SpaceX Has Bought $835 Million of Megapacks. Musk's 10 GW Target Needs Nine to Twenty-Four Times That.

SpaceX's Q2 2026 10-Q, filed August 4, discloses in Note 17 — Related Party Transactions:

"During the three and six months ended June 30, 2026, the Company purchased $295 million and $329 million, respectively, of Megapack products from Tesla, Inc. ("Tesla") recorded in Property, plant, and equipment, net in the consolidated balance sheets. As of December 31, 2025, the Company purchased $506 million of Megapack products and $131 million of Cybertrucks at manufacturer's suggested retail price from Tesla."

Three numbers, and the arithmetic on them is being done wrong in public.

First, the correction

The disclosure gives $295M for the quarter and $329M for the half. Those are not two additive line items — the second contains the first. Q1 2026 was therefore $34 million, and the first half was a $34M quarter followed by a $295M quarter.

Cumulative purchases through December 31, 2025 were $506M. So the running total is:

A widely shared post put the total "over $1.1 billion" by adding $506M + $329M + $295M. That counts Q2 twice — $329M already includes the $295M. Sawyer Merritt's figure of "over $800 million" is the correct one.

The $1.1B version is circulating faster than the $835M version, which is the usual fate of the bigger number.

What $835 million actually bought

Tesla does not disclose Megapack unit pricing, and the 10-Q is careful to note that the Cybertrucks were bought "at manufacturer's suggested retail price" — a qualifier it conspicuously does not attach to the Megapacks. So the related-party price per kWh is not in the filing.

The best traceable proxy is Tesla's own realised figure. In Q2 2026 Tesla booked $3,139M of energy revenue on 13.5 GWh deployed — about $233 per kWh, blended across Megapack, Powerwall and solar. On that basis $835M is roughly 3.6 GWh, or about 900 Megapack 2XL units at 3.9 MWh each.

That is a floor rather than an estimate: the PP&E line likely carries installation and balance-of-system cost that Tesla's revenue line does not, which would push the implied kWh down and the implied price per kWh up. Read 3.6 GWh as "no more than this."

The ratio that makes the forecast possible

To size what comes next you need storage-per-megawatt of datacenter load. Two installations are observable:

The Cortex 2 figure comes from Joe Tegtmeyer's site observation that roughly 260 Megapacks will support that facility once fully installed, on top of 140 elsewhere at Giga Texas — about 400 across the site. Tesla's own Q2 update lists Cortex 2 at >115 MW of installed AI training compute. Note that Tegtmeyer describes some of those units as supporting the cooling towers rather than the compute directly, which is part of why the Texas ratio runs richer than Memphis.

So the observed band is 3.3 to 8.8 MWh of storage per MW of datacenter load, at 3.9 MWh per Megapack 2XL.

Sizing Musk's 10 GW

SpaceX has said 2 GW of AI compute will be installed by the end of 2026 and 10 GW online by the end of 2027 (with a more aggressive internal target above that). Applying the observed band to 10 GW:

These are our estimates from observed site ratios, not company guidance.

Against the $835M spent so far, the 10 GW build implies nine to twenty-four times the cumulative Megapack purchase to date. Against the ~3.6 GWh implied by that spend, it implies nine to twenty-four times the physical fleet.

Which is when this stops being a SpaceX story

Yesterday we sized the new Texas Megafactory at 50 GWh a year of Megapack 3, taking Tesla's total Megapack nameplate to about 110 GWh a year — 40 GWh in California, 20 GWh in Shanghai, 50 GWh in Texas. Tesla's trailing-twelve-month deployments across all customers were 49.0 GWh.

Now put the table above next to those two numbers:

• The lean case, 32.5 GWh, is 30% of Tesla's entire annual Megapack nameplate — and 66% of everything Tesla actually deployed to every customer on earth in the last twelve months.
• The mid case, 43.7 GWh, is 40% of nameplate and 89% of trailing deployments.
• The rich case, 88.2 GWh, is 80% of nameplate and 1.8x trailing deployments.

That reframes yesterday's piece. We argued that the Texas factory does not create demand, it removes a ceiling, and that the open question was whether grid-storage demand would roughly double to fill 110 GWh of nameplate. One related party's 2027 datacenter plan, taken at face value, answers a third to four-fifths of that question by itself.

The concentration this creates

$295M of Megapack purchases in a quarter when Tesla's total energy segment revenue was $3,139M is 9.4% of the segment, from a company with the same CEO. On the trajectory above it would not stay at 9.4%.

We flag this without alarm and without endorsement, because it cuts three ways at once:

• For Tesla, it is a large anchor customer that de-risks the 50 GWh Texas ramp — and a related-party revenue concentration that arrives without an arm's-length price disclosed in either filing.
• For SpaceX, the Megapacks sit in PP&E, so they land on the capex line that already took free cash flow to −$25,010M for the half. $295M against $18,369M of quarterly capex is 1.6% — the batteries are not what is straining SpaceX's cash.
• For anyone modelling either company, the same physical asset is revenue on one income statement and depreciation on the other, and both are controlled by the same person.

SpaceX still carries no Rule of 40 score on this site, because it publishes operating cash flow only year to date and there is no pre-IPO quarterly report to difference against. Tesla's score is 21.6, held down by the same capex wave this demand is part of.

The number to watch is not the next quarter's Megapack line. It is whether SpaceX's disclosed purchases keep compounding at the Q1-to-Q2 rate — $34M to $295M is 8.7x in one quarter — because two or three more quarters of that puts one customer past Tesla's ability to supply it.