Market Fluctuations: Why He Insists on Being 'Present'? 5 Key Judgments for AI Investment

Written by · Hosted by Mr. Z (++@168MrZ++), 168X

Frontier AI is transitioning from selling subscriptions, APIs, and tokens to directly selling research and commercial results; the arms race in computing power will not stop due to a temporary slowdown in ARR, but Alpha is shifting from 'all AI is rising' to focusing on storage, fabrication, equipment, and biotechnology.

On October 28, 2026, 168X once again invited Rick (++@rickawsb++), discussing everything from the Anthropic IPO to Frontier Models, AI for Science, Recursive Self-Improvement, Hyperscaler CapEx, semiconductors, memory, Neocloud, and the new risks facing crypto.

Rick's core judgment this time goes beyond simply "continuing to long AI": the market is using an outdated software valuation framework to price a potentially entirely different commodity. If the strongest models no longer rely on monthly $200 subscriptions or API tokens for revenue, but instead directly create new drugs, new materials, chip designs, or even scientific discoveries, then the ARR that the market is currently focused on may not represent the ultimate core value of Frontier AI.

This has also changed his asset choices. Rick has reduced his holdings in NVIDIA significantly, preferring companies that possess manufacturing, certification, equipment, and non-compressible time barriers; storage remains his largest position, followed by chips, biotech, and equipment. In contrast, he is more optimistic about the Agentic Economy on Crypto than ever before, but due to AI's acceleration in mathematics, cryptography, and quantum computing, he has chosen to clear out most of his BTC holdings first.

1. Anthropic IPO Severely Underestimated: The Real Exaggeration is Not Revenue, but Capital Efficiency

Mr. Z: What was your biggest impression after the Anthropic IPO?

Rick: I am very optimistic about Anthropic. If it is truly valued at the level currently discussed in the market, I personally think it is severely underestimated, and I would buy it.

Rick: At first glance, people will see its revenue growth is rapid, while GAAP still shows losses, and they will think the costs are high. However, a significant portion of this is due to past financing, debt-to-equity conversions, and equity appreciation, which does not fully reflect its actual operating spending.

Rick: What really matters is how much actual money has been invested relative to current revenue, ARR, and growth rate. From a startup perspective, I believe this kind of capital efficiency growth is unprecedented.

Rick: Therefore, I won't just look at "how much it is still losing." The real question should be: how much does it need to burn to create one dollar of new intelligence, revenue, and future earnings?

2. ARR May Start to Become Irrelevant: The Next Step for Frontier AI is Not Selling Tokens, but Selling Results

Mr. Z: After Anthropic's listing, the market can finally see real revenue, profit, and cash flow. What metrics will you value the most in the future?

Rick: I actually have a rather radical view. Since August, I have been saying that in the future, ARR may not be as important as everyone thinks, and even subscriptions themselves may not be that important.

Rick: The reason is simple. When model capabilities are not strong enough, OpenAI and Anthropic can only sell subscriptions for $20 or $200 a month, or sell API tokens. But if the models begin to truly solve cutting-edge problems in mathematics, physics, materials, and biology, why would these companies still care about the small subscription fees from ordinary users?

Rick: If they can truly develop new cancer drugs, new molecules that extend life, new materials, or even new room-temperature superconductors, the amount you would be willing to pay for those results is certainly not just $200 a month.

Rick: In the early days of the electricity revolution, power companies did not sell "electricity" to consumers, but rather light. AI will ultimately be the same. It will not always sell tokens; it will start selling outcomes.

Rick: Perhaps in six months to a year, the strongest model companies will discuss token prices less and less, and more and more will directly deliver results. New drug formulas will be sold to pharmaceutical companies, new materials to manufacturers, and scientific discoveries will be directly monetized.

Rick: If this direction holds, the market's current pricing of Frontier Intelligence using SaaS ARR multiples may itself be incorrect.

3. The Strongest Models May Not Be Open at All: The Gap Between Closed Source and Open Source is Widening

Mr. Z: But on the other hand, China's open source is catching up quickly. Can the lead of closed source really be sustained?

Rick: I now feel that the gap between Frontier Closed Source and Open Source may actually be widening.

