Qinbafrank: AI Commercialization Has Only Just Begun! Watch Out for Anthropic Draining Liquidity After 11/9, as Crypto Waits for Liquidity to Return

Written and hosted by Mr. Z (@168MrZ)

The US economy is entering a new infrastructure cycle driven by AI CapEx, while AI commercialization has only just begun to cash in the compute invested over the past year; the real risk is not a Semiconductor pullback, but oil prices re-entrenching inflation, Cloud Revenue stopping its acceleration, and Anthropic's IPO starting to siphon liquidity out of the market

On October 29, 2026, 168X invited Qinbafrank (@qinbafrank) for the fourth time. The conversation started from the FOMC, the US-Iran war, oil prices and inflation, and ran all the way to Microsoft, Google, Anthropic, AI CapEx returns, Cloud, Agent and AI for Science, then on to Crypto's TGA liquidity, compliant financing and Tokenization.

Frank's core framework this time is not a simple bullish-or-bearish call on AI, but rather splitting the market into two lines that are happening at the same time: on the macro side, high oil prices and high long-end rates are still pushing up the valuation threshold for Risk Assets; on the industry side, AI Commercialization is still accelerating, and the CapEx poured in over the past year is gradually turning into Capacity that is going live, enterprise orders and Revenue.

So AI is not over, but it has entered the second half. Going forward it won't be all AI stocks rising together; instead the market has to start answering: Who actually turns CapEx into Revenue? Who owns the enterprise entry point? Who can take more profit once Tokens keep getting cheaper?

I. Why the US economy is still so strong: AI CapEx has become America's new infrastructure

Mr. Z: Let's start with the big picture. How do you see the US economy, inflation, and the Fed's rate pause this time?

Qinbafrank: I look at the US economy on two levels. The first is macro GDP, and I think there is no big problem there right now. Since the second half of 2025, the US has essentially entered the era of AI mega-infrastructure.

Qinbafrank: In 2025, large tech companies' CapEx was roughly $400–500 billion, and this year it may approach $900 billion. That means nearly $500 billion of incremental capital expenditure in a single year. That magnitude alone can provide very clear support to US GDP — roughly 1-2%.

Qinbafrank: This is actually a lot like China more than 10 years ago. Back then a big part of China's economic driver came from real estate and infrastructure; now the US has swapped in new Infrastructure, with the core being AI Datacenter, Power, Chip and Networking. As long as this round of CapEx Growth continues, it is hard for US GDP's aggregate growth to be particularly bad.

Qinbafrank: But the second level — looking at consumers, employment and how individuals actually feel — is a completely different story. Consumption among middle- and high-income groups is still strong, while middle- and low-income people are clearly weaker, and AI's replacement of Labor is becoming more and more obvious. Microsoft has been cutting jobs continuously over the past two years, and other large tech companies will keep doing similar things in the future.

Qinbafrank: So it is a very clear K-shaped Economy. AI, Semiconductor, Power and Datacenter industries are doing well, but people and industries unrelated to AI may feel something completely different.

II. The real problem with oil is no longer a lack of crude: it's the extra $30–40 per barrel of shipping cost

Mr. Z: You've consistently argued that oil prices are one of the most important variables for the next two months. Where exactly is the bottleneck now?

Qinbafrank: The important reason inflation has turned back up this year has always been energy, but the first half of the year and now are actually not the same logic.

Qinbafrank: The first half was more of a Supply Shock. After the Strait of Hormuz was blockaded, oil genuinely could not get out, so prices shot straight up. But since September, supply itself has recovered a lot.

Qinbafrank: For example, Saudi Arabia has a pipeline from the Persian Gulf to the Red Sea which, according to what I've seen, can carry about 6–7 million barrels a day; the UAE's pipeline bypassing Hormuz can supply roughly 3 million barrels. On the other hand, Middle Eastern producers have also started using VLCCs more — a single large tanker can load about 2 million barrels of crude, and then they pass through the strait concentrated in lower-risk time windows.

Qinbafrank: So by late September, the volume of crude leaving the Persian Gulf had actually recovered to close to pre-war levels. The problem is that volume came back, but cost did not.

