US Core CPI MoM Exceeds Expectations! Don't Rush to Short US Stocks, the Four Giants' Q1 $165 Billion Equipment Purchases Consume 96% of Operating Cash Flow, Net Liquidity of 5.79 Trillion Is Miscalculated—Real Figure Only 2.93 Trillion, Oil Price Impact Peaks in Three Months and Returns to Zero in a Year! Three Ledgers Openly Calculated

This article analyzes the US CPI and PPI data released in September 2026, explores the transmission mechanism of oil price fluctuations on inflation, examines the Federal Reserve's actual liquidity status, and focuses on the cash flow and free cash flow pressures faced by the four tech giants—Microsoft, Alphabet, Meta, and Amazon—due to surging capital expenditures on AI computing power. It warns investors to be vigilant about the risk that the speed of return on computing power investment is lagging behind payment obligations.

Video Link:

Full Text:

Greetings to all who seek truth and reject casual cognition. Just two days ago, on September 11, 2026, the US CPI and PPI data for August were released. The headline CPI year-on-year was 3.4%, exactly in line with market expectations, but the core CPI month-on-month rose by 0.3%, 0.1 percentage points higher than expected. Meanwhile, the PPI was released just before the CPI; the PPI year-on-year rose by 5.4%, and month-on-month by 0.4%, with diesel alone rising by 24.1%. In late July, Microsoft, Alphabet, Meta, and Amazon released their Q2 2026 financial reports. The four companies combined spent $165.05 billion in a single quarter on purchasing equipment and building data centers, an increase of 827% year-on-year. When these pieces of news are put together, what is the first reaction of many? It is that oil prices are driving up inflation, inflation is forcing the Federal Reserve's hand, and if the Fed raises interest rates, the money for AI will dry up. Today, let's examine this claim. We will clarify how the three accounts—oil prices, dollar liquidity, and computing power capital expenditure—are intertwined, and which account is currently under the greatest pressure.

First, I will state my conclusion upfront. My judgment is that what is most worth being vigilant about right now is not risk, nor oil prices, nor dollar liquidity, but the fact that the speed of recovering cash from computing power investment has clearly fallen behind the speed of payment obligations. Furthermore, the oil price shock and financing costs are compressing the time left for adjustments. In the second quarter, the combined operating cash flow of the four companies was $171.7 billion, up 33.7% year-on-year, while their cash capital expenditure in the same quarter was $165.05 billion, up 827% year-on-year. The investment speed and cash recovery speed have clearly diverged. We must also clarify the boundaries of this judgment: the data evidence we currently possess is sufficient to review returns and performance capabilities, but it is not enough to declare that AI demand has bottomed out, nor can we say the cycle has reversed based solely on a downward adjustment in capital expenditure guidance.

Let's break down the August CPI. The headline CPI year-on-year is 3.4%, exactly the same as in July, and this was also the market expectation. However, I believe this figure has been suppressed; by my calculations, it should be around 3.6%. Looking at this number alone, it seems like nothing happened this month, right? The year-on-year contribution of energy rose from 14.7% in July to 16.3% in August, and month-on-month, it flipped from a negative 1.5% in July to a positive 2.1%. Gasoline alone rose by 3.9% this month. While the year-on-year core CPI fell from 2.5% in July to 2.4% in August, the month-on-month core CPI actually rose from 2.2% in July to 3.3% in August, 0.1 percentage points higher than market expectations. Therefore, this data is being pulled in two directions. What is keeping the headline CPI pinned at 3.4% and preventing it from falling is energy, while the core year-on-year is still trending downward, but the month-on-month has started to push upward. How do we view this structure? We need to pull the lens back by one month to the July data, because the calculation method is the same every month, and the July report has been fully released.

