@aiconomics
Account based inEurope
About this account
- Account based in
- Europe
- Connected via
- Europe Android App
Account-level information from X, not a live location or the device used for a specific post.
Joined November 2025
- Tweets138
- Following12
- Followers116
- Likes184
$INDI quietly showing what “Physical AI” actually looks like in production.
This demo combines emotion3D DMS/OMS with the iND881, using real-time eye-gaze tracking plus driver/passenger detection.
The interesting part isn’t just the UI.
It’s the stack underneath: vision + edge AI + ultra-low-latency processing inside the vehicle.
That same architecture is exactly why automotive sensing names are starting to matter beyond cars.
Cameras don’t create intelligence by themselves. The value is in processing what they see, locally and fast.
$NOVT $VPG $INDI could be three under-the-radar ways to play the humanoid robotics boom.
Everyone is watching the companies building the robots.
I’m increasingly interested in the companies building their senses and nervous system.
NOVT — Novanta
Force + motion sensing.
Already has a humanoid-focused product range with products available today.
VPG — Vishay Precision Group
Precision sensing.
A humanoid customer hasn’t been disclosed, but the company has been identified as a selected supplier.
INDI — indie Semiconductor
Sensing chips.
indie has disclosed robotics design wins with Unitree and AGIBOT.
That’s what makes this interesting.
Humanoids don’t just need massive AI compute.
They need to see, feel, measure force, understand position and react to the physical world in real time.
NVIDIA may provide part of the brain.
$NOVT, $VPG and $INDI are positioned around the senses and nervous system.
And unlike many of the humanoid companies getting all the attention, these three are already publicly traded.
Everyone asks which humanoid robot will win.
Maybe the better question is:
Who gets paid regardless of which robot wins?
Interesting $INDI finding from looking at Tesla Optimus and Figure AI.
I spent the past several days reviewing a large sample of public demo footage, supplier references, automotive-grade sensing architectures, and known component requirements for humanoid robotics.
What stood out is that the potential overlap with indie Semiconductor’s existing product portfolio may be larger than the market is currently assuming.
Across the systems I reviewed, roughly 80% of the core sensing / perception functions mapped conceptually to areas where $INDI already has products or announced capabilities:
- vision / camera processing
- radar and sensor fusion
- edge perception
- connectivity
- power management
- functional safety
Tesla Optimus and Figure are obviously very different platforms, so I tried to separate tasks by function.
For locomotion-only tasks, the overlap looked relatively limited.
But once the robot was required to navigate around people, identify objects, operate in changing lighting conditions, or interact with its environment, the number of relevant sensing functions increased substantially.
The most interesting part was perception.
In nearly every complex task I reviewed, the robot depended on multiple simultaneous sensor inputs rather than a single vision stream. That architecture looks much closer to modern ADAS than I originally expected.
That matters because $INDI is already positioned around exactly that problem: taking multiple sensor inputs and helping a system understand its surroundings in real time.
I also noticed that as task complexity increased, the theoretical semiconductor content associated with sensing and edge processing rose disproportionately.
My takeaway is that the humanoid robotics opportunity for companies already exposed to automotive sensing could eventually be much more meaningful than most current models assume.
If Tesla Optimus, Figure, or another major humanoid platform were to standardize around architectures similar to advanced ADAS, the semiconductor content per robot could be significant.
Important: this post is parody/fiction. The 80% figure and any implication that Tesla or Figure uses $INDI are invented and are not based on supplier disclosures or confirmed product teardowns.
I don't doubt @OddDiligence's good intentions, but $FSLY pumping that much over a single tweet makes you wonder.
What if he just made the whole thing up?
After a few days, $INDI will do its typical 30% jump, and suddenly everyone will want in.
Don't wait, it's better than $AMBA and $VPG at these levels.
I still believe $INDI is one of the best opportunities for investing in the growth of humanoid robots, and nobody talks about this stock.
Some of the biggest humanoid robot companies in the world are already using indie’s products.