Qinbafrank: Shipping a barrel of crude from the Persian Gulf to Asia used to cost maybe just $3 or $4; now it may be $35–40. When you see Brent above $110 today, a large part of that is not the price of Crude itself at all, but War Risk + Shipping Cost.

Qinbafrank: Because tankers have to pass through high-risk waters, shipowners bear the risk of being attacked, and insurance, escorts, lightering and transfers all get more expensive. So this has already turned from a Supply Shortage into Logistics Inflation.

III. Beyond Crude, watch Refinery: refined products are recovering more slowly than crude supply

Qinbafrank: And there is a second layer now, which is Refinery.

Qinbafrank: Many people only look at crude, but US consumers ultimately use Gasoline and Diesel, and industry actually uses various Refined Products. Some refineries in the Middle East have been damaged, and Russia and Ukraine kept attacking each other's refining facilities in August and September, so global Refining Capacity has also been weakened.

Qinbafrank: This means that even if the crude transportation problem is slowly solved, refined products won't necessarily come back right away.

Qinbafrank: That's why the US later discussed Diesel Supply with Russia, including hoping other countries would increase refined product exports — essentially because fixing Crude alone is not enough.

Qinbafrank: Oil Inflation now has two layers: the first is Shipping Cost, the second is the Refining Bottleneck. Until these two problems are solved, Energy Inflation will be very hard to truly eliminate.

IV. The Fed's pause is not a dovish turn: it is the hawks saying "let's wait a bit"

Mr. Z: The Fed didn't hike this time — should the market read that as dovish?

Qinbafrank: I don't think so. I'd rather understand it as Patience within the hawkish camp.

Qinbafrank: The Fed also knows that part of today's inflation is caused by war and energy. If US-Iran relations ease and oil prices fall on their own, then Inflation in November and December may improve accordingly.

Qinbafrank: So what it is doing now is: I won't rush to keep hiking, I'll look at a few more data points, and see whether this Oil Shock is Temporary or will slowly spread into broader, more persistent inflation.

Qinbafrank: If the Middle East eases and oil prices fall clearly, then in December it could even stay on hold again and keep watching how energy prices pass through to CPI and PCE.

Qinbafrank: But conversely, if oil prices stay high, and Energy Cost starts spreading from oil to chemicals and industrial goods, and then to more commodities, then it is no longer a One-off Shock and starts to become persistent.

Qinbafrank: In that case, the worst scenario is not "one more hike in December," but possibly having to keep hiking next year.

Qinbafrank: The market can actually digest 1 or 2 extra hikes, because you can calculate that. But if the expectation was 2 and it ends up being 3, 4 or even longer, then the entire valuation framework gets repriced.

V. The real window for the US and Iran is after the midterms: Trump is now stuck by his own hand

Mr. Z: So when is a real turning point between the US and Iran most likely?

Qinbafrank: I think the period after the midterm elections will be fairly key.

Qinbafrank: Right now Trump is somewhat trapped by his own position. Iran's system means it can endure economic pain for longer than the US political system can. The US has elections, public opinion and inflation pressure at home; the suffering of ordinary Iranians doesn't necessarily translate immediately into regime change.

Qinbafrank: So earlier the US hoped that tougher economic sanctions and blockade would make Iran crack first, but in reality Iran has held up better than expected.

Qinbafrank: That's the problem. Before the midterms, if Trump backs down directly, his political image suffers badly, because it amounts to launching a conflict without winning and then retreating first. But after the midterms, regardless of the result, his political concerns will be fewer.

Qinbafrank: At that point there may be two paths: either actually carry out another, more intense military operation; or realize that continuing to fight won't change anything, and start negotiating.

Qinbafrank: I actually think the latter isn't without a chance, because you can already see more and more third-party mediation. Pakistan, Saudi Arabia and Turkey are all getting involved, and China also has an incentive to push for de-escalation, because no one can afford another round of oil price spikes.

Qinbafrank: And signs of more direct communication between the US and Iran have actually begun to appear.

Qinbafrank: So I think the biggest disagreement between the two sides now may not even be the final terms, but who takes the first step. Does Iran loosen the Strait of Hormuz first, or does the US loosen sanctions and unfreeze assets first? The biggest problem for both sides is that trust has already been destroyed — each fears that if it steps back first, the other won't deliver.