the Bureau of Labor Statistics has an official 12-month contribution breakdown table, which provides a comparison of the contributions of various items to the CPI. Food contributed 0.4 percentage points, energy contributed 0.98 percentage points, and core contributed 1.97 percentage points. Adding these three together gives 3.36 percentage points, which rounds to 3.4%. How did energy contribute? This is easy to calculate: 0.982 divided by 3.365 equals 29.2%. When many people see this 29.2%, they say that oil prices explain 30% of inflation. This statement is wrong. Where is it wrong? It is wrong because it confuses two types of quantities. What this 29.2% explains is how much of the total price increase that has already occurred is accounted for by energy. It is an accounting breakdown of a price basket. What is an accounting breakdown? It is adding up the money increased by every item in the basket for the year and then seeing what fraction energy accounts for. This does not mean that oil prices caused 29.2% of the inflation, and it certainly does not mean that 29.2% of the inflation exceeding expectations was caused by oil prices. Causal contribution and accounting share are two different things.

Moreover, within that 1.97 for core, housing accounts for 1.12, core commodities account for 1.15, and other core services account for 1.7. These three are already within core and cannot be added to core again, right? The 0.783 percentage points for gasoline is a subset of energy; changing to natural gas cannot all be blamed on crude oil, right? So how exactly do oil prices transmit to the CPI? Let's write the most direct gasoline path as a formula. Let's talk about this equation. Don't let its seemingly complex and long appearance fool you; in reality, it is just describing a very, very simple statistical fact. We have mentioned a simple conclusion before, but today let's be more precise. What was that simple conclusion? It was that for every $10 increase in the price of a barrel of crude oil, the US CPI jumps by 0.3 to 0.35 percentage points. In fact, the statistical logic of the US CPI is based on this set of formulas.

Actually, if you want to calculate the statistical contribution of any item to the US CPI, you can use this set of formulas. Look at this formula: CPI increase equals the weight of gasoline in the price basket, multiplied by the cost transmission rate, multiplied by the proportion of the crude oil price increase allocated to each gallon. We can explain this with elementary school arithmetic. Why do we need to handle 42 in the formula? Because one barrel of crude oil contains exactly 42 gallons. For every $10 increase in the international oil price per barrel, the cost per gallon increases by exactly $0.238, roughly 24 cents. But how much of a splash these 24 cents make when thrown into the equation depends entirely on the denominator, which is the original retail price of gasoline at the gas station. The weight of gasoline in the entire CPI basket is about 3.85%. Let's assume one thing: assume that refineries and gas stations don't play tricks and pass on the costs fully. If the oil price were cheap, say only $3 per gallon, then adding this $0.24 would be equivalent to an overall surge in oil prices of nearly 8%, which would then force the CPI up by about 0.3%. This 0.3% to 0.35% is where it comes from.

However, if we calculate it precisely, we have to update the data frequently. What is the benchmark now? It is $3.84, so this shrinkage magnitude immediately shrinks to...

Well, shrinking to 0.239%, the secret actually lies right here. The larger this denominator is, the more diluted the shock rate of crude oil price hikes will be. Here is another even more important distinction: after oil prices go up, meaning after oil prices rise, if the price level stays persistently at this high level with a one-time jump, then in the first year this year-over-year figure will be pushed up. However, after a full year has passed, this base effect will fade, and the year-over-year figure will pull back. Only when there are repeated shocks to oil prices, or when things like wages and rents follow up with a secondary adjustment, will the inflation rate be prolonged. Therefore, you must separate four things: the dollar shock, the percentage oil price shock, the CPI index points, and the year-over-year percentage points. You must look at these four things separately; whoever mixes them together will be led astray by this data.

What does the actual transmission path look like? I used monthly data from 2017 to 2026 to run a local projection regression, using the monthly changes in oil prices to see what? To see the cumulative changes in CPI over the subsequent 0 to 12 months. This is an actuarially calculated path. This econometric model looks quite complex, but when you break it down, what is it? Essentially, it is tracking an inflation pulse. On the left side, what do we have? We have the cumulative rate of inflation, which is starting from the rise in energy prices and looking across a full year ahead to see how many points the CPI will be pushed up like a snowball. On the right side, what we need to look at is the core coefficient BetaH. The significance of this term is that as WTI crude oil, the mother of commodities, is abruptly raised by 1% this month, how much sustaining momentum will it have in each of the following months? What kind of momentum? How much of the overall prices will be padded layer by layer. This projection equation puts the oil price increases from 2017 to 2026 and various items in the CPI into the same coordinate system to conduct a relational test.

