end target of 2 GW. When asked about future plans on the call, Musk said the internal target is 20 GW power and cooling by the end of 2027, though he personally expects about 15 GW in practice—a roughly 10x jump from 1.4 GW.

CFO Johnsen provided a figure on the call: the payback period for incremental computing power investment is now less than a year. He said capital expenditures can almost be treated as operating expenses—meaning at current customer contract prices and scales, the money put in today can be recouped in 12 months.

In the first three weeks of Q3, the company signed another $6.7 billion in incremental cloud contracts, with a service period of about half a year, starting billing in October.

But $15.8 billion is already spent. $6.7 billion are merely contracts. During the call, multiple analysts pressed the same logic: you say the payback is less than one year, but you also spent $7.7 billion last quarter; $23.5 billion spent in two quarters. When will we see the corresponding returns in the income statement?

Management gave no concrete timetable. Musk simply said that by December, calculating annualized by current month revenue, recurring revenues will reach $100 billion. He added that the actual number may be higher. The company’s goal for $1 trillion in revenue has been moved up from 2031 to 2030, and for 2029 there is a "non-zero probability"—his own words.

The market clearly isn't very interested in these long-term "non-zero probabilities." The pressing problem is more specific: tomorrow (August 6), 910 million internal shares will be unlocked, equivalent to 1.4x the current trading float. On the call, management did not proactively bring this up and only briefly responded when pressed during the Q&A.

Starship Still Burning Cash

Launch business revenue was $962 million, up 29% year-on-year, but posted an operating loss of $542 million. Thirty-eight launches were completed in Q2. The main source of losses is Starship.

Over the past 90 days, Starship V3 has completed two successful flights. The 13th flight validated core capabilities in orbital missions and returning for launch tower capture. The 14th flight delivered Starlink V3 satellites to operational orbit for the first time.

The thermal shield issue—which Musk himself described as Starship’s greatest technical barrier—has, according to him, been resolved. The next step is for the upcoming test flight at the end of the month to attempt simultaneous capture recovery of both the first-stage booster and the spacecraft.

R&D spending increased by $389 million year-on-year, and the aerospace segment posted an adjusted EBITDA of negative $200 million. Musk’s goal is to achieve at least one Starship launch per day within a year, driving orbit costs down to below 1% of traditional methods. He said, if you draw a global launch mass bar chart, other competitors’ columns would be only one pixel high—provided the SpaceX column fits on the chart at all.

Can Musk Win All Three Battles at Once?

This quarterly report shows three completely different forms of tension pulling against each other.

Starlink is making money—and making it quickly. $1.66 billion in operating profit corresponds to 12 million users; Shotwell says the goal is to become the fourth largest carrier, ARPU is stable, V3 satellites have yet to change the picture, and enterprise growth is nearly twice the consumer side. This is a great stand-alone business that can be independently valued.

AI is burning cash recklessly. $15.8 billion in a quarter, with a theoretical payback under a year; $6.7 billion in new Q3 contracts; after Grok 4.5’s release, token consumption tripled; the Anthropic compute rental deal is $1.25 billion per month through 2029. All signs point to one thing: demand is real. But that $15.8 billion is already gone from the books, and the profits have not caught up. This is the difference between what Johnsen calls "like operating expenses" and what investors see as "like a bottomless pit."

Aerospace is caught in the middle. Starship’s technical progress is tangible—the thermal shield problem is solved, two successful flights, V3 satellites in orbit—but the R&D bills are also very real. This segment has the longest payback logic: things will improve only when launch costs fall to 1% of what they used to be.

Relying on his over 82% voting power, Musk is trying to balance three fundamentally different business tensions: Starlink’s rapid profitability, AI’s aggressive expansion, and Starship’s long-term R&D. However, whether a single company can win all three wars simultaneously remains a question the market is still waiting to see answered.