Space · Tuesday, 25 August 2026

01 · Briefing · what happened

Half a billion dollars went into data centres in orbit. The rides are sold out to 2029.

Space 3 min 23 sources

Two space startups raised $250m each in one week to put AI computers in orbit, with Nvidia and Cisco joining in. At the same time SpaceX confirmed it will retire the Falcon 9, and its shared-ride slots are gone past 2028 - so buyers have started purchasing whole rockets and building their own engines.

$500m

raised in one week

$250m each for Starcloud and Muon Space, two orbital data centre startups [1][3]

2029

next free shared ride

Falcon 9 shared-ride slots are gone until late 2028 or early 2029 [1]

1,000

US launches a year by 2030

the new White House goal, against just under 200 in 2025 [10]

11,000+

Starlink satellites in orbit

passed in August; astronomers are studying plans for 1.7 million more [8][12]

At a glance

Starcloud raised a $250m extension to its funding round on 21 August, doubling its value to $2.3bn and taking the total raised to $450m [1][2].
Nvidia and Cisco both joined the round - the first time makers of ground computing gear have backed orbital computing at this size [1][2].
A second orbital data centre startup, Muon Space, announced a $250m round in the same week [3].
SpaceX will retire the Falcon 9 once Starship flies reliably, Elon Musk confirmed on 22 August, shifting factory capacity to the bigger rocket [5].
Falcon 9 shared-ride bookings are already unavailable beyond late 2028, and industry reports expect most Falcon missions to stop from 2028 [1][5].
Buyers have stopped queueing: Portal Space Systems bought an entire Falcon 9 for a 2028 mission and is reselling the spare half of the nose cone [9].
Others are building their own way up - Cowboy Space raised $275m to build its own rocket, Japan's Letara raised 2.6bn yen to scale a thruster into a launch engine, and CesiumAstro bought a chip maker to supply its planned 737-satellite fleet [1][16][22].
Rival supply is coming but not yet: Kepler switched on a commercial laser relay network and booked Rocket Lab's Neutron for 2028, and China's Landspace wants to refly its landed booster within six months [15][23].

Forces in play

Money for orbital computing High↑

Half a billion dollars in one week, and Nvidia is designing a processor specifically for space [1][2][3].

Rides to orbit High↑

SpaceX has stopped selling shared rides past 2028 while it moves its factory onto Starship [1][5].

Crowding overhead Building↑

About 15,000 satellites are up now, 11,000 of them Starlink; filings propose 1.7 million more, which astronomers say would badly damage telescope views [8][12].

Cost of a trip up Easing↓

One Falcon 9 booster flew for the 37th time on 25 August, and China's Landspace wants to refly its landed booster within six months [6][23].

In play Starcloud — raised $250m more to build AI computers in orbit; has asked to fly 88,000 of them Nvidia — invested, and is designing its first processor built for space using Starcloud's orbital test data SpaceX — will retire the Falcon 9 for Starship, which is what tightened everyone else's rides Portal Space Systems — bought a whole Falcon 9 rather than wait in the queue The White House — set a goal of 1,000 US launches and re-entries a year by 2030

How it unfolded

Nov 2025 Starcloud's 60 kg first satellite runs an Nvidia H100 chip in orbit - the first to do so [1]
Aug 20 The White House publishes a launch policy aiming at 1,000 flights a year by 2030 [10]
Aug 21 Starcloud's $250m extension lands, with Nvidia and Cisco joining [1][2]
Aug 22 Musk says winding down the Falcon 9 and Falcon Heavy is inevitable [5]
Aug 25 The 100th Falcon 9 flight of 2026 sets a reuse record on its 37th trip [6]
Jan 2027 Starcloud-2, about 100 times more powerful than its first satellite, is due to fly [1]

Where this points

Watch whether Starship reaches orbit on its 14th test flight, expected in September - it is the one event that decides whether the ride shortage is a two-year gap or a decade-long one [13][14].

Full briefing

Why the money moved this week

Nvidia’s name on the round is the tell. Starcloud is the only company known to have run one of Nvidia’s big data-centre chips in orbit. Its chief executive says the chipmaker checked the engineering far harder than any other investor before paying [1][2]. Nvidia is now designing a processor meant for space, and the readings from that orbiting chip are what the design rests on [1][2]. The investment and the chip are one project, not two.

The squeeze is one company’s factory decision

None of this is urgent if there are rides. SpaceX is winding down the Falcon 9 so it can move people and tooling onto Starship, which Elon Musk called inevitable in a post on 22 August [5]. That decision lands on everyone, because the Falcon 9 flew 165 missions last year - more than every other orbital rocket in the world put together [6]. Bookings for shared rides - where several firms split one rocket - are already gone beyond late 2028 [1].

Two counts of the same hundred

Two outlets reported a hundredth flight this month, counting different things. Spaceflight Now called a 21 August launch SpaceX’s 100th orbital mission of 2026 [7]. Space.com called the 25 August launch the 100th Falcon 9 flight of the year [6]. One count folds in Starship tests; the other does not. The tracked figure below uses the Falcon-only count.

The part with no number yet

The waste has no precedent. Ars Technica’s estimate is deliberately conservative. Strip a data-centre chip to bare silicon and ignore the cooling hardware. SpaceX’s proposed million-satellite fleet would still push roughly 1,000 tons of copper and 170 kg of gold out of Earth’s material cycle every year [4]. That is about 1% of the world’s yearly palladium and thallium [4]. Around 40,000 satellites a year would burn up in the air; the rest would be shoved outward to a distant disposal orbit [4]. Neither path has been costed or ruled on.

