Verifying the Truth of the "AI Bubble Theory" from NVIDIA, AMD, and Siemens' CES 2026 Keynotes - From Fund Garage Mr. Oshima's YouTube Stream
I am Kusuura, CEO of TechnoProducer Inc. and head of the Invention School. Today, I am sharing a portion of my research notes on "corporate analysis," "management strategy," "management," "technology strategy," and "intellectual property strategy," which I compile daily through brainstorming with AI to prepare for the lectures (seminars) and workshops I conduct on IP strategy at companies, universities, and research institutions.
This time, I will take up the "AI bubble theory."
Our company often supports the planning of new businesses (mainly technical) for companies through our "Invention School" service. During these sessions, discussions like "By the way, the generative AI boom is a bubble, so it won't last long, right?" or "Isn't it too late to start now?" sometimes come up, which can be a bit of a hassle (laughs).
Actually, I want to explain in a way anyone can understand that now is the time to start (laughs). I often refer to the YouTube streams for investors by Mr. Oshima of Fund Garage. Since he explains things in a way that is (probably) easy to understand even for non-technical people, I use his approach as a reference for my own explanations.
This time, I am posting notes that I had an AI summarize, including my own notes and insights, based on his stream explaining the keynotes by the CEOs of NVIDIA, AMD, and Siemens at "CES 2026" below.
First, I will post the AI summary of the video.
"Based on the trends of major companies at CES 2026, one of the world's largest technology trade shows, this explains how the AI business has moved beyond a mere boom into a new stage of generating actual profits. Leaders from companies like NVIDIA, AMD, and Siemens defined AI not as a single component, but as factory-level infrastructure called an "AI factory," emphasizing monetization through 24/7 continuous operation. Its role is shifting from one-off Q&A services to control systems directly linked to improving operational efficiency and reducing waste in industrial settings, and peripheral equipment such as memory, networking, and cooling technology have been elevated to essential requirements. Furthermore, the importance of Oracle for maintaining data integrity and Palantir for operational governance was highlighted as key to supporting this massive 'factory.' In conclusion, the AI bubble theory misidentifies where profits are generated, and argues that implementation as an industrial OS is the true coordinate for future growth."
The AIs used were "ChatGPT," "NotebookLM," and "Gemini." These are notes from the research I conduct daily to deepen my understanding, so I would appreciate it if you could leave a comment if you have any questions or points of confusion.
This time, I am also posting a portion of my handwritten notes. I have only confirmed the CEO keynotes for NVIDIA and AMD myself so far. I will check the Siemens CEO's keynote in the near future. I will add any points I notice, so if you want to check for follow-up reports, please "like" this for your own reference.
== Kusuura's Notes
If you want to check the overall view, please start by looking at the "AI Memo."
Here, I will write down the points I thought were "particularly noteworthy" or "worth watching in the future" after watching the video, while including AI-generated summary slides.
AI does not end with "chat"
First, starting from the most important point: "Generative AI is just chat, right?"
This is already underway, but it is the stage where AI, based on internal company data including human instructions and emails, issues instructions for inventory management and production adjustments, or actually begins to execute them. Natural language processing is necessary, but it doesn't end there; it becomes an entity that keeps the mechanism running constantly behind the scenes of manufacturing and service provision. Mr. Oshima calls this "constant inference."
Palantir, which appears in the video, is also deeply involved in the execution of US military operations and provides a mechanism where "stopping would be a problem." If you are wondering what Palantir is, please refer to the following.
Where does Physical AI make money?
This is also a common question.
It seems easy to understand that robots will increase. Along with that, the amount of data and processing will increase, so NVIDIA chips will also sell. People seem to be able to grasp it up to this point, barely.
In between, there is an 'industrial OS' dominated by companies like Siemens. To start with, you can think of it as being the same as having a PC (or smartphone), apps, and an OS. While people don't complain much if a smartphone stops for a bit, a factory stopping leads directly to massive losses, so reliability here is crucial. It won't be a one-time sale, but (likely) a 'subscription model' that is constantly managed and updated.
The main players in industrial OS and industrial AI are Siemens, Palantir, and Oracle.
This is for my own reference (laughs).
The role Palantir plays is that of an 'ontology.' A string of data has no meaning, so they provide interpretation—to put it simply—to clarify what each piece of data means.
I would like to introduce what their systems are specifically doing in a separate newsletter while explaining the patents.
(This is my reference note)
== AI Memo
The End of the 'AI Bubble' Theory Proven by CES 2026: The Day the Source of Revenue Shifted from 'Chat' to 'Factories'
I had the AI create a few versions, and I have posted the simplest one among them.
Infographic
Infographics are indeed easy to understand.
1. Introduction: Why is the term 'AI bubble' missing the point?
Currently, cold debates about whether 'AI is a bubble' are swirling in the market. Concerns such as high P/E ratios, expectation-driven stock prices, and the completion of demand cycles are likely 'noise' that investors and business leaders cannot ignore. However, the vision shared by the CEOs of NVIDIA, AMD, and Siemens during their CES 2026 keynotes proved that these anxieties are completely 'missing the point.'
Most bubble theories only discuss the high 'P' (Price) of the P/E ratio and overlook the fact that the structure generating the 'E' (Earnings) in the denominator is changing dramatically. The axis of competition has already shifted from 'peak performance' to 'operational stability and uptime.' Now that the source of profit has shifted from 'human dialogue (chat)' to 'non-stop factories,' we must redefine AI not as 'convenient software' but as 'heavy industrial infrastructure.'