Robots need the same things as cars, the ability to process multiple camera feeds at once, low power usage, and proven reliability. Karpathy even mentioned that the early Optimus robot thought it was a car for this reason.
Other robot suppliers have to develop and test their parts from the ground up. Indie’s components are already being used in cars at scale, which gives them a significant head start.
Additionally, the FCC recently blocked new foreign-built robots from being authorized in the US. Now, any Western robot manufacturer needs a supply chain they can fully document. Indie’s automotive-qualified parts, made in its Swiss and Canadian factories, are a great fit for this requirement.
Indie already has radar, vision, and driver monitoring technology, plus a $7.4 billion backlog from its automotive business.
Meanwhile, the potential of its robotics segment isn’t even being valued yet. That’s what makes this opportunity appealing to me.
The AI Investor retweeted
I still believe $INDI is one of the best opportunities for investing in the growth of humanoid robots, and nobody talks about this stock.
Some of the biggest humanoid robot companies in the world are already using indie’s products.
Robots need the same things as cars, the ability to process multiple camera feeds at once, low power usage, and proven reliability. Karpathy even mentioned that the early Optimus robot thought it was a car for this reason.
Other robot suppliers have to develop and test their parts from the ground up. Indie’s components are already being used in cars at scale, which gives them a significant head start.
Additionally, the FCC recently blocked new foreign-built robots from being authorized in the US. Now, any Western robot manufacturer needs a supply chain they can fully document. Indie’s automotive-qualified parts, made in its Swiss and Canadian factories, are a great fit for this requirement.
Indie already has radar, vision, and driver monitoring technology, plus a $7.4 billion backlog from its automotive business.
Meanwhile, the potential of its robotics segment isn’t even being valued yet. That’s what makes this opportunity appealing to me.
$INDI $3.08 → $5.
Don't miss the move.
$INDI changed up with September OPEX being behind us.
Not the biggest of fan of smaller cap names with only monthly expiries.
But right now we got
GEX KING at - $5 into January
VEX KING at - $5 into January
But watch that Negative Gamma on $7.5 on both panes for November...
$INDI quietly gets MORE interesting every time DRAM gets more expensive.
AI is consuming huge amounts of memory.
🟢 DRAM prices rising
🟢 Auto customers struggling to source memory
🟢 Memory becoming a real production bottleneck
And indie has something unusual:
The iND880 can process automotive camera data WITHOUT external DRAM.
➡️ Lower BOM
➡️ Less memory dependence
➡️ Lower bandwidth pressure
➡️ One less supply-chain bottleneck
The important part?
Management says customers are already redesigning systems around it.
Some automotive design cycles that normally take years are reportedly converting in weeks or a quarter because customers NEED a DRAM-free solution.
And that's only one part of the story:
🟢 Radar moving toward production
🟢 $25M initial radar order
🟢 Volvo design win
🟢 Unitree + AgiBot exposure
🟢 CMOS sensor acquisition
🟢 ~$135M potential cash from Wuxi sale
The hidden $INDI thesis may be simple:
AI → memory scarcity → DRAM gets expensive → DRAMless automotive vision becomes more valuable.
The worse the memory bottleneck gets…
the more interesting $INDI's architecture becomes.
If you don't believe it or don't get it, I don't have the time to try to convince you, sorry.
Full port $INDI.
I'm with Mr. Xu on this one.
He drops a ticker and goes all in on it.
It's way easier to post 20 charts a day and just cherry-pick the 2 that actually worked.
Sometimes even pulling up posts from 2025.
He undoubtedly called $AEHR.
just woke up and catching up on things so i'll reply here:
- my original comment was more trying to say it's different to share multiple ideas (with some plausible deniability like "bounce time?" which feel more like an observation than a call tbh) vs me going "all in" 1 stock with the extreme fervor and attention i bring to my picks. very different levels of conviction and public scrutiny.
- yes i exaggerated your 20 picks per day. sorry.
- tbh i do like your observations a lot. one of the better folks on this app.
