AI is in a Credit Expansion Phase: The Stronger AI Becomes, the More the Federal Reserve Needs to Cut Rates

Author: Mr. Z, 168X War Room

The ultimate mission of the Federal Reserve is to preserve U.S. Treasury bonds; the stronger AI becomes, the closer we are to rate cuts, while the bubble is still far off.

On Monday, August 24, 2026, at 6 PM, the market opened with a bang. Long-term interest rates cannot be suppressed, while two pumps are simultaneously running: tech giants are heavily investing in AI CapEx, and the U.S. Treasury is frantically issuing bonds at the same time. AI CapEx and U.S. Treasury bonds are competing for the same pool of liquidity. Is the Treasury expanding its repurchase program to save liquidity or to prolong the life of U.S. Treasury bonds?

In this episode of the 168X War Room, we welcome back Tiezhu (@TiezhuCrypto), who has a background in policy research, has worked in traditional financial institutions, and loves to understand a country's macro transactions through its balance sheet. In this episode, he makes a strong statement: the ultimate mission of the Federal Reserve is not inflation or employment, but to preserve U.S. Treasury bonds. If the country goes bankrupt, will you still consider inflation? Cutting rates can reduce the peaks of inflation but cannot lower the water level of inflation. Rate hikes are impossible by the end of the year, with a 60% probability of rate cuts. The stronger AI becomes, the more the Federal Reserve should cut rates because high interest rates cannot contain AI but will kill all other industries. The emergence of SPV and GPU financing signifies that AI has officially entered a credit expansion phase, but the bubble phase has not yet arrived. The only message for investors is this: various analyses have been smoothed out by AI, and what remains as the decisive factor is position and mindset!

1. The Truth of Independence: Adults are Independent, but Decisions are Never Independent

Mr. Z: Today is Monday, August 24, at 6 PM. Welcome back to the 168X War Room. We are honored to have Tiezhu back after three months. Today, we want to discuss several topics: Does the stronger AI become, the more the Federal Reserve needs to cut rates? Alibaba also announced this weekend that it will raise over $10 billion in the Hong Kong capital market for AI CapEx. The U.S. Treasury is simultaneously issuing bonds like crazy. Why can't long-term interest rates be suppressed? Is the Treasury expanding its repurchase program to save liquidity or to prolong the life of U.S. Treasury bonds? On the other hand, AI has already moved towards project financing, SPV, GPU financing, and private credit. Let's start with the Federal Reserve: What does the independence of the Fed really mean? This is a bit of a paradox.

Tiezhu: That's a good question; few people discuss it specifically. The independence of the Federal Reserve refers to a system arrangement: the Fed can execute monetary policy without interference from the White House and Congress, focusing on price stability and employment. However, in a legal sense, everything is independent for an adult, but the act of making decisions is not independent: you have to consider many factors and are subject to many constraints. Conversely, why does the law enforce independence? Because there are opposing forces, more collusive and private non-independence exists, so a strong mechanism is needed to constrain it. The Fed still has independence, but the real constraints it faces when exercising its powers are much greater than before, and the redundancy space is very small.

Mr. Z: When the Federal Reserve was established in the early 1900s, it was to prevent the president from interfering too much with the Fed chair. In the 1970s, Arthur Burns was pressured by Nixon: "I want to be re-elected; don't mess things up for me." As a result, the era of high inflation emerged, and the federal funds rate once soared to 18% or 19%, equivalent to today's returns on private credit. So, do you mean that the Treasury and the Fed are more like twin brothers?

