SpaceX’s $116B Lockup Tests Its All-In Nvidia Bet

Published August 6, 2026, 9:00 AM PKT. Updated with closing-bell trading data as it becomes available.

SpaceX’s $116 billion share lockup expires today, and it could not have landed at a more exposed moment. Two trading days after the company posted its first earnings report as a public company, up to 911.5 million insider shares become tradeable for the first time since SpaceX’s record-setting June IPO. That is more than three times the stock’s current public float, arriving into a market that just watched SpaceX’s AI division burn $15.8 billion in a single quarter.

If you own SPCX, lend to companies that depend on SpaceX’s compute, or track the Nvidia supply chain, today is not a normal Thursday.

The Numbers That Beat, and the Number That Spooked Wall Street

SpaceX’s second-quarter 2026 results, released August 4, told two different stories depending on which line you read. Revenue hit $7.814 billion, up 92% year over year and roughly $900 million ahead of Wall Street’s consensus estimate. The AI segment, built around Grok and xAI’s compute business, grew 247% year over year to $2.561 billion. Net loss narrowed to $541 million from just over $1 billion a year earlier. On paper, that is a strong quarter.

Then investors got to capital expenditures: $18.369 billion for the quarter, more than double revenue and nearly $5.4 billion above the roughly $13 billion analysts had modeled. Of that, $15.828 billion went to AI infrastructure alone, a 21x jump from $749 million a year earlier.

Starlink still carries the company. Subscribers doubled year over year to 12 million, even as average revenue per user slipped to $66 a month from $85, a sign that growth is increasingly coming from lower-price international markets rather than the premium U.S. base that built the business.

Why Today’s Unlock Is Different

Lockup expirations happen after almost every IPO. What makes SpaceX’s unusual is the math. The public float since the June 12 IPO has sat under 280.1 million shares. Starting today, up to 911.5 million additional insider shares, about 20% of restricted holdings and worth roughly $116 billion at recent prices, become eligible for sale. That is not a marginal increase in supply. It is a tripling.

Layer on short interest of 219.3 million shares, about 34% of the public float as of July 29, up from just 23.3 million shares in mid-June, and you have a stock where a huge chunk of the trading population is already betting against it walking into the single largest supply event of its short public life.

The stock has not needed help finding the exit. SPCX has fallen roughly 43% to 50% from its June 16 intraday peak of $225.64 and now trades below its $135 IPO price. Underwriters typically stagger IPOs and earnings dates specifically to prevent disorderly selling around an unlock. Here, the sequencing backfired: the first earnings report gave the market its most consequential data point yet (the AI capex number) just two days before the biggest supply increase in the stock’s history.

Note: Musk’s own holdings are not part of today’s unlock. Shares held by Musk and a select group of insiders remain restricted until mid-2027, well past the general 180-day backstop that expires December 8, 2026 for most other insiders.

Musk’s Nvidia-Exclusive Bet Raises the Stakes

On the earnings call, Elon Musk removed any ambiguity about where SpaceX’s AI compute strategy is headed.

“We’re exclusive to Nvidia. We think Vera Rubin architecture is the best AI computer.” Elon Musk, Founder and CEO, SpaceX, Q2 2026 earnings call, August 4, 2026

That single line moved two other stocks that had nothing to do with SpaceX’s earnings. AMD fell as much as 9% on the session, while Nvidia climbed on the reinforced commitment. It is a reminder of how thin the “second source” narrative for AI accelerators still is when one customer’s on-record preference can swing a competitor’s market cap.

The commitment funds an ambitious build schedule: 2 gigawatts of AI compute by the end of 2026, scaling toward 10 gigawatts, and potentially as high as 15 to 20 gigawatts, by the end of 2027, all running on Nvidia’s Vera Rubin NVL72 architecture, internally nicknamed “Kyber.” Musk also confirmed that prototype “Starmind” satellites, carrying Vera Rubin chips into orbit, are targeted to begin launching in 2027.

Nvidia CEO Jensen Huang has offered support, but with a caveat that matters.

