nextbig.dev
Vancouver, B.C. · Intelligence on AI and the machines that run it
nextbig.dev
The Briefing · Sunday, July 26, 2026

Alphabet's $900m SpaceX cheque became a $94.1bn stake, turning a decade-old strategic investment into a balance-sheet business large enough to move the quarter

Alphabet now values its roughly 6% SpaceX holding at $94.1bn, more than 100 times the $900m invested in 2015. About $80bn is under short-term restrictions and $14.1bn remains restricted through the third quarter of 2027, just as Google's first Project Suncatcher satellites are due to test orbital compute.

7 min read
The Rundown No. 157 · Audio Edition · 2 min All episodesRSSMP3
0:00 / 1:46
VTT
The Big Story
Alphabet's $900m SpaceX cheque became a $94.1bn stake, turning a decade-old strategic investment into a balance-sheet business large enough to move the quarter

A $900m cheque written in 2015 now sits on Alphabet's books at $94.1bn. The company disclosed that its SpaceX holding represents roughly 6% of the newly public rocket and satellite operator, making the position worth more than 100 times the original investment. About $80bn is subject to short-term sale restrictions, while $14.1bn remains restricted through the third quarter of 2027. This is no longer venture upside hiding in a footnote. It is almost large enough to equal Alphabet's annual capital budget.

The original logic was strategic. Google wanted faster global internet distribution, SpaceX needed capital for rockets and the satellite network that became Starlink, and both companies could describe the investment as infrastructure. A decade later the asset is large enough that movements in SpaceX can overwhelm ordinary operating gains or losses in a reporting period. Alphabet built a second earnings variable without adding a line of search advertising. An 11% move in the stake creates roughly $10bn on either side of the income statement.

Project Suncatcher makes the relationship current rather than historical. Google plans to launch two prototype satellites in early 2027 to test solar-powered machine-learning compute in orbit. SpaceX owns the launch system and orbital operating experience such an experiment needs. Alphabet therefore holds both a financial claim on the transport layer and an internal programme that may buy from it. The $94.1bn valuation gives that supplier relationship unusual balance-sheet gravity. Successful prototypes would make SpaceX both an asset and a cost of goods sold. They also give Google a technical reason to keep launch access aligned while the share restrictions expire, an alignment with its own option value.

The strongest objection is liquidity. Most of the stake cannot be freely sold today, and SpaceX's public price can move before restrictions lapse. Mark-to-market wealth becomes datacentre funding only when shares become cash. Alphabet has ample cash elsewhere, so the holding's nearer value is optionality: the company can hold, sell in phases, exchange shares around a transaction or keep strategic alignment without another capital contribution. Restriction dates determine when those choices become real, and $14.1bn stays locked until the same quarter in which Suncatcher should have flight data.

Builders should read the disclosure as an ownership lesson. A platform dependency bought early can compound into an asset that partly offsets the cost of using the platform later. Most companies cannot take a 6% supplier stake, but they can negotiate warrants, capacity rights or usage credits when a critical provider is young. Alphabet's return came from recognizing that global connectivity was more than a service bill in 2015. By the third quarter of 2027, it will have a liquid financial asset and the results of two orbital-compute prototypes on the same ledger. Those dates now matter more than the original cheque.

Open Hardware Changes the Denominator

A portable MRI built for under $70,000 attacks the machine before the scan

The open-source OSI2 ONE portable MRI was built for less than $70,000, below 7% of the roughly $1.1m starting price cited for a conventional full-size system. Its 3D-printed core produces a 50-millitesla field rather than the 1.5 to 3 tesla common in hospitals, then leans on computational reconstruction to recover usable images. That trade shifts cost from precision hardware into software and time, which can matter enormously where the alternative is no scanner. It does not erase regulation, calibration or diagnostic accuracy. The opportunity is a new floor for research and triage, not an instant replacement for a clinical machine whose stronger magnet collects more signal.

Peak Energy gives sodium-ion storage a US factory and a 20-year claim

Peak Energy plans a 17,000-square-metre US factory with capacity for 4GWh of sodium-ion storage a year, backed by $71m and an initial California deployment in 2027. The company claims its system can last 20 years or 20,000 cycles while retaining 80% capacity, operate at 96% efficiency and cut lifetime cost by 20%. Sodium avoids lithium's supply chain and fire profile, but the commercial gap remains visible: less than 1% of US storage deployment is expected to use it this year, and Peak currently depends on Chinese cells. A domestic plant can change procurement before it changes chemistry; bankable field data is the step that turns those claims into project finance.