VI. Warsh is remaking the Fed: the dot plot and Forward Guidance may matter less and less

Mr. Z: What do you make of Kevin Warsh's new framework for Fed communication and decision-making?

Qinbafrank: I think he genuinely wants to change things, not just put on a show.

Qinbafrank: The Fed has 8 policy meetings a year; his direction now is to reduce the number of meetings that actually discuss monetary policy, with adjustments starting next year and possibly further reductions after that.

Qinbafrank: More importantly, he clearly dislikes excessive Communication and doesn't like giving too much Forward Guidance. Things like the dot plot may also become less important in the future.

Qinbafrank: His thinking is more like: each time, genuinely judge based on the data at hand, rather than telling you 3 months in advance roughly what I'm going to do next.

Qinbafrank: He also set up several working groups after taking office, and I think those reports are what's really worth watching by year-end. I'm mainly looking at 3 questions.

Qinbafrank: First, will the Inflation Framework change? Why must it be 2%? The Powell era already changed it once, from a Point Target to an Average Inflation Target over a period, so whether there will be a new definition in the future matters a lot to me.

Qinbafrank: Second, how does AI Productivity enter monetary policy? In the long run AI itself is a deflationary force, because the same resources can produce more goods and services; but in the short run the AI mega-infrastructure buildout creates inflationary pressure in capital expenditure, materials and electricity. How exactly does the Fed balance short-term Inflationary CapEx against long-term Productivity Deflation?

Qinbafrank: Third, the Balance Sheet and the Reserve Regime. Since 2008 the US has built a very large balance sheet and an ample reserves system; will Warsh shrink the balance sheet again in the future and release more liquidity from the Fed system back to banks and the market?

Qinbafrank: If these things really change, the impact is far bigger than whether one FOMC meeting hikes 25bp or not.

VII. Stop fooling yourself with ARR: what's really valuable about Anthropic is Revenue that has already happened

Mr. Z: Let's move to AI. You've been saying lately that we should stop looking at ARR and look at real revenue instead — why?

Qinbafrank: Because ARR and Revenue that has already happened are simply not the same in terms of quality.

Qinbafrank: Anthropic's full-year revenue in 2025 was about $4.5 billion, and in the first half of 2026 about $16-odd billion. In Q3, according to figures previously disclosed by the media, it was about $16.8 billion for the single quarter.

Qinbafrank: That means its actual Revenue in the first three quarters of this year is already about $33 billion. If Q4 continues to grow versus Q3, then my earlier estimate of $50–60 billion of real Revenue for the full year is a discussable number.

Qinbafrank: That is completely different from "$100 billion ARR." ARR annualizes the Run-rate at a certain point in time; $50–60 billion is sales that actually happened this year.

Qinbafrank: So I'd say that once AI has developed to this stage, the ARR metric should gradually be pushed to the back and we should start looking at Quarterly Revenue, Annual Revenue, and real Cash Flow.

Qinbafrank: And at this pace, if Anthropic's Revenue passes $100 billion next year, I don't think that's an exaggeration. It previously said it would hit $200 billion by 2028, and at least now that no longer looks like a completely unimaginable number.

VIII. What really beat expectations at Microsoft isn't Azure's 48%: it's that enterprise customers are taking the baton from model labs

Mr. Z: Microsoft this quarter almost hit the upper end of your forward-looking framework — what do you value most?

Qinbafrank: I think Microsoft's earnings were indeed very good: Azure year-over-year growth reached 48%, Backlog was about $670 billion, of which roughly 30% is expected to be recognized in the next 12 months. Free Cash Flow was still $18.6 billion, which stands out sharply among the big cloud players.

Qinbafrank: But I think what really matters is not the 48%, but who is buying Azure.

Qinbafrank: Microsoft previously gave a data point: a large part of new orders are no longer from Frontier Model Companies like OpenAI and Anthropic, but from ordinary Enterprise Customers.

Qinbafrank: This time we also saw Foundry's business growing fast, with a large number of paying customers; Copilot's pay-as-you-go enterprise customers also continue to grow rapidly.

