Every leftover from the last orbital sentence has a different object. Google Research named two TPU satellites. Planet is the bus. Early 2027 is the year. Starcloud filed 88,000. Overview sold Meta a night-time gigawatt with a 2028 demo. Flatten those into one space trade and you get a false headline. Do not flatten them.
Suncatcher prototypes, Google Research blog, Planet as the bus partner, early 2027. Not an FCC constellation.
Two is how you find out whether a TPU dies of heat, upsets, or boredom.
A research blog that named a year
The dates are already on paper. Early 2027 on two Suncatcher birds. 30 June 2027 on a public note that an optical link carried a real TPU workload. 2028 on an Overview in-space demo. 2030 on a commercial beam. A date is not a constellation.
Planet is the exposed bus
The letter the note actually produced is a research blog, not an FCC integer. Google Research's Suncatcher paper names two prototype satellites, Tensor Processing Units, a dawn-dusk sun-synchronous orbit, optical links, and Planet as the partner, for early 2027. Planet flies buses for a living. Alphabet is. Two birds can fail in 2027. An 88,000-satellite file cannot fail until the 2030s. That is why files get written.
Dawn-dusk sun-synchronous orbit is the honest orbit for a solar-powered computer. The array sees sun almost every minute. The radiator sees cold sky. You do not get that for free in a GEO power-beaming story or in a mid-inclination comms plane. You also do not get a low-latency path to a user in Lagos. It is an attempt to put training or batch inference where the photons already are. The downlink is the product everyone will try not to talk about. A model that trains in sun-synchronous orbit still has to come home, or the job was a stunt.
Starcloud is an H100 in a small bird, then an 8-kilowatt class, then a Starship-class 200-kilowatt node, then an FCC number. SpaceX is telling the Commission about a million compute satellites and telling earnings about a Vera Rubin NVL72 in orbit. Overview wants to beam a gigawatt at Meta farms, demo 2028, commercial 2030. Suncatcher is two birds. The small dated object is the one that can fail this decade. The million-satellite file is a policy option on Starship and on a public that has not yet seen the night sky fill with radiators.
X treated Suncatcher as confirmation of Musk. It is the opposite shape. Google did not file 88,000. Google hired a bus company and named a year. That is how a laboratory spends money when it wants an answer, not a valuation.
Two Suncatcher prototypes on orbit by 30 June 2027, with a public note that the optical link carried a real TPU workload, would confirm the narrow claim. A slip past 2027 without a named hardware cause would weaken it. A Google notice that the next step is 80 satellites, with launch deposits, would be a different piece. Do not promote a research blog into a constellation. Do not ignore it because it is only two. Two is how you find out whether a TPU dies of heat, single-event upsets, or boredom.
What a research note cannot do
Cross-asset does not care which bird talks to its twin. AI abundance cares only if the middle's hosting cost becomes an orbital cost. It has not, on the invoices that page already named.
A refutation would look like this: two named Suncatcher prototypes on orbit with a public note that the optical link carried a real TPU workload by 30 June 2027; or a named slip past 2027 with no hardware cause; or a Google notice that the next step is 80 satellites with launch deposits. A blog post is none of those until the birds fly. An 88,000-satellite file is none of those.
The honest uncertainty is which object gets a number first. If the two birds fly and pass data over the optical link, Google has shown that a TPU can live in sun-synchronous orbit and talk to its twin. If they slip, the blog was a blog. If Starcloud's 88,000-satellite file grows and the two birds are still on a bench, the file moved and the prototype did not. Those two sentences can live in the same year. Two birds and a year are a fact about one laboratory. An 88,000-satellite integer is a fact about one filing. Neither is a constellation that has already trained a model.
Satellite bus on the cleanroom floor.
Wafer and a radiator plate.
Launch-slot tags on an aluminum bus.