The uplink was always the bottleneck.

On Monday 28 September, at 7:48 in the morning Central Time, Starship lifted off from Starbase in Texas on its fourteenth flight and, for the first time, reached orbit. It lost one of 33 booster engines shortly after liftoff and one of its three vacuum engines on the way up, relit a sea-level Raptor to insert itself into orbit anyway, and released 26 Starlink V3 satellites within about half an hour. The flight team cut the planned ten hours short to a little over three, then brought it down in the north Pacific.

The rocket gets the headlines. The detail we keep coming back to is in the payload. SpaceX says each V3 satellite carries 10 times the downlink of the V2 generation and 22 times the uplink, around 160 gigabits a second of it. Downlink got a big number. Uplink got a much bigger one.

Most networks are built for watching

Almost every connection a business can buy is lopsided. Home broadband, venue Wi-Fi, a mobile data plan and, until now, satellite: all of them assume you mostly receive. That is the right assumption for people streaming films. It is the wrong one for most of what we install.

A brand installation at an event sends clips and photos to a gallery and a social wall. A check-in desk in a field syncs attendance back to the office. A camera on a remote site pushes footage to be reviewed. A fleet of kiosks ships logs so we can see a fault before the client rings. None of those systems care much how fast they can download. Every one of them lives or dies on how fast it can send.

"

We rarely lose an install to a slow download. We lose it to a queue of uploads that never drains.

What the constraint has done to our designs

Because the upload is so often the narrow pipe, a lot of our engineering exists purely to work around it. We compress on the device before anything leaves. We send a thumbnail first and the full file later. We batch, queue and retry, and we design the screen so a guest never waits for the network to finish. On the busiest sites we have carried footage back on a drive at the end of the day because it was quicker than the connection.

That work does not go away because a constellation gets better. Coverage will arrive gradually, SpaceX says V3 service starts within weeks rather than days, and the satellites share their capacity across everyone underneath them. A crowded event is still a crowded event. What changes is the default. If sending becomes cheap in places that used to have nothing, some decisions we make by habit deserve to be made again.

What we would do now

• 01Ask for upload figures in every site survey, at the busiest hour, not the download number on the venue's welcome sheet.
• 02Keep the offline queue. Better backhaul makes the queue drain faster, it does not make it unnecessary.
• 03Separate what must be sent now from what can be sent tonight. A registration is urgent. A 4K clip usually is not.
• 04Price in satellite as a real option for remote and temporary sites, measured on the day, not assumed from a brochure.
• 05Revisit anything we currently process on site only because we could not send it. Some of that work may be cheaper and easier to maintain centrally.

The honest caveat

Twenty-six satellites is a start, not a network. Starship still has to fly often, land reliably and carry full loads before the capacity shows up where our clients are. The figures are SpaceX's own, and real throughput on a given site will depend on how many other people share the same patch of sky. We will believe a number when we have measured it at an install.

Still, the ratio is the interesting part. For as long as we have been building connected systems, the network has been designed for an audience. This is one of the first signs of capacity designed for people who produce. That is what an installation, an event or a field team actually does all day.