Tiezhu: In accounting, they cannot operate independently. The TGA (Treasury General Account) is an asset of the Treasury and a liability of the Fed, fluctuating between $700 billion and $1 trillion. If $200-300 billion flows out at once, the Fed must adjust and adapt its balance sheet. The Treasury relies on short-term debt and the repurchase market for financing, while the Fed must manage reserves to maintain buying power: the buyers of Treasury bonds are large banks and institutions, and their money comes from reserves. The two institutions are essentially working together to change the structure of debt supply and liquidity in the market. If you say they are not coordinating, that’s impossible. China's policy documents clearly state that fiscal and monetary policies should coordinate and cooperate. The U.S. does not explicitly say this, but it is effectively moving in that direction: when the cake grows very small and becomes a limited resource to compete for, the U.S. can no longer afford to waste and must be frugal. You can see that the U.S. is also implementing its own supply-side reforms and industrial policies, which are not inferior to China's: China's industrial policy is top-down planning, while the U.S. industrial policy is the result of negotiations among large enterprises, lobbying groups, and the government, each with its strengths, but the direction is the same.

2. The Ultimate Mission of the Federal Reserve: Not Inflation, Not Employment, but Preserving U.S. Treasury Bonds

Tiezhu: Let me say something controversial. The Fed's stated policy goals are price stability and employment, a dual mandate, but I believe its ultimate goal is to maintain the national debt. Imagine this extreme scenario: if the U.S. cannot repay its debt and cannot afford interest payments, what do inflation and employment matter? They don't matter. If the country goes bankrupt and its credibility collapses, will you still consider inflation? No, you won't. When the national debt was not so large, it was not sufficient to change the weight of the Fed's predictive factors; at this stage, the weight of the national debt factor must be higher, and the Fed must examine and consider this matter.

Tiezhu: Looking at the Treasury's recent expansion of repurchases, from the perspective of Bessenet, this is definitely not a temporary move; it must have been communicated with Waller: "Brother, the pressure on the long end is very high, which is not good for your management. I have money, so I will repurchase." From Waller's perspective, he also needs to use the Treasury's action to observe the market: using repurchases to lower long-term bond yields, will the market buy it? If not, and the sustainability is insufficient, what methods will I use to adjust later? This is our reverse thinking, deducing from their perspective.

3. Rate Hikes Cannot Contain Inflation: Rates Can Cut Peaks but Not Water Levels

Tiezhu: Recently, a U.S. bank released a report discussing three rate hikes. I shared internally: don't believe it; it's garbage; it's just a scare tactic to threaten the market. Waller is not a fool; he has an economics background, and it has been rigorously proven academically that raising interest rates cannot lower the average level of inflation, only the peaks of inflation. The real determinant of the inflation level is fiscal spending, the act of distributing money itself, not interest rates. If you raise interest rates while the Treasury is issuing money like crazy, it has no effect; it has no effect in any economic conclusion.

Tiezhu: To illustrate this, let me give an example. The U.S. is a K-shaped divided society, and I put the conclusion here: the lowest-income people are insensitive to inflation. If egg prices rise by 200% or 300%, it does not affect them; they eat and drink as usual, and if they can't, there are welfare benefits. The wealthy are even less affected. The only group that is sensitive to prices and truly feels the pinch is the middle class. Therefore, solving inflation is a complex issue that must be coordinated with fiscal policy. So why do we still raise rates sometimes, knowing it won’t work? Because it is an action for the Fed to maintain its independent credibility: this action may have no value, but it shows that I want to strongly maintain it.

4. How Will It Go by Year-End: Rate Hikes Are Impossible, 60% Probability of Rate Cuts

Tiezhu: My view at the beginning of the year was that there would be two rate cuts this year. After the U.S.-Iran conflict, I adjusted it to one. I mentioned last time I was on the show that there might be an opportunity for one rate cut by the end of the year, and that view has not changed. Let's use the process of elimination: there will be no rate hikes in September. Core inflation is in a downward trend (we need to look at trends, not single data points); September to October is a major node for the Treasury's new fiscal year budget and the Fed's economic forecast. Under this node, there is no motivation or necessity for rate hikes; if they do, it won't have any effect. So, it is highly likely that September will maintain: neither hike nor cut, and Waller will come out to shout, maintaining the shell of his independence.

