Level 1 — Absolute Beginner

Google and SpaceX are talking about a big idea. They want to put computers in space.

Computers need a lot of power and water on Earth. In space, the Sun gives a lot of power. There is no problem with hot weather.

SpaceX can send big rockets to space. Google makes very fast computers for AI.

If the plan works, AI will be smarter and faster. The first test could happen in two or three years.

Google

a very big company that runs a famous search engine

SpaceX

a company that builds rockets

space

the area above the sky where stars are

rocket

a big machine that flies into space

computer

a machine that can do many tasks with numbers and words

AI

a computer that can think a little like a person

power

energy that machines need to work

Sun

the bright star that gives us light and heat

Level 2 — Elementary

Google and SpaceX may build the world's first data centers in space. News reports on May 12, 2026 said the two companies are talking about how to do this. A data center is a large building full of computers that run websites and AI.

Today, almost all data centers are on the ground. They use a lot of electricity and water to keep the computers cool. Some cities now have problems giving them enough power. There is also worry about water in dry areas.

In space, the Sun shines almost all the time. Solar panels can give the computers free power. There is also no humid weather to make the machines too hot.

SpaceX would use its big new rocket, Starship, to take the equipment to orbit. Google would build special computer chips that can work in space. If the plan succeeds, the first small test satellites could fly within two or three years.

orbit

the path of a satellite going around the Earth

satellite

a machine that travels around the Earth in space

solar panel

a flat board that turns sunlight into electricity

electricity

the energy used by lights, computers and machines

Starship

the big new rocket built by SpaceX

humid

having a lot of water in the air

chip

a tiny piece of equipment that does the work inside a computer

test

an early try to see if something works

Level 3 — Intermediate

Google and SpaceX are in advanced discussions over a joint plan to launch artificial-intelligence data centers into Earth orbit, according to a Wall Street Journal report published on May 12, 2026, and quickly confirmed by TechCrunch. The proposal would marry SpaceX's Starship heavy-lift rockets with Google's Tensor Processing Units, the custom silicon that already powers much of Google Cloud's AI workloads.

The strategic logic is straightforward. Terrestrial AI data centers are increasingly bumping against three hard limits: gridded electricity, fresh-water cooling, and local zoning rules. In Northern Virginia, the world's densest concentration of cloud capacity, Dominion Energy has warned of multi-year delays in delivering new interconnections. Ireland and the Netherlands have begun rejecting new permits. By contrast, low Earth orbit offers nearly continuous solar illumination, radiative cooling into the vacuum of space, and zero land use.

Engineering hurdles remain serious. Computer chips in orbit must be hardened against cosmic radiation, which can flip bits and corrupt computations. Bandwidth between satellites and the ground has to scale dramatically, almost certainly using laser optical links rather than radio. And the cost per kilogram to orbit, although falling fast thanks to Starship, must drop further before space compute can rival the marginal cost of a hyperscale terrestrial data center.

The talks are reportedly part of SpaceX's preparations for a planned 2026 initial public offering rumored to value the company at $1.75 trillion. Positioning Starship as the lowest-cost path to orbital AI infrastructure would broaden SpaceX's revenue base beyond Starlink and government launch contracts. For Google, securing early access to orbital compute capacity could become a strategic differentiator in an industry where Amazon, Microsoft and Meta are all bidding aggressively for terrestrial power supply.

Starship

SpaceX's giant fully reusable rocket designed to carry large payloads to orbit and beyond

TPU

Tensor Processing Unit; Google's custom chip designed for artificial-intelligence work

low Earth orbit

the region of space relatively close to Earth, typically up to 2,000 km altitude

cosmic radiation

high-energy particles that travel through space and can damage electronics

laser optical link

a way to send data using a laser beam instead of radio waves

interconnection

the connection that lets a data center get power from the electric grid

hyperscale

very large in scale, especially in cloud computing

marginal cost

the extra cost of producing one more unit

Level 4 — Advanced

The reported negotiations between Google and SpaceX over orbital artificial-intelligence data centers, first surfaced by the Wall Street Journal on May 12, 2026, mark the most ambitious attempt to date to relocate hyperscale compute beyond the planet's surface. The architecture, as outlined in subsequent reporting by TechCrunch and Tom's Hardware, would pair successive Starship cargo flights with Google-designed Tensor Processing Unit clusters housed inside radiation-hardened bus structures, networked over inter-satellite laser optical links and rendered cheaply addressable from Earth via a meshed downlink architecture. The objective is to bypass the increasingly binding terrestrial constraints — interconnection queues, water availability, and zoning resistance — that now dominate site selection for any new gigawatt-scale campus.

Economically, the calculus hinges on three converging trendlines. First, Starship's marginal launch cost is widely projected to fall below $200 per kilogram by 2028 once full and rapid reusability is demonstrated. Second, photovoltaic conversion efficiencies for space-grade panels have crossed 35% in flight, while spacecraft radiator technologies adapted from Hubble follow-on missions can dissipate hundreds of kilowatts in low Earth orbit without active coolant. Third, frontier-model inference is becoming dramatically more bandwidth-tolerant of latency on the order of tens of milliseconds, which is well within reach of optical mesh downlinks via Starlink-derived intersat networks.

The technical obstacles, however, are not trivial. Single-event upsets from galactic cosmic rays and solar particle events require either expensive rad-hard silicon or aggressive redundancy that erodes the orbital cost advantage. Thermal cycling between Earth's shadow and full solar illumination places severe fatigue loads on substrates and solder joints. And debris flux at altitudes above 600 km — the most attractive band for thermal and radiation reasons — has grown rapidly since the 2024 Russian ASAT test, forcing every operator to budget for collision avoidance fuel and on-orbit servicing.

Strategically, the deal would advance two distinct corporate agendas. For SpaceX, monetizing Starship as the canonical compute-to-orbit pipeline diversifies revenue beyond Starlink and government missions ahead of a planned IPO that bankers have valued at roughly $1.75 trillion, with confidential S-1 filings reportedly already at the SEC. For Google, an orbital footprint addresses a vulnerability that has quietly worried its cloud leadership for two years: Amazon Web Services and Microsoft Azure are bidding aggressively for the limited supply of terrestrial power, and any partner with privileged access to gigawatts of solar-illuminated orbital capacity could regain category leadership in AI compute. Whether the partnership produces a flagship orbital cluster by the end of the decade or simply a flagship marketing announcement, the talks reflect a broader industry consensus that the cheapest electron for AI may very soon arrive above the atmosphere rather than below it.