What the crunch is already made of

The bottleneck is not only rockets. A report from PwC and the Aerospace Industries Association finds shortages in the unglamorous middle [11]. The laser links satellites use to talk to each other. Fuel tanks. Connectors built to survive radiation. In some of those, the whole United States has as few as three suppliers [11]. Aerospace factory output has grown 30% in five years, and the plants now run at about three-quarters of capacity [11]. Space bids for the same chips and skilled hands as AI, defence and power [11]. That is why a queue for rockets becomes a queue for everything else.

Also this week

Two astronauts go back outside the space station today to fit the high-speed aerial they ran out of time to install last Tuesday [18][19]. China’s Chang’e-7 Moon mission was stopped by a typhoon and will not fly in its planned window this year [20]. Astronomers dated the Milky Way’s earliest known meal - a dwarf galaxy swallowed 11.8 billion years ago [21]. And 17 spacecraft, none built as a team, together tracked one solar eruption and found it had two lobes moving at different speeds [17].

02 · Lesson · why it matters

Why a shortage builds the glut that follows it

When one supply tightens, every buyer quietly starts building its own - separately, on the same morning - and all of it arrives years after the squeeze has passed.

How it works

One supplier carries most of the traffic
It turns its factory to a new product
Its old rides stop being sold
Every buyer separately builds its own way up
None of them can see the others doing it
All the new supply lands years later, together

The twist

A shortage does not produce a steady, matching build-up. It produces one lump of new supply that arrives after the shortage has passed - and the scramble is what makes the glut.

Where you've seen this

Computer chips

the 2021 shortage started factories everywhere; the finished plants met a slump

Nursing

training places fill after a staffing crisis and the graduates arrive years later

Housing

builders all start in the boom and the flats complete into a falling market

Cattle and crops

high prices this year become the surplus that crashes the price next

The catch

Sometimes demand really does keep growing and the build-out is exactly right. You cannot tell which from inside it - and whoever waits to find out has no ride at all.

Full lesson

A man buys a whole rocket

Jeff Thornburg runs a small spacecraft company. It needs a ride to orbit in 2028. Rather than book a share of one, he bought the entire rocket. He is now selling off the half of the nose cone he does not need.

Nobody wants to buy launch that way. It is what you do when the queue closes.

The queue closed for an ordinary reason

SpaceX has not run out of rockets. It has pointed its factory at a different one.

The Falcon 9 flew 165 missions last year, more than every other orbital rocket on Earth put together. Starship is meant to take over. The same buildings, tools and people make both, so the new rocket only grows if the old one shrinks. That is why shared-ride bookings stopped years before the last Falcon 9 will fly.

It is a reasonable decision about one company’s floor space. It is also the road to orbit narrowing, for everyone, at once.

Everyone reaches for a fix on the same morning

Watch what the buyers do next.

Portal buys a whole rocket. Cowboy Space, which wanted computers in orbit, decides to build a rocket instead. A small Japanese firm that made satellite thrusters scales the same engine up into a launch engine. A satellite maker buys a chip company so its parts stop being someone else’s problem. Starcloud spends part of its new money on factory space and on buying rides in advance.

Every one of those is a sensible private answer to the same public fact. And every one is decided without knowing what the others are deciding. There is no room where they compare notes. There is a price, saying the same thing to all of them on the same day.

The answers do not arrive when they are needed

Here is the part that catches people.

A shortage feels like a gap you can fill. But nothing on that list can be delivered this year. A rocket takes years to design. A factory takes years to build. The scaled-up engine has to be tested, then flown, then trusted.

So a shortage does not produce a steady trickle of supply that keeps pace with demand. It produces a lump. Everyone starts at the same moment, because they all read the same signal. Everyone finishes at roughly the same moment too. And the lump lands after the squeeze that caused it.

That is why shortages so often end in gluts. The glut is not a separate event. It is the shortage, arriving late.

You have seen this without a rocket in it

Chip factories broke ground during the shortage of 2021 and opened into a slump. Nursing courses fill up after a staffing crisis, and the nurses qualify three years later. Flats are started in a boom and finished in a fall. Farmers breed more cattle when prices are high, then sell into the price their own herd pushed down.

The pattern needs only two ingredients. A shortage everybody can see. And a fix that takes longer than the shortage does.

Being right is not the safe option

There is no clean way out of this.

Wait to find out whether the demand is real, and you may have no ride at all. That is the position the smaller founders are in right now, the ones who cannot buy a rocket outright. Move early, and you may end up owning a factory nobody needs.

Most of the people carrying that risk are not in the story. The money in these funding rounds comes from investment funds, and those funds hold, at some distance, ordinary savings. The satellites they pay for carry weather data, maps, and internet to places with no cable. If too much gets built, the loss is spread thin across people who never heard of Starcloud. If too little gets built, so is the shortage.

Each company sees a scramble for something scarce. What is actually happening is a few hundred separate decisions, made in the same week, by people who cannot see each other. Nobody in the picture has the view of the whole picture - and that includes anyone writing it down afterwards.

03 · Lab · your turn

The Queue and the Lump

Rehearse choosing between waiting, buying and building when a shortage is visible to everyone at once.

04 · Hope · carry this

The scramble that looks wasteful from inside is often how a scarce thing turns into an ordinary one. Every glut in history began as a queue that a lot of people separately refused to accept.