# data center revenue — X 热门讨论 (2026-09-20 14:28 UTC)
## @HunterAllen4 (THE GAP FATHER) · 09-20 05:50 · ♥32 ↻1 💬14 Physical AI.
The $NVDA stack.
NVIDIA is attacking almost every layer with Jetson Thor, Isaac, Cosmos, GR00T, Omniverse and Halos. The shift is from AI that generates answers to AI that can sense, reason, decide and physically act.
The adoption curve is getting harder to ignore. 542,000 industrial robots were installed globally in 2024, taking the operating fleet to 4.66M, up 9%. Asia accounted for 74% of new installations, with China alone at 54% and more than 2M robots already in operation.
Mobile robots are projected to grow from just under $5B in 2024 to ~$14B by 2030, around 19% CAGR. Capgemini found 66% of executives consider Physical AI a high priority, while 65% expect to reach scale within five years.
And the capital is following. Physical AI and robotics startups pulled in $47.4B in H1 2026, roughly 4x H2 2025 and more than the combined total from 2022–2024.
At the platform level, NVIDIA just posted $96.2B quarterly revenue, up 106%, with $89B from Data Center, up 117%. That is the financial engine behind the compute, simulation and edge infrastructure being deployed across this ecosystem.
That creates a MUCH bigger ecosystem than NVIDIA alone.
Think AI compute → ARM CPUs → sensors → perception/models → planning → control → certified safety → real-time OS → OT security → actuators. $ARM has 80+ companies in its Physical AI ecosystem and shipped billions of Arm-based chips into adjacent markets.
$PANW sits further downstream securing connected factories, OT environments, AI infrastructure and machine identities.
The ramps are what matter. $NVDA’s Automotive revenue hit $2.3B in FY26, up 39%, while Physical AI expands across robotics, AVs and industrial systems. $ARM generated $2.61B in FY26 royalty revenue, driven by growth across Edge AI, Physical AI and Cloud AI.
$PANW is attacking the security attach as autonomous machines become connected agents. Different layers, same machine-count explosion.
NVIDIA’s Halos for Robotics spans compute, sensor connectivity, safety software and certification, and its ecosystem explicitly includes QNX for the real-time operating environment and embedded safety layer. Early configurations support Linux + QNX OS for Safety 8.0.
That’s the piece I think the market is still underestimating. $ARM benefits as more intelligence moves to the edge. $PANW benefits as those machines become connected and autonomous.
QNX benefits when certified edge control becomes mandatory. It doesn’t need to replace Linux or NVIDIA it can operate alongside the AI workload where deterministic, safety-critical execution matters.
The global ramp is much bigger than humanoids. China already dominates industrial-robot volume, Japan remains a robotics powerhouse, the U.S. is pushing models and orchestration, and Europe brings industrial standards and certification.
NVIDIA is building the simulation and digital-twin layer so fleets can be trained, tested and virtually commissioned before deployment. The next KPI isn’t another robot demo — it’s fleet utilization, cost per task, uptime and payback.
Remember $FPS? I was pounding the table on that hidden infrastructure layer ahead of earnings.
Now I’m seeing another setup where a massive platform is expanding into a new market while a much smaller company sits in a critical part of the architecture.
I don’t need it to become NVIDIA. I need the market to realize what layer it owns and how much Physical AI can expand that opportunity.
If you want the actual setup before I start talking about it publicly, get inside the Gap Father Swings channel NOW.
Don’t wait for the ticker to start moving and then ask where I found it. Get in, see the setup, and do your own DD.
The picks and shovels beneficiary…
Know someone who needs to see this? TAG THEM BELOW. REPOST THIS. DONT SLEEP. DONT MISS
Spread the word. Let’s find the next Rip Salad together.
🥗 JOIN HERE THE GAP FATHER SWINGS CHANNEL BELOW.
https://t.co/0dNvIlw1Ey https://x.com/HunterAllen4/status/2101549476787699788
## @blackroomsec (BlackRoomSec) · 09-19 02:24 · ♥30 ↻3 💬5 Hi again Bee!
First, a correction, yesterday I said this using the term hyperscaler interchangeably with an AI server:
This is computer code on a hard drive on a server which is very expensive called a hyperscaler.
Hyperscalers are installed in specialized server racks vertically.
They are drilled into the server racks and each hyperscaler can be accessed by opening up a door in the front of those racks.
I apologize. I was writing it quickly and was very tired so before I talk about why I believe they have to IPO, I need to actually CORRECTLY define a hyperscaler for you.
A hyperscaler is the company which owns the datacenters where the AI servers are located, NOT the actual servers itself, even though the majority of the servers the hyperscaler owns, are, at this juncture, and for all intents and purposes, AI at this point.
Prior to four years ago, a hyperscaler data center was just regular non AI servers and such.
So my brain just equates the two as one and the same.
Still, I used the term incorrectly and I need to hold myself accountable here.
Two big hyperscaler companies are Microsoft and Amazon.
With that said, AI data centers are built in a very specific way and are made up of ridiculously expensive hardware that a typical, traditional data center with regular servers we've all been using for years? Does not have.