Look at the logarithmic function on the left, what is it? It is an approximation drawn from the percentage of the CPI cumulative asset rate. Then, things like job openings and non-farm payrolls are all lumped into the letter Z. That gamma, called the regression coefficient vector, to put it bluntly, is splitting up every factor that can affect the CPI, breaking down the logarithm on the left—that is the ratio of pt plus H to pt minus 1—completely into what? Breaking them down into the cumulative values of the elements constituting the CPI, turning it into an addition problem. Under this method, the relationship between oil prices and CPI can be solved. As for the remaining residual, it is purely the real fluctuations that the model cannot explain. If you plug historical data in and run it for a round, the shock rhythm you calculate is extremely instructive. My actuarial conclusion is this: if crude oil prices suddenly surge by 10%, its transmission curve to the price level is very steep. In the month when energy prices rise, inflation will correspondingly tick up by how much? 0.139%. And transmitted to the third month, the entire shock will peak completely, and the cumulative push to this peak can reach how much? 1.97%. By the sixth month, this momentum begins to exhaust, falling back to 0.15%. Lengthening it to the twelfth month, the entire impact not only completely dissipates, but even slightly pulls back to a negative 0.053%.

What does this mean? It means that energy shocks are never an endless impact on inflation; rather, what are they? They are a very, very short-lived yet extremely forceful heavy blow. We can understand it this way: around the three-month mark, the force of this punch begins to wane; half a year later, the entire impact is cut in half; by about a year, the entire shock is basically completely cleared and zeroed out by the economic system. But here I must say one thing: why oil prices rise in reality actually has many possibilities. It could be production cuts by oil-producing countries, which is a supply shock; it could be economic overheating where everyone is scrambling, which is a demand shock; or it could be geopolitical friction, scaring everyone into hoarding goods, which belongs to safe-haven sentiment. This model does not have a God-like perspective either; in other words, it cannot turn oil price increases into a completely independent exogenous variable. What does it calculate? What it calculates is merely an averaged trajectory of how, after hundreds of thousands of complex games in history, when oil prices move, prices follow suit. It by no means assumes that oil prices are permanently locked in at the moment of a surge, and then looks at how the economic structure behaves—this ultimate truth does not provide such information here.

And let us not forget that right now in 2026, he just went back to doing that long-term payment which has only just begun, which forces the latter half of the model's long-term coefficients to still rely on an earlier historical sample. Let me add one more casual word: in the market, there are always people who like to grab a statistical variance and casually spout off, saying things like, look, 55.8% of this wave of inflation is entirely the fault of oil prices. This is purely substituting concepts. This 55.8% can at best prove what? That once the current oil price is packaged and plugged into the equation, the entire regression line will align better with past historical data and be drawn more smoothly. But what is this called? It is merely called higher drawing fit. As for how much true, unexpected inflation oil prices actually pounded into the market that caught everyone off guard, this has nothing to do with it at all, because these are fundamentally two different matters.

So what does the official model say? The Dallas Fed released a model on April 17, 2026, which factored what? It factored the supply gap caused by the closure of the Strait of Hormuz together into a simultaneous model of U.S. gasoline prices, personal consumption expenditure prices, and inflation expectations. Under the scenario of a 15% global supply gap, if the strait is closed for one quarter, comparing the prices of the fourth quarter of 2026 with the fourth quarter of last year year-over-year, the headline PCE inflation will be 0.6 percentage points higher than without the shock, and core PCE will be pushed up by 0.2 percentage points. If the strait is closed for three quarters, then headline PCE inflation will be pushed up by 1.1 percentage points, and core PCE inflation will be 0.3 percentage points higher. This is the conditional result given by the Dallas Fed model; it is not a realization probability.

