SpaceX Reclaimed Its $135 IPO Price on 10 August 2026: The CFO Put $100bn of ARR on a $31bn Run-Rate — and the Contracts That Have to Land by December

The stock is back above where it started. SpaceX closed at $138.74 on Monday 10 August, up about 4%, its first close above the $135 IPO price since 15 July — and it got there through the week the whole market was pointed at. Up to 911.5 million restricted shares became eligible to trade on Thursday 6 August, roughly $98.7 billion at Wednesday's $108.27 close and more than the 638.9 million floated in June's IPO. The feared cliff-edge selloff never arrived: shares closed 6.1% higher on Thursday, rose 15.83% on Friday to $133.11, then added about 4% on Monday. Doubling the tradable supply of a company coincided with one of its largest advances, and the reason is not that the market ignored the supply. But the number that now sets the terms is the one chief financial officer Bret Johnsen put on the earnings call: $100bn of annualised recurring revenue by the end of the year, measured from a run-rate of $31bn.

What actually happened

The tranche opened on the second full trading day after the first earnings release, allowing holders to sell 20% of their eligible restricted stock, up to 911.5 million shares. CNBC put the size at just over 911 million shares, about 7% of shares outstanding and more than the 639 million sold in the record IPO. Bloomberg reported the shares steady after the expiry.

The arithmetic across those three sessions is the headline: $108.27 to $138.74 is a 28.1% recovery, and it began in the exact window in which the tradable share count more than doubled. CNBC reported that Monday's close was the first above the $135 IPO price since 15 July, and that the shares had briefly touched the benchmark in early trading. Note what that milestone is and is not. Reclaiming an issue price means the average buyer in the June book is no longer underwater; it does not mean the stock has recovered its post-listing move, because it remains well below its 16 June intraday high. This is a recovery inside a much larger drawdown that has now reached its starting line, not one that has passed it by any distance.

The move was not confined to one ticker. Rocket Lab closed up 9.46% at $82.83 and AST SpaceMobile rose 6.80% to $71.94 in the same session, which tells you the market was repricing a sector's risk premium rather than reacting to a single company's share register. It landed inside a strong tape: the S&P 500 took a record close of 7,757.64, up 0.62%, capping its strongest week since April at +3.6%, with the Nasdaq Composite up 1.3% at 26,690.62 and the Dow up 151.83 points at 54,036.93.

Why doubling the float lifted the price

Three things sat between an expiring restriction and an actual share changing hands, and all three worked in the same direction this week.

The first is willingness. Employees and early investors hold at cost bases far below the current price — Ryan Lee, senior vice president for product and strategy at Direxion, told CNN that "with these employee share unlocks, you're going to have natural sellers entering the market", people who held "far before it was printing at $135 on the IPO". That cuts both ways: a low cost basis creates a seller still in profit at any plausible price, but also a holder under no pressure to act in any particular week.

The second is staggering. SpaceX and its underwriters did not use a conventional 180-day cliff but nine tranches, which converts a single wall of supply into a sequence of smaller ones — and gives a holder who wants out a series of later exits rather than one.

The third is that the overhang was already in the price, and its removal is itself a source of demand. Bernstein analysts noted before the event that "we have found many investors unwilling to buy SpaceX shares in the interim, based on fundamentals, given the lock-up overhang that will exist at least into December". Buyers who stood aside for that reason return as each date passes without incident. That is the unglamorous mechanism behind a well-flagged unlock trading firm: the event does not add buyers by making the company better, it adds them by removing a reason to wait.

The denominator that halves short interest

This is the part of the arithmetic that is easy to miss and hard to argue with, and it is the best explanation for why Friday's move was as large as it was.

Before the unlock, short interest had become extraordinary for a company of this size. CNBC reported, citing S3 Partners, that short sellers held an estimated 206 million shares — 32.2% of the publicly tradable float, worth about $23.6 billion notionally, overtaking Tesla's roughly $22 billion. When the stock began trading in June, only about 40 million shares were sold short, around 5% to 7% of float.

Now hold that 206 million position completely still and change only the float. Against 638.9 million tradable shares it is 32.2%. Against roughly 1,550 million it is about 13%. Nobody covered. No short seller changed their mind. The single most cited statistic about this stock was cut by more than half by an event that had nothing to do with short sellers at all.