Rick: The public models that everyone sees are just what can be used externally. The models that OpenAI uses internally to solve cutting-edge mathematical problems have not even been seen in the market yet. Why would they need to release them? If this model can create extremely high value directly, there is no need to open the API for competitors to distill and learn from it.

Rick: Open Source is still strong, but you will find that many open-source models are still competing in benchmarks, coding, and SWE-bench, while the strongest closed-source models have already begun tackling the most cutting-edge issues in mathematics, physics, materials, and life sciences.

Rick: So the so-called "pacing the frontier"—I would even be willing to make a rather harsh interpretation: Frontier Labs continues to push forward, but everyone forms a kind of tacit understanding that the truly strongest models may not necessarily be open to the public.

Rick: If the strongest models in the future do not even sell APIs, the revenue cannibalization from open source may not be as direct as everyone imagines, because what both sides are selling is no longer the same product.

4. The True Endgame of RSI: After Models Iterating Models, Chips, Fab, and Energy, Leaders May Forever Stay Ahead

Mr. Z: You have been talking about Recursive Self-Improvement lately. What does it mean to truly enter RSI?

Rick: My framework has always been three layers.

Rick: First, models consume applications. As model capabilities become stronger, many so-called AI applications may eventually turn into model features.

Rick: Second, computing power consumes models. Companies without sufficient computing power will find it hard to remain at the frontier. Because for models to achieve recursive self-improvement, they inherently require more and more compute.

Rick: The third and most critical layer is the formation of a complete closed loop: models begin to iterate the next generation of models, while models also help design chips, and further optimize fabs and energy infrastructure.

Rick: Once this loop is formed, a very frightening positive feedback loop will emerge. Leading in computing power leads to leading in models, and leading in models allows for faster chip design and faster construction of computing power, with new computing power continuing to widen the gap in models. Leading will create more leading.

Rick: After that point, latecomers may not just be "12 months behind," but may almost never catch up.

Rick: Musk is currently working hard on compute, chips, energy, and models, essentially hoping to complete a curveball before this complete closed loop is formed.

5. Why Has Microsoft Laid Off 6 Times in Two Years? Hyperscalers are Cutting All Costs Just to Preserve AI Capacity

Mr. Z: But the market is currently very worried about CapEx. How do you view Microsoft's situation of laying off employees while continuing to burn computing power?

Rick: I actually think this is particularly worth watching. Microsoft has already gone through multiple rounds of layoffs in the past two years. Normally, if a large company knows it needs to cut 20% over the next three years, it would tend to do it all at once, because small, frequent layoffs can actually be more damaging to morale.

Rick: Why is Microsoft laying off repeatedly? My non-professional understanding is: the demand for capacity growth has reached a point where the original budget can never keep up with the changes.

Rick: Therefore, what the CFO needs to do is cut employees, cut other business expenses, and allocate more resources to AI data centers and compute.

Rick: This is also why I am not too worried that a short-term slowdown in ARR will cause hyperscalers to stop CapEx. This is an arms race that no one dares to lose. You can go all out and still lose, but if you don’t go all out, you will likely be eliminated.

Rick: The real challenge for hyperscalers moving forward is how to maintain balance between debt, equity dilution, cost-cutting, and capacity expansion while being as aggressive as possible.

6. Will Anthropic's IPO Drain Semiconductor Resources? Short-term Possibly, Long-term IPO Money Will Flow Back to Compute

Mr. Z: As one of the first truly pure-play frontier model stocks to go public, will Anthropic directly siphon off the AI premium from semiconductors?

Rick: It’s possible. In the past few years, why have people been buying memory, optical, GPU, and equipment? Because the public market did not have a truly pure frontier model stock, so everyone could only buy AI exposure through the supply chain.

Rick: If Anthropic goes public, it may become the purest AI asset, and in the short term, it could indeed siphon off some funds from hardware.

Rick: But we must also look at the other side. What will the money raised from the Anthropic IPO be used for? Still to buy computing power.

Rick: So this is a very interesting two-way relationship. On one hand, it may siphon off the Valuation Premium from the Supply Chain, while on the other hand, the money it raises will eventually turn back into orders for GPUs, Memory, Networking, and Datacenters.

Rick: I don't know how short-term funds will flow, but from the perspective of the entire industry chain, I don't think Anthropic's IPO is purely a negative event for Hardware.