“The economics are poor today, but it’s going to improve over time.” Jensen Huang, CEO, Nvidia, Q4 earnings call, reported by Business Insider

The Skeptics: Valuation Math and Orbital Physics

Not every analyst is buying the AI story at face value. Glenn Thum of Phillip Securities, a five-star-rated analyst per TipRanks, initiated coverage with a rare Sell rating and a $75 price target.

“AI carries the valuation but not the earnings.” Glenn Thum, Analyst, Phillip Securities

Thum’s underlying data point is hard to argue with: SpaceX’s AI division generated $3.2 billion in revenue in 2025 against a $6.4 billion operating loss. Even after a strong Q2, neither the AI segment nor the Space segment is profitable on its own. Starlink is still doing the heavy lifting.

There is also a harder problem than accounting: physics. Musk’s Starmind plan depends on cooling AI servers in orbit, where there is no air or water to carry heat away, only radiation. Dylan Taylor, Chairman and CEO of rival space infrastructure firm Voyager Technologies, laid out the challenge to CNBC earlier this year.

“It’s hard to actually cool things in space because there’s no medium to transmit hot to cold.” Dylan Taylor, Chairman and CEO, Voyager Technologies, CNBC interview, February 6, 2026

Independent engineering analysis backs up the concern. IEEE Spectrum calculated that a single AI server rack in orbit needs roughly 80 square meters of radiator area, about the size of a pickleball court, and that a 100-megawatt orbital facility would need something like 2,500 of those radiators. Taylor, whose company competes with SpaceX in the same space infrastructure market, called a two-year deployment timeline “aggressive.” That competitive angle matters, but the underlying thermodynamics do not care who is saying it.

There is a third risk that gets less attention: circularity. A meaningful share of SpaceX’s AI revenue comes from compute-leasing arrangements with Anthropic and Google, the same hyperscalers facing their own scrutiny over AI capex spending. When your growth partly mirrors your customers’ spending cycles, a slowdown anywhere in that chain shows up everywhere in it. For more on how that scrutiny is playing out elsewhere, see our piece on the Anthropic Claude security incident that rattled enterprise AI customers earlier this month.

What to Watch Over the Next 6 to 18 Months

Three things will tell you whether today’s unlock is a one-day liquidity event or the start of something longer.

• Unlock-day volume and closing price. Watch whether insiders actually sell into weakness or whether pre-unlock declines already priced in the supply increase. This will be visible by market close today.
• Q3 AI capex versus Q3 AI revenue. If the spending-to-revenue gap widens again next quarter, expect more downgrades along the lines of Phillip Securities’ call. If it narrows, the bull case gets easier to defend.
• Starmind’s first prototype flight in 2027. A successful, on-schedule launch would be the first real evidence that SpaceX can solve the cooling problem at scale. A delay would validate Taylor’s skepticism.

Wall Street’s consensus rating still sits at “Moderate Buy,” with 27 Buy, 6 Hold, and 2 Sell ratings across roughly 33 to 35 analysts, and an average 12-month price target near $223 to $232. That implies upside of 95% to 110% from current levels. The spread between that consensus and Phillip Securities’ $75 target tells you how unresolved this stock’s identity still is: is it a satellite and launch company that also does AI, or an AI company that happens to own the world’s best rocket fleet?

Frequently Asked Questions

The Bottom Line

SpaceX proved this week that its AI division can grow fast. It has not yet proven it can grow profitably, or that a Nvidia-exclusive bet on orbital compute is more than a spending commitment with an unsolved cooling problem attached. Today’s unlock does not change any of that math. It just adds 911.5 million shares of investors who now get to vote on it with their wallets.

Watch the close today, watch Q3 capex against Q3 AI revenue, and watch whether Starmind’s first prototype actually leaves the ground on schedule in 2027. Those three data points will tell you more about SpaceX’s next decade than this week’s headline numbers ever could.

Subscribe to The Neural Loop for same-day breakdowns of the AI and big tech earnings that move markets.