That matters because crowding is what makes a stock reflexive, and reflexivity has no preferred direction. Borrow utilisation near 95% meant nearly every lendable share was already out on loan — the condition that let a $1.43 trillion company fall 14% on Wednesday. It is also the condition that lets it rise 15.83% on Friday when the news flips, because a crowded short base facing a positive catalyst is a source of forced demand. The float doubling does eventually damp this — more lendable shares, lower utilisation, more depth on both sides — but that is gradual, and it had not taken hold by Friday. The largest up-day since the IPO happened while the old crowding was still mostly intact.

The capex line got a name and a postcode

Wednesday's 13.9% fall was not about the unlock. It was about a single line: $18.369 billion of capital expenditure against $7.814 billion of quarterly revenue, with about 86% of it in the AI segment, against $12.73 billion of capex across all of 2025. Every segment beat, by roughly $880 million in aggregate, and the stock fell anyway — because a beat of $880 million is a rounding error next to a step-change of that size in the capital programme.

Then, on 6–7 August, that number acquired an object. Tesla and SpaceX confirmed that Terafab, their joint advanced chip plant, will be built in Grimes County, Texas, with a $16.8 billion first phase, a planned workforce of at least 3,000 and a total footprint the companies put above 100 million square feet, bringing logic, memory, packaging and testing onto one site. Texas extended a $30 million Texas Enterprise Fund grant to SpaceX for it. Argus Research raised its rating on the stock the same day, and Bernstein lifted its own estimates.

The mechanism here is worth naming carefully, because it is not "good news happened". An unexplained $18.4 billion is a funding requirement of unknown duration and unknown return. A sited, phased, named project with a budget is the same money with a schedule attached — and a schedule is what lets anyone model it. Note the scale: the entire announced first phase, $16.8 billion, is about nine-tenths of what the company already spent in the June quarter alone. Terafab does not shrink the capital programme. It converts part of it from a number into a plan, and markets price those two things differently even when the cash is identical. Whether the plan earns its cost is a separate question that Friday did not answer, and a chip fab whose principal customers are its own two owners carries a demand risk that an external order book does not.

Vertical integration into memory and logic also puts these companies on the other side of the shortage that has been repricing the whole hardware chain — the squeeze we unpacked in the Samsung and SK Hynix selloff, and the reason Chinese DRAM has been turning up inside brand-name laptops. Building your own fab is the most capital-intensive available answer to a supply constraint, and the one with the longest lead time.

The $100bn is a run-rate, not a year

The most consequential thing said in the quarter was not in the tables. On the earnings call, chief financial officer Bret Johnsen said the company is on pace to reach $100 billion in annualised recurring revenue by the end of 2026. Set that against the starting point: $7.814 billion of revenue in the June quarter annualises to roughly $31 billion.

Taken as a claim about calendar-year revenue, that would be impossible, and reading it that way is the single most common error made with this number. ARR is an exit run-rate — a final period's rate multiplied out to a year, not cash booked over the year. So the question is not whether the company earns $100 billion in 2026. It is whether the December rate of billing, multiplied by twelve, prints near $100 billion. That is a far smaller thing to ask, and it is a question about a handful of contracts with published start dates rather than about growth in general.

Four of those pieces are disclosed and dated.

Annualising every disclosed piece at its stated full rate gets to roughly $67 billion. That arithmetic is ours, not the company's — SpaceX has not published a component build-up, and the figures above mix a reported quarter with contracted rates that begin at different times. It is still the useful exercise, because it locates the gap rather than arguing about the target. Deutsche Bank's Edison Yu put the cloud contribution alone at "$45-50bn of ARR exiting December" and called the overall target "likely very achievable, driven mainly by neocloud and Cursor contribution". Analysts at Wolfe Research, looking at the same report, wrote that while there were "big beats" in the quarter, "we would advise not misunderstanding aspirations of mgmt from most likely outcomes".