Tiezhu: Looking at the end of the year: employment is still declining; besides coding and agents, other new scenarios for AI have not emerged; the debt is so large that after expanding repurchases, the long end has quickly risen again. In this situation, how can you raise rates? You can't. Either you raise them early, dragging it out until later becomes impossible, or you might as well cut rates directly. Another thing that everyone hasn't realized: the U.S. real estate market is also under tremendous pressure. This has nothing to do with Waller but is related to Bessenet and Trump. In summary: rate hikes are a significant obstacle for him, while rate cuts are a widely accepted idea. By the end of the year, rate hikes are impossible, and the probability of rate cuts is 60% in my view. Of course, with Trump as a professor, nothing can be predicted: who would have thought the U.S.-Iran conflict would come so quickly and violently? We can only explain based on current conditions.

5. The Stronger AI Becomes, the More the Federal Reserve Should Cut Rates

Mr. Z: So, from a certain perspective, the stronger AI becomes, the more the government should cut rates?

Tiezhu: You are absolutely right; this is also the conclusion I have been thinking about recently. The characteristic of K-shaped differentiation is that everything is unevenly distributed, with a stark divide between hot and cold. When the return on AI is very high, funds flock to AI, both in the capital market and in industries: at this point, your high interest rates cannot suppress its expansion; AI companies continue to develop, guiding the real interest rates in the long end of the market to remain high. However, other industries like real estate and small and medium-sized enterprises face rising financing costs, making their survival more difficult, leading to a more severe K-shaped differentiation, and to a certain extent, massive eliminations, with a recession not ruled out. The only situation that can return to equilibrium is when the productivity effects of AI are truly released. In the strongest industries, whether you raise rates or not has no impact; 5% or 6% doesn’t matter, just like the real estate market of the past; the weakest industries can no longer bear it: you cannot suppress the high ones, and if you raise rates on the low ones, they will die. How can you raise rates at this point?

Tiezhu: Therefore, the stronger AI becomes, the less sensitive it is to interest rates, and you should cut rates instead: the goal is not to inflate the AI bubble further but to maintain economic stability and take care of other industries. The U.S. government should use other policies to regulate and cool down AI. Some economists are calling for cooling down AI, which I think is a good idea. The lesson from the past is China’s real estate: only when it was about to collapse did they intervene, and a policy came out that directly caused a shock, while other industries had not developed enough to take over, making it useless to cut or raise rates at that time, which was very painful. Learning from history, the AI industry is no different.

6. Tech Debt is Not National Debt: AI is a New Real Estate Insensitive to Interest Rates

Mr. Z: I saw a rather unique argument from you. Everyone explains the high long-term interest rates as inflation or AI CapEx; however, you say that the way these two types of debt generate returns is different: the government borrows money based on taxes and disciplined fiscal spending, while tech companies issue long-term debt based on high technological growth. The maturity is the same, but the cash flow sources are different. So, does the issuance of so much long-term debt by tech companies squeeze the liquidity of U.S. Treasury bonds?

Tiezhu: There is definitely an existence. In the absence of significant liquidity expansion, everyone is using existing money: those doing fixed income are merely buying Treasury bonds or credit bonds. The key is that they are competing for long-term money in the market. The credit bonds issued by tech companies are all long-term, while the supply of long-term U.S. Treasury bonds has increased rapidly in the second and third quarters. A significant reason for the steepening of this yield curve is the mismatch of supply and demand. Market money is divided into short, medium, and long terms: bank money is short-term because it attracts savings, and no one saves for a lifetime; long money comes from insurance (a policy bought for decades), sovereign funds, and pensions, which have always been scarce. Issuing long-term debt must match with long-term money; the liquidity crisis caused by mismatched terms is unpredictable. Another overlooked squeeze is the siphoning of real physical resources due to large-scale construction. Engineers, construction, transformers, electricity—building data centers requires these, and building houses requires these too; AI giants have strong cash flows and can afford rising prices, while other companies may not be able to bear it.