7. The Next Wave of AI is Not Chatbots: Bio, Materials, and Energy are the True Earnings Hotspots

Mr. Z: If the next round of AI stocks is not driven by Narrative but by Earnings, where will profits emerge the fastest?

Rick: I think one increasingly clear direction is Bio.

Rick: Why are companies like OpenAI and Anthropic starting to collaborate with life sciences companies? Because when models enter mathematics, physics, and scientific research, the most direct economic value downstream is not just "helping you write emails," but helping design proteins, drugs, molecules, and experiments.

Rick: Many growth stocks in Bio have actually seen a rally in the past two to three months, and there is a reason for that.

Rick: The significance of AI for Science lies here. You are not just making a person work 20% faster; you are starting to create discoveries that did not exist before.

Rick: And this won't stop at Bio. There are also Materials, Energy, Nuclear Fusion, Quantum, and Superconductors. What AI ultimately brings is not an industrial revolution, but a machine that continuously creates new industrial revolutions.

Rick: In the past, a technology cycle might last decades; now, as intelligence accelerates, all technology maturity curves will be compressed. The market will experience more corrections in the future, but every time the next round of technological revolution restarts, the rise may come faster than in the past.

8. The Biggest Side Effect of AI May Be Interest Rates: 10Y Heading Above 6%, Real Estate and Emerging Markets Under Pressure First

Mr. Z: Are your views on future interest rates more extreme than the market?

Rick: Yes. I think many people are only discussing how much Earnings AI will create, but they haven't seriously considered how much AI Infrastructure will actually cost.

Rick: Hyperscalers, Neocloud, Datacenter, Energy, and Semiconductors all require significant financing. This is a new capital demand that did not exist before.

Rick: If the AI industry continues to expand Debt Issuance in the future, then long-term interest rates may not just be Higher for Longer, but Much Higher for Longer.

Rick: I wouldn't be surprised if the 10-year U.S. Treasury yield reaches 6%; I wouldn't think 7% or 8% is completely impossible either. Of course, this is a very extreme scenario, but the logic behind it is simple: the new capital demand far exceeds the past.

Rick: This will lead to two very important results. First, emerging markets will face tremendous pressure. When U.S. risk-free assets can provide extremely high returns, why would global dollars remain in riskier markets? Historically, when the 10Y was at high levels, Southeast Asia, Latin America, and peripheral European markets experienced significant capital pressure.

Rick: Second, real estate may be the biggest loser. On one side, there are high interest rates, and on the other, future AI Agents may impact a large number of white-collar jobs, leading to high Financing Costs and increased unemployment, which puts double pressure on real estate.

9. The Market is Transitioning from K-Type to I-Type: In the Future, Only the Strongest Assets in AI Will Be Worth Holding

Mr. Z: So if long-term interest rates remain high, what stocks can we still buy?

Rick: I have always talked about a K-type market, but my current view may be even more extreme. What comes next may not be K-type, but I-type.

Rick: What does that mean? It's not just a differentiation between AI and non-AI, but even within AI, there will be further stratification.

Rick: The truly strongest assets can still be held, but if an AI company's Moat is not strong enough, then even if it has an AI label, I would not want to touch it.

Rick: In the future, you can no longer buy stocks with the logic of "as long as the AI Supply Chain goes up, everything will rise."

Rick: As intelligence becomes stronger, it will start to consume Software, Applications, and eventually Chip Design. Every layer of the value chain must be re-evaluated: Is your moat something that AI itself cannot quickly replicate?

10. Why Rick Has Sold Off Most of His NVIDIA: The CUDA Moat is Narrowing, What is Truly Hard to Replace is Fab and Equipment

Mr. Z: Do you now feel that even NVIDIA's moat is not as wide as it used to be?

Rick: I sold off most of my NVIDIA two months ago. I still have some left, but a big reason is that part of the stock has not been held for a year, so there is no tax reason to move it now.

Rick: I'm not saying NVIDIA is bad, but I feel it is no longer the best Risk/Reward.

Rick: One of NVIDIA's biggest Moats in the past was CUDA. But today, you will constantly see teams developing CUDA Alternatives; Google, Amazon, OpenAI, and even Frontier Labs are all working on Chips.

Rick: More importantly, when models can start helping design the next generation of chips, the development cycle of Chip Design itself will continue to shorten.