- that actually wasn't rage bait or big account baiting. it was a reply. it was more just trying to defend myself a bit against that other replier.
- $AEHR was on my radar for a long time. i noticed the exact dip to $75 at ~10:50am PT on 9/3 and bought it. your post was at 8:18pm PT on 9/1 which may have entered my subconscious as something to pay attention to. but the buy itself i determined it.
yes $AEHR was a good short-term bounce play. yes my trades can be short or medium term depending on people's definition of those words.
i like having thesis bc every stock needs a thesis in order to enter otherwise its just luck.
i'm allowed to change my mind when the facts change. fedwatch odds of rate hike went from 70% to 50% last Thursday so that's why i flipped bullish to bearish. i gave much more details on why to my subs.
sorry for the rough morning. i really didn't mean anything against you in particular.
it was more general commentary about my style of going "all in" 1 stock in my challenge account and everyone knows it and publicly scrutinizes it vs the general practice of talking about "multiple" stocks via call outs / observations throughout the day with no position attached.
which just feels like different levels of playing this game.
I’ve seen several comments claiming $CBRS insiders are “dumping shares” and that Cerebras CEO Andrew Feldman sold 93% of his position.
Yes, there has been insider selling.
No, he did not sell 93% of his stake.
The real figure is closer to ~1.7%.
@andrewdfeldman exercised roughly 238K options, converted those shares from Class B into Class A, and then sold those same shares under a pre-arranged 10b5-1 trading plan.
Some insider trackers then look only at his Class A balance:
➡️ ~254K Class A shares before
➡️ ~238K sold
➡️ ~17K Class A left
And automatically report:
🚨 “Ownership down 93%”
That is technically describing one share bucket.
It is not describing Feldman’s total ownership in Cerebras.
After the sale, Feldman still owned roughly:
🟢 **13.9M Class B shares directly**
🟢 Additional Class B shares through trusts
🟢 Other equity/options
Class B shares are economically convertible 1-for-1 into Class A.
So the relevant comparison is not:
238K sold vs. 254K Class A.
It is:
238K sold vs. roughly 14M total shares of economic exposure.
That puts the sale at roughly ~1.7% of his common equity exposure, not 93%.
Huge difference.
There’s another nuance.
Some recent insider sales were also sell-to-cover transactions for taxes, not discretionary selling.
That does not mean insider selling should be ignored.
There is real selling at $CBRS.
Some executives are monetizing part of their holdings after the IPO, and that is worth tracking.
But there is a massive difference between:
“Insiders are taking some liquidity.”
and
“The CEO dumped 93% of his company.”
The first is true.
The second is not.
**Real number: ~1.7%, not 93%.**
$CBRS quietly gets MORE attractive every time HBM prices go up.
The AI industry is in an arms race for memory.
🟢 HBM demand keeps exploding
🟢 DRAM pricing keeps moving higher
🟢 $MU and $SKHY are printing extraordinary margins
🟢 NVIDIA GPUs increasingly depend on massive HBM bandwidth
🟢 Advanced packaging remains one of the critical bottlenecks in AI infrastructure
Cerebras largely designed around that bottleneck.
The Wafer Scale Engine does something radically different.
Instead of constantly moving model weights back and forth between compute and expensive external HBM…
Cerebras puts enormous amounts of SRAM directly next to the compute.
➡ 44GB of on-wafer SRAM
➡ ~21 PB/s internal memory bandwidth
➡ 900,000 AI cores
➡ No traditional HBM stack
➡ No dependency on CoWoS in the same way GPU systems do
That means something very important:
The more expensive HBM becomes, the more valuable Cerebras’ architectural choice becomes.
Think about the economics.
NVIDIA architecture:
GPU
* HBM
* advanced packaging
* interconnect
* switches
* networking
* increasingly expensive memory
Cerebras:
Wafer-scale compute
* SRAM
* ultra-fast local data movement
This doesn’t mean HBM disappears.
Training, prefill, huge contexts and high-throughput workloads will continue to consume enormous amounts of HBM.