Tie Zhu: The essential difference lies in the source of repayment. The credit entity of the government is the U.S. government: fiscal revenue, plus a portion from the intangible credit of military assets, operating in a continuous rollover mode, where new debts pay off old debts in a snowball effect. Technology companies are different: as a creditor, I look at your cash flow coverage ability, debt-to-equity ratio, and whether future revenue growth can cover interest expenses. Moreover, there is a rule: I cannot lend money to companies that are sensitive to interest rates. In the real estate era 20 years ago, real estate was not sensitive to interest rates at all; loans were given even at 13% or 15% interest because the future returns were high, and investments could be recouped immediately. Today's scenario is identical in the AI industry: AI is the new real estate that is not sensitive to interest rates. When I buy credit bonds, I am looking for the yield that is higher than government bonds; otherwise, why would I buy them?

7. SPV and GPU Financing: Credit Expansion Has Begun, but the Bubble Is Far Off

Mr. Z: AI CapEx initially relied on the balance sheets and cash flows of tech giants, but it is becoming increasingly market-oriented, even starting to engage in Project Finance: establishing an SPV to combine equity and debt in the same company, using this SPV for large-scale construction, with creditors assessing how much revenue the project can generate each year. Tech giants are increasingly relying on project financing and bringing private credit giants into the game. What do you think of this financing game?

Tie Zhu: Let me first state a major rule: capital markets always lead industries by one version. The industry stage may still be in its early phase, while the capital market has already entered the mid-stage and is beginning to create bubbles; this is a characteristic of all new industries. The project financing you mentioned is exactly what real estate did back then: project companies were established in various locations to finance as independent entities. The only difference is that back then, the target was banks, while today, AI's targets have shifted to private credit, corporate bonds, and off-balance-sheet financing. Why not seek banks? When I was still in traditional institutions, I had bosses coming to us for financing with a batch of A100s. The biggest challenge in GPU financing is that depreciation cannot be calculated: if you cannot repay the money, banks cannot sell the GPUs they hold, making them inherently unsuitable for bank credit. Therefore, private credit must be sought: it is backed by the long-term capital I mentioned at the beginning, willing to invest and able to bear term risks.

Tie Zhu: From a cyclical perspective, industry financing occurs in stages: the first stage relies on the credit of the entity, company revenue, and cash flow as sources of repayment; the second stage involves expansion, using different projects and SPV structures to raise more money. The emergence of project financing signifies that the AI industry has officially entered the stage of credit expansion. However, credit expansion does not equate to a bubble: it only indicates that capital has strong confidence in industry expansion, with more money flowing into this sector, which also means more risks. The endgame of all financial maneuvers is the same: when the bubble begins to burst, a bigger story is told to encompass everything, using a larger narrative to raise money and repay previous debts. The logic behind Evergrande opening branches everywhere was just that; recently, Xu's boss has also entered the fray. However, these two stages are still quite far from where we are now.

Mr. Z: However, houses are ultimately to C, directly reaching end consumers; GPUs are very B to B. Who knows how GPUs depreciate? It has only been in recent years that GPUs have started to be treated as assets similar to real estate. Therefore, we need to look further downstream at the application side: it’s just like Anthropic's IPO in October, valued at 2 trillion, surpassing SpaceX in June, with news saying they want to raise 100 billion; for a 65 billion ARR, the P/S is around 30 times. Is there really that much money left in the market to do this?

Tie Zhu: The difference in models is a fact, but it ultimately depends on one thing: demand. B-end demand will eventually translate into C-end improvements; we ultimately only measure whether supply and demand are matched.