If oil prices continue to transmit downward, the next station is where? Corporate profits. In the August PPI, final demand goods rose by 1.1%, and U.S. statistical agencies stated what? They said 40% of it came from energy. But in the same report, trade service prices dropped by 0.2%, while the retail margin indicator for fuel and lubricating oil dropped by 11.3%. What do trade service prices measure? They measure the profit margin in the retail and wholesale sectors, not a price sitting on a shelf. This provides very, very solid evidence. What evidence? That energy price hikes and the distribution sector eating into profits happen simultaneously. But mind you, you cannot extrapolate from this to say what? That all industries have lost pricing power. Let us look further into how much this compressed profit margins. We can use a precise cross-bar formula to calculate this: mu is the gross profit margin, which equals what? Selling price minus cost, divided by selling price. Tau is the pass-through rate, which is how many cents of price increase a company dares to make for every dollar increase in cost. The final formula is the change in gross margin. Don't look at it as complex; I will walk through the numbers and you will understand immediately.

For example, suppose the selling price is 100 yuan, and the unit cost is 70 yuan. What is the gross margin? 30 yuan, right? So the gross margin is 30 divided by 70 plus 30, which is 30%. An energy-related cost shock causes the unit cost to rise by 5 yuan, becoming 75 yuan. In this case, the company only dares to raise the price by 2 yuan, making the selling price 102 yuan. At this time, tau is 2 divided by 5, which equals 0.4. What is the new gross margin? It is 102 minus 75, which equals 27 yuan. Thus, the new gross margin is 27 divided by 102, which equals 26.4%. The gross margin was compressed from 30% down to 26.4%, a compression of 3.53 percentage points. Look at this formula: what is the denominator? It is p plus tau added to c, right? What does this represent? It represents the new selling price. And what does one minus tau multiplied by one minus mu in the numerator talk about? It talks about how much of the increased cost was not covered by the price increase. These 3.53 percentage points are calculated clearly from the accounts; it is more precise than what? Subtracting CPI from PPI to get that result.

Well, the ledger of oil prices is settled. Let's look at dollar liquidity. There is a popular net liquidity calculation in the market—I don't know if you've heard of it—which takes the Federal Reserve's total assets, minus the Treasury's checking account balance at the Fed (the TGA), and then minus the overnight reverse repo (ON RRP), and then comes up with a number saying what? Saying this is the US dollars entering the stock market. I will just pull out the numbers from the balance sheet and calculate, and everyone can take a look and see. Some people on the internet talk about this every day, saying this is the Fed's net liquidity indicator, which is this N, and then tell everyone that this is the water level gauge for the stock market's rise and fall. But if you were to treasure this, you would be completely fooled by this number. Its derivation formula looks very intuitive, using the Fed's total assets A minus the Treasury's current deposit at the central bank TGA, and then minus the overnight reverse repo network, but once you open up the real ledger to look at it, the situation is like this: the Fed's total

Assets are 6.737 trillion dollars, while how much money is currently parked in the Treasury account? It is 0.944 trillion dollars. Meanwhile, the overnight reverse repo pool has been almost drained, leaving a mere 5 billion dollars, which is completely negligible because it has basically dropped to zero. According to this formula, subtracting these gives a so-called net liquidity of as high as 5.792 trillion dollars. Looking at it, it simply looks like a grand spectacle of massive capital flooding, but obviously this calculation is wrong. The true liquidity should be based on that other formula, and it is not this number at all. Everyone will see shortly what this number compares closely to.

If you strictly take the real balance sheet data of the Federal Reserve, then taking total assets A, you must not only deduct the Treasury's TGA, but also deduct the physical paper currency and cash super C of entities in circulation throughout society. Just doing that isn't even enough; you also have to subtract the massive repo pool of all central banks including foreign central banks, which is the RFO indicator, as well as various miscellaneous liabilities and capital O. If this account is calculated to the end, the real figure is only 2.929 trillion dollars. In other words, how large is the exact gap between these two figures? 2.863 trillion dollars. This is already a huge chasm. So where did this nearly 3 trillion dollars of evaporated money go? It is all settled in the actual cash of ordinary people, as well as other types of depository institutions, and the accounts of foreign official institutions. If this N is regarded as real money that can rush into the stock market to take over positions at any time, this is essentially forcibly daydreaming a crude accounting subtraction into real-world capital flows.