But inference is starting to split.
And low-latency decode is increasingly becoming its own hardware category.
That’s why:
🟢 OpenAI committed to 750MW of Cerebras inference capacity
🟢 AWS is pairing Trainium with Cerebras for disaggregated inference
🟢 AMD is working with Cerebras on the same prefill/decode split
🟢 NVIDIA itself is moving toward SRAM-heavy inference through Groq
Notice what’s happening.
The industry isn’t saying:
“HBM is useless.”
It’s saying:
“Why use the most expensive memory in the system for workloads that can be architected around it?”
And this is where $CBRS becomes extremely interesting.
If HBM stays scarce and expensive:
➡ Cerebras becomes relatively cheaper
➡ Its supply chain looks more differentiated
➡ Its SRAM-first architecture becomes more valuable
➡ Customers have another path to scale inference without competing for the exact same HBM bottleneck
And if AI agents explode?
The value of fast decode only gets bigger.
Agents don’t make one model call.
They make:
Model → tool → model → code → model → search → model → verification…
Over and over again.
Latency compounds.
So the real $CBRS thesis may not be:
“Cerebras replaces NVIDIA.”
It may be:
Cerebras becomes the specialized inference engine used wherever HBM-heavy GPUs are economically inefficient.
That is a MUCH bigger distinction than most investors realize.
$MU and $SKHY can absolutely keep winning as AI memory demand explodes.
But ironically…
the higher HBM prices go, the stronger the incentive becomes for engineers to design around HBM.
And Cerebras spent more than a decade doing exactly that.
$CBRS quietly gets MORE attractive every time HBM prices go up.
The AI industry is in an arms race for memory.
🟢 HBM demand keeps exploding
🟢 DRAM pricing keeps moving higher
🟢 $MU and $SKHY are printing extraordinary margins
🟢 NVIDIA GPUs increasingly depend on massive HBM bandwidth
🟢 Advanced packaging remains one of the critical bottlenecks in AI infrastructure
Cerebras largely designed around that bottleneck.
The Wafer Scale Engine does something radically different.
Instead of constantly moving model weights back and forth between compute and expensive external HBM…
Cerebras puts enormous amounts of SRAM directly next to the compute.
➡ 44GB of on-wafer SRAM
➡ ~21 PB/s internal memory bandwidth
➡ 900,000 AI cores
➡ No traditional HBM stack
➡ No dependency on CoWoS in the same way GPU systems do
That means something very important:
The more expensive HBM becomes, the more valuable Cerebras’ architectural choice becomes.
Think about the economics.
NVIDIA architecture:
GPU
* HBM
* advanced packaging
* interconnect
* switches
* networking
* increasingly expensive memory
Cerebras:
Wafer-scale compute
* SRAM
* ultra-fast local data movement
This doesn’t mean HBM disappears.
Training, prefill, huge contexts and high-throughput workloads will continue to consume enormous amounts of HBM.
But inference is starting to split.
And low-latency decode is increasingly becoming its own hardware category.
That’s why:
🟢 OpenAI committed to 750MW of Cerebras inference capacity
🟢 AWS is pairing Trainium with Cerebras for disaggregated inference
🟢 AMD is working with Cerebras on the same prefill/decode split
🟢 NVIDIA itself is moving toward SRAM-heavy inference through Groq
Notice what’s happening.
The industry isn’t saying:
“HBM is useless.”
It’s saying:
“Why use the most expensive memory in the system for workloads that can be architected around it?”
And this is where $CBRS becomes extremely interesting.
If HBM stays scarce and expensive:
➡ Cerebras becomes relatively cheaper
➡ Its supply chain looks more differentiated
➡ Its SRAM-first architecture becomes more valuable
➡ Customers have another path to scale inference without competing for the exact same HBM bottleneck
And if AI agents explode?
The value of fast decode only gets bigger.
Agents don’t make one model call.
They make:
Model → tool → model → code → model → search → model → verification…
Over and over again.