Qinbafrank: This signal is especially important, because in the past everyone worried: is this Cloud Demand just OpenAI borrowing a pile of money to buy Microsoft GPUs, Microsoft then investing in OpenAI, ultimately forming a Circular Trade that just cycles money back to itself?

Qinbafrank: Now more and more orders come from ordinary enterprises, which shows demand is genuinely spreading from model labs to Enterprise. That is real AI Commercialization.

IX. Cloud will become the AI Operating System: enterprises don't care whether it's Claude or DeepSeek, only results and cost

Qinbafrank: There's another point from the Microsoft call that I think is especially important.

Qinbafrank: Microsoft now has its own Models, while also connecting to OpenAI and deploying various Open Source Models. An analyst asked: how do users actually choose?

Qinbafrank: Satya Nadella's answer was essentially dynamic selection by Workload.

Qinbafrank: Enterprises don't Care whether the underlying layer today uses Claude, OpenAI, DeepSeek or Microsoft's own Model. Enterprises Care about only two things: first, can this task be completed stably and at high quality; second, how much does completing this task cost.

Qinbafrank: Complex tasks may call a Frontier Model; high-frequency, low-value, relatively mature work just switches straight to a cheap Open Source or in-house Model.

Qinbafrank: This is exactly what I said before — Cloud will ultimately become an AI Operating System.

Qinbafrank: It handles Model Routing, Data Connection, Permission Management, Security and Workflow. The underlying models can be swapped out, but the enterprise's entire system runs on your platform.

Qinbafrank: Especially since a lot of traditional companies are not Technology Companies and don't have the ability to build their own full Infrastructure. So the commoditization of Models doesn't mean Cloud gets bypassed — it may actually make Cloud more valuable.

X. Why Google both blew up and wasn't that bad: Cloud at 103%, the problem is really Search Monetization

Mr. Z: Google's results this quarter got a very poor market reaction — what's your take?

Qinbafrank: The headline was bad, but once you break it down it isn't that bad.

Qinbafrank: Google Cloud grew about 103% year over year in Q3, already more than $30 billion of Revenue for the quarter, and Operating Profit was also very high, with Margin still improving compared with Q2.

Qinbafrank: So Cloud itself is very strong; the problem is mainly the legacy core business. AI Search now has very high Usage, but Monetization hasn't fully kept up.

Qinbafrank: Previously, when users searched for something they'd see a lot of ads; now users talk directly to AI, and if after AI gives the answer there is no equally strong ad monetization, then you are Cannibalizing your own original Search Revenue.

Qinbafrank: The market also saw that its Free Cash Flow was very poor, but there were some one-off factors in there, such as an increase in TPU Inventory and Europe-related settlement expenses. Add those back, and actual FCF isn't as dramatic as the headline.

Qinbafrank: But the market is just that picky right now. No matter how strong your Cloud is, if the core business or FCF has flaws, I'll hit your valuation first.

XI. AI CapEx's "time lag" is starting to pay off: money spent in 2025 is only now becoming Revenue

Mr. Z: So is this finally the Time Lag you've been talking about starting to pay off?

Qinbafrank: Yes. This is very key — AI CapEx isn't spent today and turned into revenue tomorrow. Buying land, building Datacenters, building power facilities, ordering GPUs, installing Servers, and then actually going live all take a long time.

Qinbafrank: Large tech companies really started raising CapEx at scale around Q3 2025, so a lot of the Revenue growth at Microsoft and Google Cloud today actually corresponds to money spent a year ago or even earlier.

Qinbafrank: In other words, the Capacity that began accelerating construction in the second half of last year is only now entering its production phase, and for the past several quarters every cloud provider has been saying the same thing: Demand > Supply, not enough compute.

Qinbafrank: If demand is still growing and whatever new Capacity you bring online gets absorbed immediately, then Revenue will naturally keep accelerating in the coming quarters.

Qinbafrank: So I don't think commercialization is peaking now — it has only just begun to be delivered at scale.

XII. The real inflection point: when Cloud Revenue Growth starts to exceed CapEx Growth

Qinbafrank: But you can't just look at Revenue forever; the second stage is about ROIC.