8. Demand Has Not Peaked: The Great Explosion of AI Awaits the Post-00s to Take Center Stage

Tie Zhu: I have been paying attention to one piece of data: the token consumption of OpenRouter. It cannot represent all global calls, but at least it reflects whether token consumption is still on the growth curve, which is very valuable for reference: there was a period when its growth curve slowed, and at that time, bearish comments began to emerge. Let me also mention the corporate research I did this weekend: a very small, yet-to-be-listed company had a monthly token consumption of over a million dollars, with just one backend department burning 1,000 dollars a month, which completely surprised me. I asked him: can you really convert this into actual revenue? If you can’t convert it, your boss cannot be a fool; he will definitely reassess. He said that is indeed the case; everyone is confused, but they all know they must get on this boat, so they can only grit their teeth and go. The reality of this era is: the 60s and 70s still hold wealth; they don’t know how to deal with AI, but their logic is: you first spend money for me, and once you show results, then we can cut back.

Tie Zhu: The other side of widespread confusion is that, apart from coding, people have not found a second path, but exploration has already begun: companies like Xiaopeng in Guangzhou are already building platforms with their past data, creating knowledge bases, and AI-ifying. Let me provide another perspective: the internet emerged in the hands of those born in the 70s, but the real explosion and entry into various industries occurred when those born in the 80s and 90s took to the workforce. The same logic applies to the great explosion of AI; even the peak period must wait for those born in the 00s and 10s to truly become the main force in various industries: they have been using AI since birth and have the deepest understanding of it, and such transformation will come the fastest. It will only take three to five years. Relying on those born in the 80s and 90s won’t work; they are aging.

9. China's Answer: Alibaba Leads in Debt, the Open Source Trilogy, Miracles from Strength

Mr. Z: Back to China. Alibaba announced over the weekend that it is raising 80 billion HKD (over 10 billion USD) for AI CapEx. Will we see Tencent issuing bonds and Xiaomi raising funds in a few weeks? How do China's open-source models, such as Kimi K3 from Moonlight, Alibaba's Qianwen, and DeepSeek, compete with the U.S.?

Tie Zhu: There is a 99% probability that this will happen. I used to do policy research and have been paying attention to China's top-level design: in recent national regular meetings and economic conferences, my major judgment is that the consensus among China's top leadership on technology and AI is extremely strong, and there is anxiety and urgency: consumption has been so poor this year, but if you extract keywords like "technology, AI, investment" from the reports, their proportion is continuously rising; real estate has already gone blind, and results must be achieved. The views of the banking and capital markets are also consistent: traditional industries and assets can no longer yield returns; funds can only, and only want to flock in this direction. I have previously mentioned: the best way to eliminate capital is to let capital fight it out in a single track: let the duel between capital obliterate some of it, leaving the capital I want. This is the most effective way for a powerful government to control capital. Moreover, China's reserve of engineers is indeed richer than that of the U.S.; the U.S. advantage often lies in producing extraordinary geniuses.

Tie Zhu: The endgame can be compared to semiconductors: the only country that can compete with the U.S. is China. The path is a trilogy: first, deconstruct the underlying elements, whether through distillation or other technologies, using a large amount of manpower and computing power to crack it, achieving miracles through strength; second, lower prices; in this price war, Americans will undoubtedly lose; third, start implementing blockades and technical barriers, replicating the competitive model of semiconductors. Recently, there have been such messages emerging domestically; this is not a rumor, but a fact, because it is a strategic industry.

Mr. Z: Then let’s discuss a key issue: can Apple buy memory from Changxin? There is currently a severe shortage of memory, and Xi Jinping may visit the U.S. in September; is this a sign of goodwill from both sides? Can Tencent and Alibaba openly purchase leading chips like NVIDIA's H200?

Tie Zhu: My inference is that Apple must have contacted Changxin and Yangtze Memory, but this matter has not yet been finalized. This is indeed a negotiation tactic for both sides to use as leverage, but American tech companies are rational: buying some low-end products to control costs and increase competitiveness is beneficial for Apple; I believe they are even lobbying the White House to agree to this. If you are an investor, you can be 60% optimistic about this matter. The endgame will definitely involve both fighting and negotiating: just like NVIDIA, first tightening control, then loosening it a bit when you catch up, and tightening again. The competitive and confrontational relationship between China and the U.S. will not disappear; it will still exist even after Trump leaves: everyone will put forth their dishes and see who is ultimately satisfied.