Rick: So I would ask: If in the future Frontier Model companies can design chips faster and faster, how wide is the moat of a pure Chip Design Company?

Rick: In contrast, I prefer Fab, core equipment, and manufacturing processes.

Rick: TSMC, Intel, ASML, and many companies in production and testing equipment have moats that are not just technical, but also incompressible time.

Rick: Once you enter the certification of a production line, customers will not switch you out just because someone offers a 5% discount tomorrow, because the costs and potential losses of re-certification and re-validation are too high. AI can compress Software time, but it cannot temporarily compress Fab Qualification time; this is what I see as a more reliable moat.

11. How to Buy Optical/Equipment: COHR, VECO, AMAT, KLAC, FORM, TER Start to Enter Watchlist

Mr. Z: If we break down from Fab to upstream and downstream, what are you currently looking at?

Rick: In Optical, I have recently bought Coherent (COHR).

Rick: I also bought Veeco (VECO), which makes MOCVD and other equipment, which can be understood as part of the upstream production equipment in Optical. I see the logic of these companies not as suddenly exploding 50% tomorrow, but as having insufficient capacity, a limited number of suppliers, and once customers introduce the equipment, they will not easily switch.

Rick: In Testing, Packaging, and Semicap, I will look at Applied Materials, KLA, Amkor, FormFactor, and Teradyne.

Rick: The biggest advantage of these companies is that many Processes, Qualifications, and Equipment require long-term validation. No matter how smart the model is, it cannot be directly replaced on an operating production line the day after it is designed.

Rick: So I am increasingly fond of this Time Moat. The market is looking for "what AI can do," but I am more interested in finding "what AI cannot temporarily compress time for."

12. Storage Remains the Largest Position: MU, SK Hynix, Samsung are Rick's Most Asymmetric Bets

Mr. Z: You have lost quite a bit on Storage this round, but you still haven't moved?

Rick: Storage has always been my largest position, and I have basically not moved the underlying stocks. I have indeed lost quite a bit on options this round, but I am slowly looking for opportunities again.

Rick: My core judgment has not changed. I have positions in Micron, SK Hynix, and Samsung.

Rick: The most common Bear Case in the market now is that "Memory is a Cycle, and after rising too much, it will definitely go down." But the problem is that today's Demand Curve is different from the past.

Rick: What you are seeing now is that a lot of Capacity for 2028 and 2029 has already been locked in early, and the actual prices of Memory and GPUs in the market are still rising.

Rick: More importantly, as model capabilities improve, the same unit of Compute can create more Intelligence, and more Intelligence can create more economic value.

Rick: If the same card can help you do more Coding, more Video, and more Drug Discovery because of model upgrades, then why should this card necessarily become cheaper?

Rick: So I think the market is making a big mistake by valuing Memory as a traditional Commodity Cycle.

Rick: For me, this is a very asymmetric Bet. If the market is right, I lose my position; if the market is wrong and intelligent value continues to accelerate, I could earn far more.

13. Has Neocloud Been Wrongly Punished? The Market Only Sees Debt but Does Not Price Power Contracts

Mr. Z: The biggest concern about Neocloud recently has been high Debt and high interest rates, and some have even started discussing bankruptcy. What do you think?

Rick: I think the market's FUD in this area may also be excessive.

Rick: Everyone only looks at the balance sheet, saying these companies have high Debt and financing rates are getting more expensive. But much of the debt that has already been borrowed has a locked-in interest rate, and the income contracts and electricity costs are also locked in.

Rick: The real issue is not the old debts, but whether it wants to continue expanding with expensive funding in the future. If a management team borrows under a financing cost of 15%, knowing that the project returns won't be high enough, then they should definitely go bankrupt. But I don't think a normal management team would do that.

Rick: What is often overlooked is the Power Contract. These companies have already signed Power Pipelines that have not yet been put into production, which may represent a scarce resource that is not fully reflected in accounting assets, and the price of Datacenter Power continues to rise.

Rick: Therefore, the market only focuses on Debt but does not price Power Pipeline, which I think may be a mistake. Neocloud's most important asset may not be the GPU, but the electricity it locked in early.

14. Long-term Most Bullish Crypto, Why Clearing BTC Now: AI, P vs NP, Quantum and Q-Day

Mr. Z: Your changes in Crypto this time are significant. How are you allocating now?