Qinbafrank: 2026 is the year CapEx explodes. Last year it was hundreds of billions, this year it's close to doubling — that kind of CapEx Growth can't be 100% every year.

Qinbafrank: What would a truly beautiful inflection point look like? CapEx Growth starts to naturally decline in the future, while Cloud Revenue Growth keeps accelerating.

Qinbafrank: Take Google: Cloud Growth used to be only 40%, 50%, then into the 80s, and now above 100%. If its Revenue Growth starts to exceed CapEx Growth, then the math gets better and better.

Qinbafrank: The market also used to worry a lot about GPU Depreciation. People said GPUs become obsolete in two or three years, so Hyperscalers extending depreciation schedules is an Accounting Trick.

Qinbafrank: But the reality is that older-generation GPUs still have very strong economic value today, and some long-term contracts even extend the use of old cards further out.

Qinbafrank: If the useful life of servers, Datacenters and power infrastructure is longer than the market originally expected — Chips can be replaced while the Facility keeps being used — then the real payback period for the whole CapEx cycle also improves.

Qinbafrank: So over the next few quarters, what I most want to see is: how Revenue Growth, CapEx Growth, Depreciation Life and FCF start to converge.

XIII. Coding isn't AI's only use case: Enterprise, Agent and AI for Science are all picking up the baton at once

Qinbafrank: Another market worry earlier was: besides Coding, what else does AI have?

Qinbafrank: Why did Coding explode first? Because it's highly standardized, has giant Repositories like GitHub, and has a clear Feedback Loop, so Models can improve quickly.

Qinbafrank: But the second stage is not "finding another single use case as big as Coding" — it's many use cases exploding together.

Qinbafrank: The first category is Horizontal and Vertical Workflow inside enterprises: Finance, Legal, Tax, Marketing, HR and all kinds of enterprise processes. Individually these may not be as big as Coding, but together they're enormous.

Qinbafrank: And as Coding itself gets stronger, it also paves the way for these use cases, because a lot of enterprise processes ultimately need Code to Integrate.

Qinbafrank: The second is Agent. Since September, the major companies have all been pushing their own Agents. Personal Agents and Enterprise Agents will both keep increasing Token Consumption.

Qinbafrank: The third is high-value use cases like AI for Science, AIDD (AI-Powered Drug Discovery) and Drug Discovery. OpenAI, Anthropic, Google and ByteDance are all investing into Bio and Life Science.

Qinbafrank: So it's not that there's no second use case after Coding; rather, the second use case has already split into many different use cases appearing at the same time.

XIV. How to trade Anthropic around 11/9: before the Roadshow is a window, after the Roadshow it starts draining

Mr. Z: What everyone cares about most now is probably the Anthropic IPO. Why do you see around November 9 as a dividing line?

Qinbafrank: November 9 is a date I worked out by walking through the SEC process; it is not something the company has officially announced.

Qinbafrank: After Anthropic publishes its S-1, under the rules it must pass a certain number of calendar days at the earliest before it can start the Roadshow. Counting it out, that lands around November 9.

Qinbafrank: Before that, I think the market may still have a relatively comfortable window.

Qinbafrank: There are a few reasons. First, the market has already fallen for several weeks and a lot of sentiment has been released; second, Microsoft and Google's Cloud Revenue at least proves that AI Commercialization hasn't broken; third, if Amazon's earnings also show no obvious problems, fundamentals will give the market some support; fourth, there is also seasonality around the midterms themselves.

Qinbafrank: So before the Roadshow, the market may see a rebound first. What really needs to be watched is when institutions start seriously preparing their subscriptions.

XV. Why even a great IPO can crush the market: $400 billion of subscription money can't appear out of thin air

Qinbafrank: A lot of people don't understand what "draining" means. Draining is not saying Anthropic is bad — it may be precisely because it's so good that every institution wants it.

Qinbafrank: Let me use the SpaceX situation at the time as an analogy. Subscription demand for a big IPO usually ends up being several times the actual deal size.

Qinbafrank: If Anthropic ends up being a large deal on the order of a hundred billion dollars, the market may need to prepare several hundred billion dollars of subscription money.

Qinbafrank: The problem is that institutions don't just keep several hundred billion dollars of Cash sitting in their accounts waiting for it to list. So where does the money come from? Selling positions.

