# AI capex — X 热门讨论 (2026-09-19 12:17 UTC)
## @Akewusolaf (A.F Adewale) · 09-19 08:44 · ♥31 ↻1 💬33 Zero CapEx Sensor Swarm
The more I look at @vangrid_io, the more the Zero CapEx Sensor Swarm makes sense
Physical AI has a data problem
Robots can have better models, more compute, and better hardware, but they still need to understand the environments they actually operate in
And the physical world is constantly changing
✅️A loading area gets blocked ✅️A construction site changes ✅️A warehouse gets rearranged ✅️A sidewalk gets damaged ✅️A temporary obstacle appears ✅️A map collected months ago might still tell you where something is, but not necessarily what is there right now
That is where Vangrid’s approach gets interesting
Instead of requiring contributors to deploy another expensive sensor device, the idea is to turn existing edge devices into a distributed capture layer
That is the Zero CapEx part
The hardware barrier moves dramatically lower because the network can leverage devices people already carry
But collecting images alone is not enough.
The bigger question is whether the data can become useful ground truth
Vangrid describes a pipeline built around edge computed privacy, cryptographic provenance and multi view ingestion, with the goal of producing spatial data for autonomous systems and other Physical AI applications
That distinction matters
You are not simply creating another giant folder of random footage
You are building infrastructure around the capture, verification and delivery of spatial information
Think about what this could mean at street level
A robot needs to understand the geometry of a narrow passage
An autonomous delivery system needs fresh information about a loading zone
A warehouse system needs to understand a changing layout
An embodied AI model needs more than a satellite image to understand the environment from the perspective of something actually moving through it
This is why I keep coming back to the sensor swarm idea
The network does not need every contributor to own specialized mapping hardware
The network needs enough distributed capture capacity to collect useful information where and when it matters
That changes the economics of spatial data collection
Instead of building a centralized fleet first and expanding city by city, the infrastructure can potentially grow through the devices already distributed across the physical world
The difficult part, of course, is not simply getting more captures
It is maintaining quality, provenance, privacy and usefulness at scale
That is where @vangrid_io infrastructure thesis becomes more interesting than simply calling it a mapping app
Physical AI needs a connection to reality
And reality keeps changing
If everyday devices can become part of a distributed sensor network that continuously updates our understanding of the physical world, the smartphone stops being just a communication device
It becomes part of the data infrastructure for machines
That is the Zero CapEx Sensor Swarm thesis I’m watching
✅️Not more hardware ✅️More distributed sensing ✅️Not just bigger datasets
Fresh spatial ground truth
And not simply mapping the world for humans, but helping machines understand the world they have to operate in.
What caught my attention about @NucleusCodes is the separation between reputation and contribution
Reputation looks at what you’ve already built through your on chain history
Contribution looks at what you are actually doing for a campaign now
That distinction matters
Your wallet history can show a track record, but current contribution shows whether you are adding value today
Instead of reducing Web3 participation to follower count or campaign activity alone, Nucleus is trying to build a broader reputation layer around the signals people leave across the ecosystem
That makes the leaderboard more interesting because the history and the current work are being treated as different signals. https://x.com/Akewusolaf/status/2101230828290236492
## @HunterAllen4 (THE GAP FATHER) · 09-18 15:53 · ♥30 ↻1 💬8 $IMOS is exactly the kind of semiconductor name I’m hunting for.
I’ve posted multiple times but very few have even heard of or tried to look deeper into $ASX stack.
ITS BEATEN DOWN FROM 80$.
If you haven’t heard of this one yet? Subscribe guys. 1$ for x ideas 5$ for swing channel link in bio.
I posted ChipMOS at $45 on 7/29, and there are still only a handful of eyes on it. AI-driven memory demand is staying strong, DDR5 is ramping, and management expects 2H26 to be stronger than 1H26.
August sales were up 33.3% YoY, while Q2 revenue reached a quarterly high dating back to 2014.
But the bigger picture is what has me watching it. ChipMOS is positioned in the memory backend / OSAT layer, and it’s increasing capex around memory assembly and test bottlenecks, AI/automation, silicon photonics testing and AI ASIC work.
Management is also pointing to high customer visibility through 2026 and adding capacity against committed demand.
This is the type of underappreciated AI infrastructure play I’m constantly digging for.
Not chasing the obvious names after everyone discovers them finding the companies underneath the theme while they’re still flying under the radar, then watching for the technical turning point.
If you’ve ever wondered why you’d subscribe to my research, THIS is why.
I’ve got almost 700+ legends inside my X subscription, while the Gap Hunter Swing Channel is 55 members strong for more calculated, selective swing ideas.
I appreciate every single one of you. Whichever route you choose, just know I’m working harder than ever to find the next potential Rip Salad before everyone is talking about it. 😤❤️
DONT MISS. DONT SLEEP. DONT FADE. https://x.com/HunterAllen4/status/2100976504687509979
## @MikeLongTerm (Mike) · 09-19 03:25 · ♥30 ↻0 💬2 $PLTR | I think about this a lot.
The implication of what if Frontier Labs do what Dr. karp said.
What will happen to the Compute Demand or overall AI CapEx.
I don't think AI Progress gonna stop.
I'm thinking more about of a slower AI CapEx growth YoY (2026-2027).
We will find out. > 引用 @MikeLongTerm: BREAKING $PLTR Dr. Karp Full Interview CNBC (Brandnew) on AI Safety🆕🚨🚨 09-17-2026
Dr. Alex Karp CEO of @PalantirTech discussing @AnthropicAI @OpenAI AI safety drama.
~ Palantir customers biggest risk is IP theft ~ Sovereignty AI with $NVDA partnership is very important to Palantir and customers on using AI Model ~ Customers get to control their future ~ This is not about Safety vs regulation, this is about IP protection/liability ~ Wealth tax will kill entrepreneurship, u have to find another way ~ Palantir dont build data center🤣 ~ Businesses should be responsible before releasing the product. First step disclose it and what do you do abt it. Regulate them can be very complicated. ~ "Nationalize us" to protect from liability 🧐 ~ Unlimited risk when all customers IPs move into a few institutions ~ What Anthropic Said is not what they meant ~ Palantir thrives because we make these models to work in precisely and complex workloads or use case.
~ We have to follow a set of rules ~ Businesses worry about liability ~ We are going to own Ontology ~ Businesses want Sovereignty ~ Palantir is so undervalued ~ Demand is so high and we have limited capacity. Hiring & training to be full Palantirian is a process (take time) ~ Dario is one of the most important entrepreneurs, I have a lot of respect for him ~ Countries/Businesses will use close and open models
Source & Credit: CNBC & Dr. Karp https://x.com/MikeLongTerm/status/2101150572460040515