A Rocket Shortage, AI Food Orders and a Chip That Uses Light

A launch shortage, few AI food orders, a light-based data chip, and plain meanings for common AI words.

Four tech stories stood out this week. A new report says the world outside China is short on rockets. Very few people order food through AI helpers yet, a new chip moves data with light, and AI news keeps using words that need a plain explanation. You do not need to do anything today, but each story is worth watching.

A shortage of rockets

A report from the research firm Quilty Space, covered by Ars Technica, says the world outside China faces a big shortage of rockets over the next five years. Satellites need rockets to reach space. Right now, more satellite owners want launches than there are rockets to carry them.

The firm estimates a gap of about 150 launches between now and 2030. If US military needs are added, and SpaceX does not extend its Falcon 9 and Falcon Heavy program, the gap could double. The report expects the crunch to be worst in 2028, then slowly get better.

The report lists several reasons:

• Russian rockets left the Western market because of the war in Ukraine.
• Amazon's low Earth orbit satellite project needs many launches.
• New rockets, such as New Glenn, Vulcan, Neutron, Terran R and Nova, have had delays.
• Falcon 9 is being retired as a rocket for commercial service.

Caleb Henry, research director at Quilty Space, put it simply: "If you haven't booked your launch already, and you need one, you're late."

What to watch: SpaceX's Starship. It carries very large loads and could be reused quickly. But the report says SpaceX "has also stopped affirming Starship as a commercial launcher." Its current door is narrow and built for SpaceX's own Starlink satellites. The head of Telesat, a satellite company, said his company's future satellites may need to fit that same dispenser. Henry also pointed to how often Blue Origin's New Glenn can launch.

Few people order food through AI helpers yet

An AI agent is software that does a task for you, like placing an order. DoorDash CEO Tony Xu said outside AI agents have made "several hundred orders" on DoorDash since they started, Ars Technica reports. In the third quarter of 2026, DoorDash handled just under 1 billion orders. Xu called agent orders a "very low use case."

It is still early. Less than a month ago, DoorDash released a connector that gives outside AI tools a standard way to order. Only a small number of early test partners can use it so far, mostly for company lunch orders.

These numbers do not include DoorDash's own AI tool, called Ask DoorDash. The company said carts built with it come together five times faster. It also said those orders average nearly 50% higher value, and nearly half go to local places the customer had never tried.

Why it matters: Xu thinks agents may help most with big decisions that take time. For simple tasks like food, he said, people tend to go straight to the app. Ars compared this to Pizza Hut's first online orders in 1994. Today, more than half of Pizza Hut orders are placed online.

A chip that moves data with light

PsiQuantum, a quantum computing company, built a switch that sends data as light in AI data centers, Tom's Hardware reports. A data center is a building full of computers. AI runs on chips called GPUs, and moving data between them is becoming a slow point.

Regular switches turn light into electric signals at each step. That uses power and adds delay. This kind of switch passes light straight from one glass fiber to the next.

Every switch loses some light. The loss is measured in decibels, or dB. Typical data center parts can handle losing half the light. In lab tests, PsiQuantum's switch typically lost about a quarter of the light. It did this without an amplifier, a part that boosts the signal, and with no moving parts.

What to watch: size. The test chip has eight inputs and eight outputs. Other switches handle 32 to 300, and Google's has 136 ports. PsiQuantum says larger devices are in development.

Words you will see in AI news

AI news uses many new terms. A helpful guide from Tom's Hardware explains several. Here are the basics in plain words:

• Model: the AI program itself. It learns from huge amounts of text from the internet, books and transcripts.
• Tokens: small pieces of text, like words, parts of words, code or punctuation.
• Weights: numbers that show how strongly tokens connect. They only change if the model is trained again.
• "8B" or "20B": how many billions of weights a model has. More is not always smarter.
• Open-weight: models whose weights are shared free online, such as Llama, DeepSeek and Kimi. Downloading one does not show how it was trained.
• FP16, INT8, FP4: sizes of numbers a program uses. Memory is limited, so developers try to use the smallest size that works.

Knowing these words makes AI news much easier to follow.

Have a question about anything here? Email me at sales@appwt.com or call (888) 565-0171, and I will answer it.

Sources

• Analysts find looming launch crisis is "more severe" than anticipated, Ars Technica, 2026-10-30
• Basically no one is using AI agents for their DoorDash orders, Ars Technica, 2026-10-30
• PsiQuantum turns its photonics know-how into an optical switch for supercharging AI data center GPU networks, Tom's Hardware, 2026-10-30
• A primer on key AI LLM concepts: KV cache, weights, quantization, and how inference actually works, Tom's Hardware, 2026-10-30