1. The console jockey and the job

William Gibson’s Neuromancer opens with Case jacked into a deck, doing a very specific kind of work for a very specific kind of customer. The deck is not the point. The point is the job. A compute attachment on the desktop is the same: the hardware is only as interesting as the workload it runs.

Entry 218 defined a minimum viable compute attachment and left three questions open. This entry picks the workload question and tries to answer it.

2. Three candidate workloads

A first compute attachment could run at least three kinds of jobs.

Earth observation processing. Sensors on the desktop or on nearby spacecraft produce raw data. The compute attachment turns it into useful products: cloud-free mosaics, change detection, feature extraction. The customer gets answers instead of raw pixels.

Communications routing. The desktop receives data from multiple sources — ground stations, other satellites, customer payloads — and routes it to where it needs to go. The compute attachment runs the switching, scheduling, and protocol conversion.

Manufacturing control. If the desktop hosts an additive manufacturing or assembly attachment, the compute attachment runs the toolpaths, monitors the process, and logs quality data.

All three are plausible. All three cannot be the first workload.

3. The selection rule

The first workload should satisfy four tests:

  • It must be technically achievable with commercial components and software-level fault tolerance.
  • It must have customers who already pay for the same service on the ground or on existing spacecraft.
  • It must not depend on other unproven attachments.
  • It must produce a measurable result that validates the desktop.

4. Earth observation processing wins the first slot

Earth observation processing passes all four tests.

Technically, it is a solved problem at the algorithm level. Containers and frameworks for image processing are mature. The challenge is running them in orbit with limited power and tolerance to radiation errors, not inventing the algorithms.

Customers already pay for imagery and derived products. Government agencies, agriculture, insurance, and environmental monitoring all buy this. The market exists.

It does not depend on manufacturing attachments or complex robotic servicing. It needs sensors, which can be a separate attachment or a customer payload, and compute.

The result is measurable: reduction in downlink bandwidth, faster delivery of products, or new products that cannot be made on the ground because the data never leaves orbit.

5. Communications routing as the close second

Communications routing is also strong. It is technically simpler in some ways and has clear customers. But it tends to become a commodity service offered by relay constellations. The desktop’s advantage is not just routing; it is co-locating routing with processing and storage. Earth observation processing makes that advantage visible first.

6. Manufacturing control waits its turn

Manufacturing control is the most futuristic of the three. It depends on having a manufacturing attachment, which is not the first thing to build. It remains on the map, but not in the first slot.

What this changes

  • The first compute attachment is for Earth observation processing, not general computing, communications routing, or manufacturing control.
  • This choice simplifies the power and storage requirements because the workload is known.
  • The next entry can narrow the power budget and board count using this workload as the reference mission.
  • Communications routing and manufacturing control stay on the roadmap as later capabilities.