# AI capex — X 热门讨论 (2026-10-10 09:40 UTC)
## @ZsomaBro (ZSOMA) · 10-10 05:25 · ♥39 ↻1 💬47 Good Morning Legends! ☕️🌅 Happy Saturday!
Pouring a fresh cup, getting a heavy lifting session in, and locking in for the weekend grind. Let's break down how @vangrid_io is rewriting the rules for spatial intelligence and embodied AI.
Traditional mapping requires billions in capital expenditure for vehicle fleets, whereas Vangrid turns existing smartphones into a scalable, zero-capex sensor network.
Led by CEO Robert Brighton from Microsoft’s HoloLens and Quantum groups and CTO Khalid Benrafik with a track record of handling over $1B in annual transactions, the team brings elite engineering execution to Amsterdam.
The protocol is fully live with active network metrics, robust backing from investors like Animoca Brands and HashKey Capital, and official membership in NVIDIA Inception.
Step away from networks that monetize your data footprint. Build your edge on infrastructure designed for total digital sovereignty and absolute censorship resistance.
Are you positioned in infrastructure that captures real-world utility, or are you still farming empty narratives? https://x.com/ZsomaBro/status/2108791117042889158
## @qqa16809 (亞子) · 10-10 06:55 · ♥32 ↻2 💬26 麦当劳股价跌破2022年最低点了，真得厉害。
不知道的还以为美股股灾了。Capex对非AI行业资金的misallocation，创造了很多受害者
投资泡沫结束后，这些公司应该是反弹最猛烈的 https://x.com/qqa16809/status/2108813579453608325
## @Vanquan_titans (Van Quan) · 10-10 03:01 · ♥31 ↻1 💬27 Everyone is debating when humanoid robots and autonomous agents take over the physical world.
The question isn't whether the AI models are smart enough. It's: where do they actually get groundlevel spatial data when real environments change overnight?
You can scrape text and code off the web. You cannot scrape an updated loading dock, a temporary barrier, or an awkward pedestrian underpass. Satellites don't see beneath roofs, and burning tens of millions on dedicated fleets of camera cars simply doesn't scale.
This is why @vangrid_io caught my eye. Instead of heavy capex on custom hardware, they are crowdsourcing groundlevel 3D spatial captures straight from ordinary smartphones, turning real footage into verifiable data for Physical AI and settling bounties in USDC.
Robots don't operate inside clean lab simulations; they have to navigate the messy physical world. Sourcing that data on demand without massive infrastructure overhead is where the real commercial bottleneck sits. https://x.com/Vanquan_titans/status/2108754643157397970
## @PhotonCap (Photon Capital) · 10-10 05:49 · ♥32 ↻2 💬0 - Morgan Stanley identified testing as a key CPO manufacturing bottleneck on September 3.
- Integrating HBM, XPUs, and optical engines increases test complexity and automation requirements.
- At Morgan Stanley's March conference, the 2025 semiconductor test equipment market was estimated at ~$9B.
- $TER's CEO explained that higher HBM stack counts drive wafer-sort demand, while HBM4 increases post-stack testing requirements.
- $TER announced production deployment of Titan HP with integrated SLT and burn-in on October 6.
- $AEHR announced ~$6M in follow-on Sonoma orders for a next-generation AI processor on October 7.
> As AI chips become more valuable and packaging becomes more complex, known-good-die screening, yield management, and reliability testing become increasingly critical.
> Rising test CAPEX creates opportunities for $TER, $FORM, $COHU, and $AEHR, but actual upside depends on market share and customer adoption at each testing stage. > 引用 @PhotonCap: AI capex is moving downstream.
First GPUs. Then power. Now packaging and test.
The bottleneck keeps migrating. https://x.com/PhotonCap/status/2108797138574418109