Then what is even more fun? After I reconstruct this true liquidity by breaking down the Fed's consolidated report item by item, the final calculated error compared to commercial bank reserves is only on the scale of a few million dollars. In other words, in the world of true liquidity, a tiny error can make the water temperature feel like two completely different extremes of ice and fire. Then, looking further, U.S. commercial bank reserves returned to the level of 3.04 trillion dollars this week. Where did this hundred-plus billion come from? Yes, it flowed from the TGA account into commercial banks. Moreover, the SOFR repo rate RFO is now -1. What does this indicate? It shows that the overall liquidity environment is not particularly tight right now either. I think the current market is also quite interesting. First of all, policies are still deadlocked, as mentioned before, between the Treasury and the Fed. So it can be said that as long as the Fed does not shrink its balance sheet now, the short-term shock to the capital market will only be temporary. But everyone also needs to pay attention to one thing: starting from mid-September, the Treasury will densely conduct debt auctions again, so the fluctuation amplitude of the money in this account will be very large by next month, and market volatility will also be relatively large at that time. I think what Bessent wants to do has basically become clear at this stage. How will he do it? He will dynamically adjust the reserve account to fluctuate back and forth above the 2.95 trillion dollar level, keeping the market in what kind of state? An amplitude where it neither chokes to death nor starves to death, damn it.

Moreover, with such timely rate hikes now, I don't think the action of raising rates now will cause inflation to decline, because right now it is not just a matter of oil prices, but also the issue of capital expenditure return. Next, let's look at the third account, which is the cash coverage ratio of major model manufacturers. This is the core of the entire logic. Everyone look at this: the cash property, plant, and equipment (PPE) acquisitions on the balance sheets of the four giants—meaning the money spent on cash to buy factories and equipment—divided by operating cash flow. Let's look at this data: how much was it in the first quarter of 2023? It was 33.9 billion divided by 66.7 billion, equaling 50.8%. For the full year of 2025, it is 376 billion divided by 580.5 billion, equaling 64.8%. Coming to the second quarter of this year, it is 165 billion divided by 171.7 billion, equaling 96%. What does this mean? It means that for every 100 yuan of operating cash earned back, 96 yuan has to be spent again to buy equipment. In other words, the capital expenditure of big tech companies buying equipment and building data centers surged by 827%, but the operating cash flow actually earned back in their pockets only grew by 33%. In other words, the speed of shelling out money to burn on equipment has left the speed of generating cash far behind by several streets, and the newly added input is much faster than the newly added cash.

Let's break down the caliber even finer and look at the companies individually. Microsoft's operating cash flow is relatively good; its cash flow is 55.4 billion, and cash acquisitions are 35.8 billion, leaving how much comparable free cash flow? 19.6 billion. Alphabet's is relatively deteriorated; its operating cash flow is 39 billion, acquisitions cost 44.9 billion, and comparable cash flow is a negative 5.8 billion. Meta's operating cash flow is 31.8 billion, acquisitions are 30.1 billion, and what remains is a little over 1.7 billion, less than 1.8 billion. Amazon's operating cash flow is 45.3 billion, acquisitions cost 54.2 billion, and comparable free cash flow is a negative 8.8 billion. The comparable free cash flow here is uniformly calculated by subtracting balance sheet cash acquisitions from operating cash flow, because each company's own caliber is also different. For example, Meta's caliber also needs to deduct 962 million in finance lease expenses; if this part is deducted, only about 780 million remains. Meanwhile, Amazon's company caliber also needs to add back 1.13 billion of what is called asset disposal depreciation and interest income; if this part is added, it is a negative 7.6 billion, not yet 8.8 billion.