Control Stays on the Device

Apple is making privacy the product requirement for glasses that can always see

Apple is reportedly preparing to reveal smart glasses at WWDC next June ahead of a late-2027 launch, with privacy positioned as a primary difference from rival devices. The category forces the issue because a camera and microphone worn all day create data before a user consciously opens an app. On-device processing, visible capture signals and narrow retention policies are therefore product architecture, not settings-page copy. Apple has the silicon and distribution to make local inference credible, but the claim will be tested by what still leaves the device. A privacy advantage exists only when developers can identify which frames, transcripts and embeddings never cross the network.

GrapheneOS documents how a locked phone can keep extraction tools outside

GrapheneOS published a detailed account of the protections it applies against data extraction from locked Android devices, treating post-reboot and ordinary locked states as two distinct security boundaries rather than one binary condition. The work matters because mobile forensics vendors attack implementation seams: available keys, peripheral access, background services and the time before credentials are required again. A hardened device reduces that surface by minimizing what is running and what secrets remain available while locked. The transferable lesson for agent products is precise. Sensitive state should expire into a smaller privilege domain when nobody is actively authorizing work, even if the process itself remains online.

Quick Hits
The Takeaway

Alphabet's SpaceX disclosure, a sub-$70,000 open MRI and Peak Energy's planned sodium-ion factory all describe the same financial move: shift a hard dependency from a recurring bill into something you partially own or can reproduce. Alphabet bought equity in its connectivity supplier. The MRI team published the machine. Peak is bringing cell and system capacity closer to the customer. Apple and GrapheneOS show the security version, where the valuable dependency is control of data and keys on the device. Ownership does not remove execution risk. Alphabet's shares are restricted, a 50-millitesla scanner collects less signal than a clinical magnet, and Peak still needs 20-year claims to survive real projects. It changes who captures the upside and who has an alternative when a supplier tightens terms. The practical question for a builder is which dependency is large enough, early enough and strategic enough to deserve equity, an open implementation or a local fallback instead of another invoice.

The Call C-20260726

By November 15, 2026, Alphabet will report a pre-tax quarterly gain or loss of at least $10bn tied to its SpaceX holding, making the decade-old supplier stake a material driver of reported earnings rather than a venture footnote.

The case

A $94.1bn public equity position needs only an 11% price move to create a $10bn mark. SpaceX's newly public shares lack a long trading history, while most of Alphabet's holding remains restricted and cannot be sold to damp the exposure. Search and cloud can perform normally while this one asset moves the income statement by a figure comparable with a major operating segment. The accounting consequence should arrive before the strategic relationship changes.

What proves us wrong

If Alphabet's next quarterly filing shows less than a $10bn pre-tax SpaceX-related gain or loss, or shows that the holding is not remeasured through reported earnings, the call is wrong.

Settles by November 15, 2026
The Tape T-20260726
◆ Watch GOOGL Alphabet medium conviction

We sharpen the Alphabet watch from July 24. The company now has two infrastructure exposures moving in opposite directions: leased datacentre capacity can pressure cash flow, while a $94.1bn SpaceX stake can create enormous non-operating gains. The stake does not solve a power queue, but it gives Alphabet optionality around a launch and connectivity supplier that Project Suncatcher may need in 2027.

The SpaceX position equals roughly 6% of the supplier and more than 100 times Alphabet's 2015 investment. Its restrictions limit immediate liquidity, while the size makes quarterly remeasurement a material source of volatility.

Wrong if A filing that keeps SpaceX movements immaterial, or a sharp decline before restrictions lapse without any strategic benefit to Suncatcher, would remove the balance-sheet cushion from the thesis. Settles 9 months
◆ Watch AAPL Apple medium conviction

We hold the Apple watch opened July 14. Smart glasses make the device-layer thesis testable: privacy can differentiate an always-on camera only when inference, retention and consent are visibly enforced on the hardware. Apple's custom silicon makes that architecture possible. The risk is that useful assistant features still require cloud context, leaving the company with a marketing claim that developers cannot verify.

A reported 2027 launch gives Apple time to make local processing and capture controls part of the platform contract. Rivals already define the category, so privacy must change data flow rather than merely brand it.

Wrong if Developer documentation showing routine visual or audio streams sent to cloud services by default, without granular retention controls, would break the claimed device-level advantage. Settles 9 months
Desk signals from the day's verified wire — falsifiable, dated, settled in public. Analysis, not individualized investment advice.

Get this briefing in your inbox

What changed in AI and compute, what it costs, and what to build. One email per week. No spam, unsubscribe anytime.