Tie Zhu: As for whether embodied intelligence can replicate the model of electric vehicles taking over the world: there is a desire to do so, but the capability is lacking; the timeline cannot be compared to automobiles. Cars succeeded because they are labor-intensive, providing local governments with employment, tax revenue, and GDP, aligning with officials' KPIs; now local finances are strained, and central constraints on localities are much greater than before, making officials cautious and demanding more without being able to deliver quickly. Moreover, China's financing structure heavily relies on indirect financing from banks, whose risk appetite only recognizes government credit; the belief in local debts has not changed to this day, and it is unrealistic to rely solely on private capital. Yushu is a model, but when it comes to real money, everyone is in a tight spot: with a price-to-earnings ratio of one or two hundred times, can it lead the industry like internet giants with huge cash flows? I have my doubts.

10. Disenchantment and Positioning: After AI Levels the Field, What Remains Is the Winning Hand

Mr. Z: Lastly, Tie Zhu, do you have any words of caution for everyone?

Tie Zhu: It’s not about caution; let’s encourage each other and discuss three things. First, all investors must disillusion themselves about the U.S.: the capital management of large countries will inevitably move towards a state of state capitalism, a non-war, wartime control model at a certain stage; Europe is the same, and the climate strategies of the past are no longer effective. Therefore, whether investing in A-shares or U.S. stocks, it is crucial to pay close attention to industrial policies and the judgments and trends of senior government officials: what the U.S. supports and opposes, what China supports and opposes, and one must stand on the correct path.

Tie Zhu: Second, position management surpasses everything: it surpasses your analysis and many other factors. A friend asked me over the weekend: theoretically, buying recognized assets during market panic and selling them when the cycle rises is a surefire win; why can’t some people do it? This reflects the true state of market games: you can see, understand, and comprehend, but you may not be able to act; the gap in between is position. When positions are small, rationality can recover at any time; when positions are heavy, cortisol levels rise, stress hormones increase, leading to poor or even basic decision-making. Moreover, with the emergence of AI, it has leveled one thing: analysis. Today, everything we discuss can be analyzed by AI, and the conclusions drawn are not much different from mine or those of top investment banks like Goldman Sachs. When everyone arrives at the same analytical conclusion, why are there still people making money and others losing in this market? The winning hand is your mindset and position management, which is of utmost importance.

Tie Zhu: Third, I firmly believe that AI currently has no bubble: credit expansion has just begun, and the bubble period has not yet arrived, which presents a huge space and a great era for all investors. It is essential to study AI and the related industrial chain, as there will always be opportunities to find. Also, do not be afraid of missing opportunities; if you miss this one, there will be others: when storage prices rise, everyone complains about missing out, but when prices fall, are you brave enough to buy? The reason I started dollar-cost averaging into stocks I am optimistic about in the long term is that looking at it over three to five years, this trend has not disappeared: supply is expanding, OpenRouter's token consumption is rising, and demand from enterprises is increasing. This is an opportunity given by the times, worthy of everyone's study.

Mr. Z: A conversation that spans from the independence of Wosh to the AI era of the post-00s, discussing TGA accounts and Evergrande-style credit expansion. 168X is a talk show that connects the crypto world with AI and the U.S. stock market: when the market is good, we discuss it seriously; when the market is bad, we huddle together for warmth. We invite you to revisit last week's three discussions with Rick, Chen Guilin, and the AI industry excavator, laying out the logic from both bullish and bearish perspectives. This Wednesday, Mr. Beg will guest on the War Room to discuss Bitcoin: my own house view is that Bitcoin might be making a comeback. Every weekday evening, we look forward to seeing you!