# AI capex — X 热门讨论 (2026-10-11 08:32 UTC)

## @Iam_T_HUY (H U Y) · 10-11 06:07 · ♥53 ↻0 💬62 Vangrid is turning phones into the eyes of Physical AI.

Satellites only see roofs. Mapping cars stick to roads. Neither gets inside warehouses, stairwells, markets, or loading bays.

@vangrid_io fixes that.

Anyone can record short multi-view videos with a normal phone. Faces and plates blur on-device. Each capture is fingerprinted and anchored on Base so buyers can verify it instead of trusting a company.

Those videos become clean 3D models delivered through an API.

AI agents can even post a bounty, pay in USDC, and get the reconstruction back—no accounts or keys needed.

100K+ captures already. $9M raised. Ordinary phones as a zero-CAPEX sensor swarm for robots, logistics, and world models.

Physical AI needs ground truth. Vangrid is building the rail that delivers it. https://x.com/Iam_T_HUY/status/2109164061476995580

## @alphaticaio (Alphatica) · 10-10 18:25 · ♥32 ↻7 💬3 Every capex bubble in US history popped. Almost none popped on schedule.

1. Railroads, electrification, dot-com. All overbuilt. All burst.

2. The pattern: booms tend to break once cumulative spend on the new technology reaches roughly 25% of GDP. Railroads hit it before the 1873 panic. Internet infrastructure hit it before 2000.

3. Today that line sits near $7.5 trillion. AI is well short of it. At the current pace it doesn't get there until the early 2030s.

4. The spending is also starting to pay. Last quarter NVIDIA guided to 70% revenue growth for fiscal 2028 and Alphabet Cloud grew 82%. Booms keep running while capex turns into profit.

The bubble is real. History says it ends badly.

It just rarely ends at the quarter pole, and that's where AI is.

$NVDA $SPY https://x.com/alphaticaio/status/2108987217129398751

## @mehtasdeepak (Deepak Mehta) · 10-10 14:42 · ♥31 ↻0 💬3 I sold $IREN at $40–46 in the spring. Just bought back in at $38. Cheaper price. Stronger story. Here’s why I’m back

The gap nobody’s pricing in

-FY26 total revenue: $707M -Contracted ARR for 2026 capacity: $4B -ARR running today: only $1B -Management expects ARR above $4B by the end of the December quarter. The contracts are signed; the revenue just hasn’t shown up yet.

The demand is real

-AI Cloud revenue went from $16.4M to $128.8M in FY26, about 8x in a year. -Q4 alone hit $70.5M, more than double Q3. -Customers include Microsoft ($9.7B deal), a frontier AI lab, Perplexity, Cohere and Figure AI. 2026 capacity is basically sold out.

The moat is power -You can order GPUs in months. Grid power takes years. -IREN has 4.5GW+ of secured power, and recent contracts are earning $20M+ per MW.

In an AI world short on electricity, that’s the scarce asset.

The track record -Revenue grew from $59M (FY22) to $707M (FY26), an 86% CAGR. -Analysts expect the real ramp over the next 3 years 📈

The risks, honestly

• FY26 net loss of $702.6M, mostly ~$639M in non-cash write-offs from retiring mining rigs • FY27 capex guided at $25–30B • Revenue lags ARR until GPUs are live and accepted Execution is everything.

My plan In at $38. Holding as the ARR ramp shows up in results. If they execute, I think this can double.

I own $IREN. Not financial advice. Re-rating or overhyped? https://x.com/mehtasdeepak/status/2108931151381364907