Qinbafrank: Sell some Semi, sell Momentum, sell Software, sell other AI Positions, free up Cash, and get ready to subscribe. So the real Liquidity Drain tends to appear after the Roadshow begins, because institutions start formally doing Portfolio Adjustment.

Qinbafrank: By IPO day, some of the money that didn't get an allocation or got locked up is released, and the situation changes again; but after listing there is a second layer of draining, namely retail and other capital thinking: since there is now the purest Frontier AI stock, why should I still buy indirectly through the Supply Chain?

Qinbafrank: So Anthropic's listing can be very bullish for AI in the long run, but in the short run it can absolutely drain the entire AI sector first.

XVI. The AI main line has already shifted: Memory → Optical / Software → Agent / AI for Science

Qinbafrank: There's one more thing I keep wanting to stress: don't fall into the fixed mindset that "AI = Semiconductor = Memory," because the AI main line shifts at every stage.

Qinbafrank: In Q1 and Q2 this year, Memory was the strongest; after August and September, Optical started to strengthen, and at the same time a batch of software and SaaS names that had been beaten down badly also started to come back.

Qinbafrank: Why? Because as Models get commoditized and Token Price comes down, the beneficiaries are not only upstream Hardware.

Qinbafrank: The cheaper the Token, the easier the Deployment, and Cloud, Software, Enterprise Integration and Agent actually benefit more and more. The next stage may be AI for Science and Bio.

Qinbafrank: So what trend investing really requires is not clinging to the previous Winner, but continuously identifying: where is the next layer of the profit pool shifting to?

Qinbafrank: It's not that Memory is finished, and Semi isn't dead either — it's just that the phase like the first half of the year, when everything went straight up, may be over in the short term.

XVII. In a high-yield era there is no more broad rally: from now on the market only rewards real Cash Flow

Qinbafrank: There's also a bigger backdrop now: we have entered an era of relatively high yields.

Qinbafrank: Oil prices push Inflation higher, fiscal deficits mean a lot of Treasury issuance, and AI Infrastructure itself needs to issue a lot of Corporate Debt. All of these are competing for money in the bond market at the same time.

Qinbafrank: When Bond Yields are high, stocks naturally have to offer a higher Risk Premium too — that's why the market is getting pickier and pickier.

Qinbafrank: If Microsoft's earnings have no obvious flaws, the market gives you positive feedback; Google Cloud is clearly good, but if your Search and FCF have problems, the market still sells first.

Qinbafrank: So when picking stocks from here on, you must clearly distinguish: which companies are just Narrative, and which companies' businesses have genuinely delivered — Revenue is real, Cash Flow is real, and the growth room for the next 2-3 years can still be calculated.

Qinbafrank: If a big company gets its valuation hit because of short-term headwinds, I actually think that's a good thing, because it gives you a price to get back on board. High rates don't mean no stock can rise; it means the market will increasingly "pick the best of the best."

XVIII. What to really watch in 2027 Q1 isn't Semi, but whether AI Commercialization decelerates for consecutive quarters

Mr. Z: Looking ahead to 2027 Q1, what's the judgment you're most confident about?

Qinbafrank: The judgment I can make is that at least through Q1 of next year, I think AI Commercialization is still accelerating and hasn't reached Saturation.

Qinbafrank: Because more and more enterprises are starting to implement AI, and everyone's Token Consumption will ultimately show up in Compute, and a large part of Compute is held by Cloud Providers. So at least at the Cloud layer, I think growth is still there.

Qinbafrank: When would I really start worrying about the AI Thesis? Not when Semi drops 20%, and not when some stock drops 30%. It's when you see Cloud Business Growth Rate at Microsoft, Google and AWS clearly falling for two consecutive quarters.

Qinbafrank: That's when you have to ask: have all the enterprises that should use AI already used it? Have users spent about as much as they're willing to spend? Are the Workflows that can create economic value starting to Saturate?

Qinbafrank: That is the origin point of the whole AI Story. As long as people are still using it, enterprises are still paying, and Token Consumption is still growing, the big picture isn't broken.