Rick: Long-term, I am actually more bullish on Crypto than ever before.

Rick: I believe that the future Agentic Economy will be difficult to establish on today's fiat currency system; it will likely require Crypto-native Settlement, Ownership, and Machine-to-machine Economy. From this perspective, I even think the long-term beneficiary is more likely to be Ethereum rather than Bitcoin.

Rick: But the problem is that in the short term, the risks I see are too great, so I have already cleared most of my BTC. The biggest risk is not the Fed, but the acceleration of AI on mathematics and Cryptography.

Rick: The security of many modern encryption systems is fundamentally based on the assumption that certain mathematical problems are "computationally very difficult." If AI starts to rapidly advance Complexity Theory, P vs NP-related issues, and even further accelerate Quantum Computing, the timelines for these security assumptions that everyone thought were far off may suddenly be compressed.

Rick: I'm not saying Bitcoin will definitely be cracked tomorrow. The real danger is when the market starts to believe it "might be cracked." BTC's market cap is large, but the actual liquidity available for trading relative to its market cap is not that large. As long as a small portion of long-term holders start selling due to Q-Day, Quantum, or Cryptography FUD, it could create a very large drawdown.

Rick: So my logic is that while I remain very optimistic about Crypto in the long term, I want to avoid this period of high uncertainty first. The future outlook could be very bright, but before taking off, it may first dip very deep.

15. Q4 2026 to Q1 2027: Don't Guess V, U, or W, the Biggest Risk is Not Being on the Bus When the Reversal Happens

Mr. Z: To be direct, how will the market move from Q4 2026 to Q1 2027?

Rick: My short-term judgments have been wildly off. Last time I was guessing whether the Semiconductor would be a V or U recovery, but now it seems it could be a W, or even several more Ws.

Rick: But that doesn't matter to me; my strategy has always been Time in the Market, not Timing the Market.

Rick: As long as the basic thesis has not reversed, I would rather endure a 20% or 30% drawdown than risk being off the bus when the real reversal happens for the sake of gaining that extra 10% or 20% in a swing.

Rick: Similar things actually happened at the end of last year. At that time, everyone thought Hyperscaler CapEx was too high, and AI had no applications beyond Prompt, but after the Coding Agent came out, the entire market accelerated again.

Rick: Now the market says there is no second use case beyond Coding, but we are already starting to see AI for Science. The growth rate of model capabilities this time is even faster than last year, so I am even less willing to bet that I can accurately time the bottom.

Rick: If you miss the most violent month of a Semiconductor, you might miss 70% of the gains. Storage is the same; if you weren't on board during the craziest two months, you might lose most of the annual gains.

Rick: So I will take the spirit of holding coins from before to invest in AI. My current order of positions is very simple: first is Storage, second is Chip, third is Bio, and fourth is Equipment.

Rick: The market may continue to fall, it could be V, U, or W, it doesn't matter. What I fear more than a further 30% drop is not having a position when the Singularity Curve accelerates again.

Mr. Z: To wrap up this episode, I will condense Rick's framework into five judgments. First, the biggest significance of the Anthropic IPO is not that there is one more AI stock, but that the market can directly price Frontier Intelligence for the first time. However, measuring it today with ARR, Subscription, and Token Revenue may still severely underestimate its ultimate value; second, the real RSI is not a model writing a few lines of code by itself, but rather the closed loop completed after Model → Chip → Fab → Energy, leading to continuous production of more leading technologies; third, AI Trade has entered type I differentiation, with Rick reducing his holdings in NVIDIA and leaning more towards assets like Storage, Fab, Equipment, Optical, and Bio that have real capacity, certification, and time moat; fourth, the enormous financing needs of AI Infrastructure may push long-term interest rates much higher for longer, putting greater pressure on real estate, emerging markets, and weak assets; fifth, Rick is extremely bullish on Crypto and the Agentic Economy in the long term, but due to AI accelerating mathematics, Cryptography, and Quantum, he chooses to clear BTC in the short term to avoid the uncertainty of Q-Day. By Q4 2026 and Q1 2027, he is not preparing to guess V, U, or W, but to continue Time in the Market, waiting for the next round of intelligent acceleration to push Earnings and valuations up together. Thank you, Rick, for coming back to 168X!