Latency compounds.
So the real $CBRS thesis may not be:
“Cerebras replaces NVIDIA.”
It may be:
Cerebras becomes the specialized inference engine used wherever HBM-heavy GPUs are economically inefficient.
That is a MUCH bigger distinction than most investors realize.
$MU and $SKHY can absolutely keep winning as AI memory demand explodes.
But ironically…
the higher HBM prices go, the stronger the incentive becomes for engineers to design around HBM.
And Cerebras spent more than a decade doing exactly that.
Over the next 8 quarters, the sell side expects SK Hynix to generate $492 billion of net income.
Over the next 8 quarters, the sell side expects Microsoft ($MSFT) and Apple ($AAPL) to generate a combined $606 billion of net income.
SK Hynix enterprise value is $940 billion.
Microsoft and Apple combined enterprise value is $8.4 trillion.
Over the next 8 quarters... SK Hynix is expected to generate 81.2% of the net income of Microsoft and Apple (combined) with just 12% of the enterprise value.
Wild stuff!!!!
$OUST is quietly turning into one of the most compelling Physical AI stories in the public market. 🚀
And the numbers are starting to get ridiculous.
🟢 Revenue: $55M, +56% YoY
🟢 Product revenue: $53M, +51% YoY — RECORD
🟢 14 straight quarters of product revenue growth
🟢 17,000+ lidar + camera sensors shipped in Q2
🟢 Lidar unit volumes: +70% YoY
🟢 GAAP gross margin: 49%, +400 bps YoY
🟢 Non-GAAP gross margin: 53%
🟢 Adj. EBITDA loss: just $4M, improving from $7M in Q1
🟢 Q3 revenue guide: $54.5–57.5M
🟢 $263M cash/investments at Q2-end + ~$192M NET raised in July
🟢 150,000+ lidar sensors shipped historically
🟢 90,000+ cameras shipped
🟢 10,000+ customers
But the real story is what’s happening underneath the financials.
REV8 may be the inflection point.
🟢 World’s first patented native-color lidar
🟢 Up to 2X the range + resolution of Rev7
🟢 Auto-grade, cybersecure + designed for functional safety
🟢 Manufacturing capacity now exceeds 100,000 units/year
🟢 Planned 10-year production life
🟢 NVIDIA DRIVE Hyperion qualified
🟢 Integrated across NVIDIA Jetson for robotics + edge AI
🟢 Build America, Buy America compliant — opening federally funded infrastructure deployments
And customer adoption is accelerating FAST.
🟢 Multiple $1M+ REV8 orders from major heavy machinery, autonomous agriculture and AV customers
🟢 Utah DOT: several HUNDRED intersections — its largest lidar deployment ever
🟢 BlueCity deployed across 72 highway/intersection locations around the 2026 World Cup
🟢 John Deere-owned GUSS plans to integrate REV8 into autonomous orchard machines
🟢 GeoCue is integrating REV8 into survey-grade drone mapping
🟢 FieldAI is using REV8 for autonomous robots in complex real-world environments
🟢 Stereolabs cameras are being embedded into Trossen’s Physical AI / robot-learning platforms
🟢 ZED X Nano was the most successful product launch in Stereolabs history
This is the key:
$OUST is no longer just a lidar company.
Lidar + stereo vision + AI compute + sensor fusion + perception software + AI models.
➡ Robotics.
➡ Autonomous machinery.
➡ Smart infrastructure.
➡ Agriculture.
➡ AVs.
➡ Drones.
Every Physical AI system has the same fundamental problem:
It needs to SEE and understand the real world before it can act in it.
Ouster is building the sensing layer underneath that entire transition.
➡ Revenue is accelerating.
➡ Units are exploding.
➡ The product stack is expanding.
➡ Manufacturing is scaling.
➡ The balance sheet is loaded.
Physical AI is hitting an inflection point and $OUST is scaling directly into it.
This is starting to look a lot less like a sensor company…
…and a lot more like a Physical AI platform. 🚀