XIX. Why Crypto fell in October: the Narrative didn't die, the TGA is draining liquidity

Mr. Z: Let's finish with Crypto. It had a surge in September and then fell back in October — what's your read?

Qinbafrank: On the Crypto side I think it's actually clearer — it's Liquidity. When we talked at the end of August, my framework was that late August through mid-to-late September would be better, and late September into October would see a pullback because liquidity would start tightening.

Qinbafrank: That's basically how it has played out. TGA (Treasury General Account) went from roughly $800–900 billion in August all the way up, and now it's above $1 trillion — essentially the Treasury pulling Cash out of the market.

Qinbafrank: So Crypto being weak in October is not because all the Narratives suddenly died, but because the Money Supply environment is wrong.

Qinbafrank: What to look at next is very simple: will the TGA start releasing in November? If the Treasury starts putting Cash back into the system and market Liquidity improves, then Crypto may naturally restart.

Qinbafrank: The framework I gave before was that liquidity would gradually bottom around late October, and if it is released again in November, the coin market has a chance to turn around with it.

XX. Crypto's biggest main line for the next stage: compliant ICO + Tokenization, TradFi and Crypto are converging

Qinbafrank: On regulation, I think things actually are changing — it's just that a lot of it takes time to land.

Qinbafrank: The first line is Crypto Financing. In the past, US companies raising money either did an IPO, or a Regulation D private placement, or used exemption channels like Regulation A and Regulation CF. But Crypto Projects couldn't raise money from the US public normally for a long time, so everyone running a Token Sale blocked Americans out.

Qinbafrank: Now regulators are studying a new digital asset financing exemption — essentially giving Crypto Projects a path to legally raise money from the public in the future.

Qinbafrank: This won't land immediately, because there's still public comment and the regulatory process, but if it is actually implemented, I think it can be understood as a new kind of Regulated ICO.

Qinbafrank: The second line is Tokenization. Even if the CLARITY Act doesn't advance smoothly at the congressional level, that doesn't mean regulators can't do anything. The SEC and CFTC, within their own authority, can still push rules on Custody, Tokenized Securities, DeFi, AMM, Derivatives and so on.

Qinbafrank: I think the Tokenization trend is already irreversible. More and more Treasury, Stock and Security will enter the On-chain Financial Market.

Qinbafrank: So what really benefits in the long run is not just "tokenized stocks" themselves, but the underlying Infrastructure, Settlement, Custody, Liquidity and DeFi Rail. The eventual direction of Crypto and TradFi is not to replace each other, but to gradually become the same financial market.

Mr. Z: To close this episode, let me condense Qinbafrank's framework into 6 judgments.

1. The US economy is not in a full-blown recession for now; AI CapEx has become America's new infrastructure investment, but the K-shaped divergence will keep widening as AI replaces labor.

2. The most critical macro variable in the next 1–2 months is still Oil, and it has now evolved from a simple crude supply issue into a dual problem of Shipping Cost + Refinery Capacity. Whether the US and Iran can enter genuine negotiations after the midterms directly determines inflation and the Fed's next step.

3. AI Commercialization is not peaking; the time lag of the past year's CapEx is starting to pay off. Microsoft, Google and Anthropic's Revenue all prove that enterprise Adoption is still deepening, and the most critical next question is when Cloud Revenue Growth truly exceeds CapEx Growth.

4. AI has entered the second half of the industry cycle; Memory and Semi won't disappear, but the main line will keep shifting toward Optical, Cloud, Software, Agent, Enterprise Deployment and AI for Science. In a high-rate environment, the market will only reward real Earnings.

5. Before the Anthropic Roadshow, the market may still have a window of sentiment repair, but once the Roadshow starts, large-scale institutional subscriptions will require selling other positions to free up Cash, and the short-term Liquidity Drain is worth being highly alert to.

6. October's Crypto correction is essentially still liquidity contraction; whether the TGA releases in November is the most direct Trigger, while the longer-term structural main line is compliant financing and Tokenization pushing TradFi Assets fully into On-chain Finance.

The signal that would really turn the AI Thesis bearish is not Semi falling another 20%, but Cloud Commercialization clearly decelerating for several consecutive quarters. Thanks to Qinbafrank for coming to 168X for the fourth time!