Two features of that order book deserve naming, because they are structural rather than a matter of opinion. The first is concentration: the entire $1.6 billion of second-quarter neocloud revenue came from one customer, Anthropic. Renting compute is a business with very few possible buyers, since almost nobody else needs it at this scale, and revenue that arrives from one or two counterparties reprices on their capital-allocation decisions rather than on any broad demand cycle. The second is that part of the jump is inorganic. Cursor's roughly $2.6 billion does not represent anything the existing business built; it arrives with the $60 billion all-stock acquisition of Anysphere, announced on 16 June and expected to close in the third quarter subject to regulatory approval. All-stock consideration is paid in shares, which is the same variable the rest of this piece is about — new shares change the outstanding count, and where they sit relative to the free float determines whether the index arithmetic below sees them.

Supply and index demand are still the same variable

Nasdaq's index methodology does not weight the Nasdaq-100 on total shares outstanding. To preserve investability, a company's total shares are capped at three times its free-floating shares, and securities with less than 20% free float are scaled to that float. That rule is the entire explanation for the otherwise absurd fact that a company valued near $1.43 trillion carried about a 1% index weight.

Three times a 638.9 million float is about 1.92 billion capped shares; three times a 1,550 million float is about 4.65 billion. TD Securities estimated to CNN that the weight could rise above 3.5% when the index rebalances in September — a review that will capture the 20 August tranche as well. The price has changed the other input twice now: at $133.11 the market capitalisation was about $1.75 trillion against roughly $1.43 trillion on Wednesday, and Monday's $138.74 takes the same share count to about $1.82 trillion. The rebalance applies a bigger multiplier to a bigger number, and passive funds tracking NAS100 and NQ buy at whatever the price is on the day, not the price when the estimate was made. That is the awkward feature of an estimated index weight: it is quoted as a fixed figure and computed from two inputs that both keep moving, which is why the September review is the only date on which any of it becomes real.

So the two flows remain opposed and non-simultaneous. Insider selling can begin the moment a tranche opens; index weight changes only at the benchmark's scheduled review. This week showed what it looks like when the first flow fails to show up on schedule. It says nothing about the second, which has a published date.

Where the currency channel is, and what is still unsettled

This is not a foreign-exchange story, and forcing it into one would be the wrong read. No major currency has meaningful sensitivity to a single company's float.

There is one honest indirect thread, and Terafab strengthens it. Capital spending at this scale is a real component of US business investment, and the datacentre and semiconductor buildout is now large enough to register in growth, construction and electricity-demand data. Growth is one of the five factors the meter scores, and the discount rate applied to revenue expected years out is set by the policy path, which is the interest-rate factor. Both are tracked on the USD currency page. That is the extent of it — a slow-moving contribution to the American growth and rates picture, not a channel that carries a lock-up expiry into a currency pair. The same distinction applied to AMD's record data-centre quarter, which also fell more than 8% on the shape of its spending rather than the size of its revenue.

What these sessions settled is narrow: the first and largest tranche cleared without disorder, the float is now roughly 11.8% rather than 4.9%, the capital programme has one named project inside it, and the stock has traded back above the price at which it was sold in June. What none of that settles is how much stock insiders actually sold, whether the buildout earns a return, whether Starlink's ARPU stabilises after falling to $66 from $85 a year ago, and how the eight remaining tranches land — 319 million shares on 20 August, roughly 700 million in September, with Bernstein estimating the float could reach as much as 40% of shares outstanding by December. Musk's more than 6 billion shares, carrying the majority of the voting power, are restricted until June 2027.

The December ARR target adds a second set of checkable dates to the first. It is worth being clear about what a hit or a miss would actually tell you. Reaching $100 billion of exit run-rate would confirm that a handful of compute contracts ramped on schedule; it would not by itself say the $18.369 billion quarterly capital programme earns its cost, because a run-rate is a billing rate and returns are a margin question that no ARR figure answers. Missing it by a quarter would mean a contract started late. Neither outcome resolves the thing the capex line actually asks, which is whether demand for rented compute persists at this price once the buyers have finished building their own.

The framing that survives all of it is the one the index rulebook already encodes. SpaceX is a very large company with a small tradeable slice, the slice is scheduled to grow on dates that are already published, and the past week proved only that a published date had been priced. The next one starts from a different price and a different level of crowding, which is precisely why it is not the same experiment.

To learn how Pip Theory builds its fundamental scores, see the methodology overview.

Educational macro context only — not investment advice.