2. A Shocking Turning Point: AI from 'Talking Services' to 'Non-stop Factories (AI Factories)'
The AI boom until now has centered on event-driven services like 'chatbots' that only respond when a human asks a question. However, the core discussed at CES 2026 is the transition to 'continuous inference (AI agents)' that operate 24/7.
In a world where AI agents autonomously monitor factory sensors, manage inventory, and continue to make decisions, the measure of value changes fundamentally from 'number of users' to 'uptime.' It is a shift from a model where you are charged only for the time humans are using it, to a model where the non-stop operation itself generates revenue.
AI is not just convenient software; it has entered a phase where it is designed as a factory—an 'AI factory'—and is being monetized as an AI factory.
Due to this paradigm shift, AI monetization has left the software context of 'advertising' or 'subscriptions' and landed on an extremely physical and robust profit structure, such as improving on-site yields and reducing costs.
3. The era of talking about 'servers' is over. From now on, we speak in terms of production lines called 'racks'
The evaluation criteria for hardware have also completely shifted from individual chip performance to 'integration at the rack level.' What NVIDIA's 'Rubin' and AMD's 'Helios' symbolize is the fact that AI is no longer a 'box' (server), but a vertically integrated 'production line' (rack).
NVIDIA Rubin (NVL72): Integrates 72 GPUs, 36 CPUs, and 6 types of chips including networking and DPUs at the rack level. The entire unit functions as 'a single compute domain.'
Physical reality: The NVL72 rack brought to the venue weighs as much as 2.5 tons including the water cooling system. AI is no longer 'light, thin, short, and small' software, but heavy industrial equipment.
KPIs of the new era: The metrics investors should focus on have shifted from 'top speed' to **'throughput per watt' and 'latency stability.'**
Since a 'rack equals a factory line,' individual speed is meaningless. Operational capability—the ability to produce stably at any time—is what creates true competitive advantage.
4. The keyword of the new era: 'Context Memory': Why the plumbing has been promoted to the lead role
Some in the market view the growing attention on memory and networking (the plumbing side) as a 'slowdown in GPU demand and sector rotation,' but that is a major misunderstanding.
The concept of 'context memory' introduced by NVIDIA is an essential design requirement for operating an AI factory in production. Here, 'context' refers not just to conversation history, but to all of the following:
Business context: Customer information, contract terms, internal rules, and inventory status.
Thought process: The 'intermediate progress' and 'reference grounds' when an AI agent makes multi-step decisions.
The more an AI factory runs 24/7 and performs complex tasks, the more important the 'plumbing' that retrieves this vast context at high speed becomes. In other words, the increase in demand for memory, cooling systems, and networking is not a replacement for GPUs, but a bullish signal that **'the GPU factory is finally ready for full-scale operation.'**
5. The true winners of 'Industrial AI': The triangle formed by Siemens, Palantir, and Oracle
In an industrial world where 'plausible lies (hallucinations)' are not tolerated, monetizing AI requires 'absolute correctness' rather than 'approximate answers.' To clear this high hurdle, the 'Industrial AI Triangle' formed by the following three companies is necessary.
Siemens (Digital Twin): Perfectly reproduces the physical site digitally, dramatically reducing downtime and defect rates through simulation. It governs the 'reality of the field' that directly impacts the P&L (Profit and Loss statement).
Palantir (Ontology): Applies common meaning (ontology) to disparate field data, aligning the coordinate axes of business operations. It provides 'governance and audit logs' of who decided what and when, functioning as an OS to hold AI accountable.
Oracle (System of Record): Holds the 'System of Record' for corporate data. In the industrial world, replacing existing core databases is virtually impossible due to the fatal risk of errors during migration. This **'high switching cost (fear)'** becomes the strongest moat in the AI era.
Rather than 'flashy chat,' it is these 'unassuming foundations' that are the key to elevating AI from a demonstration to an 'inescapable industrial infrastructure.'
6. Conclusion: We are standing on the main stage called 'Implementation'
What was shown at CES 2026 is the summary that AI is no longer a 'trend' swayed by the economy or sentiment, but a 'physical prerequisite' for solving industrial constraints.
The current market has been organized into the following robust three-layer structure.
AI Factory: A heavy-duty computing infrastructure that never stops, 24 hours a day.
Industrial OS: Operational design that clarifies meaning, governance, and accountability.
Physical AI: Implementation that eliminates waste on the front lines and directly rewrites the P&L.
Now that this structure is complete, the time for daydreaming about 'what AI can do' is over. From here on, a gritty yet certain phase of monetization begins: 'how quickly can we implement it and increase utilization rates?'
Is your business or investment judgment still stuck at the software level of 'chatbots'? Only those who face the reality that AI has begun operating as a '2.5-ton factory' will be able to grasp the next massive source of revenue.
I plan to add more insights and research findings as I go, or publish them as separate notes. I would appreciate it if you could 'like' or 'follow' this as a bookmark!
Sincerely, Kusunoura
P.S.
I also distribute related information via my email newsletter. I will also send out updates on my note posts.
https://www.techno-producer.com/ehatsumeijuku-tsushin/