# AI capex — X 热门讨论 (2026-09-26 13:30 UTC)

## @_JaxCapital (Jack Hurley) · 09-26 00:36 · ♥108 ↻4 💬0 My takeaways from $PL upcoming launches and cadence:

PL incremental read-through: the factory is becoming the product. Planet just sent Pelican-12 to Cape Canaveral less than 3 months after Pelican-11 launched. Same week they opened Berlin. Easy takeaway: this is no longer “another imagery company launching cubesats but rather manufacturing AT SCALE high resolution and AI enabled satellites quicker than peers.

1. Capacity
Berlin is live. First next gen Pelican targeted before year-end. Stated scale: up to 60 Pelicans a year, roughly doubling production while SF keeps running. That capacity is for Satellite Services, not just more PlanetScope.

Pipeline: $4B+ identified, ~$1B near-term. Backlog ~$815M. Cash ~$865M. FY27 capex raised to $100–$115M to pull this forward.

2. Cadence
The number that matters is contract-to-orbit: • Sweden: ~4 months from signature to launch, then handover in the quarter • Germany: dedicated capacity in ~2 months • Pelican-12: design learnings from P11 already on a pad in under 3 months They are also splitting coasts. Vandenberg stays SSO volume (Tanager-2 + 18 SuperDoves, Oct 1). Cape gets the first Pelican onto a mid-inclination Bandwagon flight. That is a product move for mid-latitude revisit, i.e. the customers who pay for tasking.

3. Quality
Launch quality looks fine: contact, first light, birds marked active. Pelican-2 commissioned in days. The real quality test is next. P11 was a Gen 2 demo and not a commercial data bird. P12 is the updated 30 cm-class pathfinder, plus edge compute and intersat comms, on a new orbit class. If that works, pricing power. If it slips, capex and margin noise. Watch gross margin. Q2 was 59%. FY guide is 55–57% because services mix can dilute the software-like data margins even as revenue scales.

Bottom Line- $PL is trying to industrialize a 30 cm bus fast enough to sell sovereign constellations. The last 90 days say the factory is starting to keep up with the pipeline.

Next proof: Berlin ships a finished bird this year, and Gen 2 actually sells at a higher price.

I’m constructive under $20, but to be honest, I would require as low as a price as possible given my cost basis is <5$ for me to bite for more. That said, I’m active in options just to have an opinion on overly pessimistic days.

Planeteers Unite! $PL 🚀 https://x.com/_JaxCapital/status/2103644957433053243

## @danielnewmanUV (Daniel Newman) · 09-26 12:37 · ♥55 ↻1 💬4 We see AI Capex at $12-14 Trillion cumulatively between now and 2030. By 2032 it could be $10 Trillion per year. Many still don’t realize how big this investment cycle is. 👀 > 引用 @Kalshi: BREAKING: AI data-center spend is projected to hit $10.3 trillion by 2032, per WSJ https://x.com/danielnewmanUV/status/2103826376658968995

## @HakiBTC (Haki) · 09-26 04:18 · ♥33 ↻0 💬23 I'll start by checking the latest posts from @vangrid_io so the article matches current news, then pick a fresh topic angle.Latest posts lean toward agents hiring humans for real-world capture. I’ll pull the bounty and x402 details so the article stays accurate.Latest docs (Sep 24–25) highlight agent-posted bounties paid in USDC on Base via x402 that is today’s angle. Article first, then the illustration.

Physical AI can plan in simulation. It still fails on a loading dock that changed this morning. That is why @vangrid_io exists: a spatial data layer and ground-truth rail for Physical AI. Contributors record short videos on ordinary phones. Those clips are reconstructed into 3D models, each fingerprinted at capture content hash, coarse location, timestamp then batched and anchored on Base so a buyer can check provenance on a public ledger instead of trusting a vendor.

The new loop is software hiring the street. An agent posts a bounty with one HTTP call, pays the reward in USDC on Base through x402, and a person walks to the aisle, stairwell, or warehouse bay and films it. No account. No API key. The caller needs a wallet. Escrow locks before work starts. Acceptance releases payment onchain and returns the reconstruction through the same API.

That is not a mapping product with a chatbot wrapper. It is an agent-native market for places satellites never see and fleets never drive. Satellites see roofs. Camera cars see roads they can legally occupy. Physical agents need the interior, the alley, the temporary obstacle. Three billion phones already sit in those places. Vangrid turns them into a zero-capex sensor swarm that software can commission in minutes.

For robotics labs, logistics stacks, and world-model teams, the scarce input is no longer compute. It is verified, ground-level observation with a settlement path machines can complete alone. Agents that can pay for reality will outrun agents that only read it. https://x.com/HakiBTC/status/2103700590543171686