They have gone ALL IN on the AI.
And, because of that, they need to feed their habit with new investor money.
AI servers have video cards inside them that we refer to as GPU. You may have heard the term. GPU is Graphical Processing Unit.
Gamers purchase expensive video cards so they can play their games. The GPU does math and renders images really fast, faster than a traditional CPU (Central Processing Unit) on computer motherboards.
A decent video card will run several hundred dollars. The faster you want, the more you pay.
Mine was around $1500, as an example. It's a "Titan Class" video card but is not an actual Titan card. They don't make them anymore but we cling to the term for nostalgic purposes.
It's colloquial, metaphoric, like my "hyperscaler" use earlier.
NVIDIA, the company that makes the GPU that the hyperscalers and AI companies buy, ironically enough, once made a series of video cards for regular gaming computers with Titan in the name.
Gamers will sometimes refer to their non-NVIDIA high end video cards by the term "Titan Class" , even though they technically aren't.
AI hyperscalers have millions of AI servers and the GPU (video cards or ASICs) which do the math to help the AI answer questions and perform tasks, really fast.
But THOSE video cards the hyperscalers own are around $40,000 each.
Let's say they have a million GPUs. That is 40 billion dollars.
They do not reveal to the public how many servers they have (both AI and traditional) for security reasons but it is estimated in the millions.
But, and here's the catch that early investors who are NOT technical probably didn't understand when they invested.
These GPUs for AI?
Have to be replaced every two to three years. They wear down very quickly.
In traditional data centers, you can get away with stretching your servers for up to seven years if you extend the warranty and the vendor supports it.
Most companies are not replacing every server they have every seven years though. It's expensive but not 40 billion dollars expensive.
The AI companies? MUST replace the AI servers every two to three years. They have no choice.
Now for four years they have been dangling a carrot, trying to get every human on earth to use their products, but the majority of their users are FREE, not paid.
And, they have to pay for staff, operating expenses, these damn video cards, etc, so the costs rack up rather quickly.
Last week, Anthropic put out a press release or a statement saying that if you overlook their operating expenses, payroll, etc, they make 80% margin.
Fintech (The Financial Tech sector of the stock market) responded with uproarious laughter to this ridiculous statement.
When I spoke on it, I pointed out that Apple and Microsoft operate at 50% margins.
Margin is how much money on the dollar the company makes in profit AFTER operating expenses.
80% means, they keep 80 cents on the dollar.
That is preposterous given the near TRILLION dollars both companies have spent on operating costs in the last four years AND both companies have paid users in the 15% range at the highest whereas their FREE USERS are at 85%.
Meaning almost all of their users are not sending them any money to use their product.
They are almost ALL using it for FREE.
How do they stay in business?
By taking other people's money.
The IPO is critical here.
It is misleading for them to say they have this margin because if they make 100k in profit but their operating expenses are 120k, they are LOSING 20k, not making it.
Whereas MS and Apple, DO have 50% margins, because they have multiple revenue streams across dozens of markets and have been in business for quite some time, each.
OpenAI and Anthropic only have one revenue stream: AI.
Let's pause here because I'm ready to crash. I don't think I'll need to make any corrections on this post but if I do, I will do so tomorrow. I want to see what you say next.
Be well. > 引用 @blackroomsec: Reposting so I know to respond to this tomorrow. https://x.com/blackroomsec/status/2101135304191103147
## @Mojo_flyin (Mojo) · 09-20 03:45 · ♥32 ↻3 💬1 The Street consensus on $INTC Q3 26 earnings is:
𝗘𝗣𝗦 𝟬.𝟯𝟵 𝗖𝗘𝗡𝗧𝗦
𝗥𝗘𝗩𝗘𝗡𝗨𝗘 $𝟭𝟲.𝟯𝟲 𝗕𝗟𝗡
I will complete my analysis and publish my estimate again a few days before earnings. But I can tell you now they are wrong on both.
This quarter will be very interesting as factory execution could deliver unpredictable upside surprises.
Demand for data center is firing from both #AI and enterprise. @Intel isn't margin hit by HBM unlike $AMD $NVDA $AVGO but supply is highly constrained.
Meanwhile PC softness is creeping in. Additionally, $AMD willingness to give away client parts at declining margin for the last few qtrs means share and margin are not both easily supported. Panther Lake on @Intel_Foundry 18A does bring some benefits to both positionimg and margin.
Judgement on numbers will be even more complex in Q3.
It will be interesting to see if I can go 3 for 3 on my forecasts.
@Intel @Intel_Foundry $INTC > 引用 @Mojo_flyin: $INTC EARNINGS ARE IN AND THIS TIME I HIT BOTH REVENUE AND EPS FCST
𝗤𝟮 𝗿𝗲𝘃𝗲𝗻𝘂𝗲 $𝟭𝟲.𝟭 𝗯𝗹𝗻
𝗘𝗣𝗦 $𝟬.𝟰𝟮 https://t.co/mMqJNJzO2Y https://x.com/Mojo_flyin/status/2101518150101803250