Then let's continue to look at whether investment is still accelerating. We can use differencing to look at it: k subscript t is the cash input in the 1st quarter, and the first-order difference is delta k. The significance of this delta k is how many more dollars were spent this quarter compared to the previous quarter. And the second-order difference is delta squared k. The significance of delta squared k is whether the extra money spent itself is getting larger or smaller, which is the change in the increment. What can truly peel back the underlying psychological state of big tech companies? It is this second-order difference, delta squared k. In physics, what is this called? Acceleration. What does it measure? It measures whether the momentum of spending more money itself is accelerating in expansion or contracting. If the second-order difference is negative, what does it indicate? It indicates that although expenditures are still accelerating, the pace has already slowed down. But once the second-order difference is positive, it means these giants are not only spending money, but also slamming the accelerator all the way down into the fuel tank to spend.

Everyone look at how fierce the battle was in the second quarter of this year: the first-order difference delta k combined for the four major giants was 35.3 billion, meaning they spent 35.3 billion more in the second quarter than in the first quarter. The second-order difference, delta squared k, unexpectedly surged directly by 24 billion dollars. This shows that in terms of the acceleration of spending more money, they were over 24 billion fiercer than the previous quarter. Let's look at the delta squared k of individual companies separately: how much is Microsoft's? It is 3.92.6 billion. Alphabet is 1.42.7 billion. Amazon is 5.3 billion. The most aggressive is Meta, which single-handedly blasted out 13.5 billion. In other words, all four are glowing red as positive, and everyone is generating an extremely intense sense of G-force pushing you back into your seat. Of course, there is also a slight seasonal factor here, which is that the second quarter happened to coincide with the concentrated delivery of many data centers, plus a bunch of lumped-together payments. The unadjusted seasonal raw delta squared k might be biased by seasonality. In other words, there is a slight overestimation component here. Therefore, to squeeze out the water, I calculated an even harsher figure: the quarter-over-quarter change in year-over-year increments. That is, taking the extra money spent in the second quarter of this year compared to the same period last year, and subtracting the extra money spent in the first quarter compared to the same period last year, just to filter this part out. As a result, after filtering this way, the number is still a positive 18.956 billion. This positive number fiercely slaps a heavy blow to the notion that it is purely seasonal noise, substantiating the fact that these giants are indeed stepping hard on the accelerator. Although other factors are mixed in here, such as rising supply chain costs and land purchases, we can still see a state where the major tech companies would rather push their cash flow to such an extreme limit than desperately scramble for computing power. This gives off somewhat of a gambler's posture. Therefore, under this differencing structure, it has been exposed completely bare.

Why am I calculating these things? To let everyone see one thing: it is not yet time to be bearish on U.S. stocks, and don't ever hear something like rate hike expectations and treat it like the boy who cried wolf, even if they actually raise rates. When they raise rates at a time like this, what should you treat it as? Treat it as an aside, unless you think rate hikes can immediately reprice these tech companies. Otherwise, if you foolishly go short on U.S. stocks at this time, just think about what kind of composition the selling pressure coming in with that aside is, especially when the market is in such extreme contradiction. Therefore, I believe that even if they really raise rates, it will just be a short-lived pulse. If a large number of short orders cause crowding at this time, it is entirely possible to move towards a short squeeze, because shorts are also liquidity.

I'll give a fairly recent example: remember that wave at the end of 21? I remember starting from September, pretty much around this exact time, the entire market expectation was entirely rate hike expectations, and inflation at the time had already broken through 5 percent. Oh, ghost stories were flying everywhere just like now, and then, watching that speculation

The result of a massive concentration of airdrops is simply this: when airdrops cluster together, the fuel clusters together. Once the fuel clusters and the liquidity of the airdrops is exhausted, what follows is a short squeeze. Interest rates were indeed hiked in 2021, but those who followed in to short the market at that time were also wiped out. The reason I am presenting so much calculation data today is to let everyone know that no single factor can influence this market; the current focus must return to liquidity itself. Alright, I will stop rambling here for today. If you like my show, please click the bell icon below. If you would like to support me, please join the membership channel. The membership channel features four in-depth programs every month. Welcome to join the ranks of the good guys. See you in the next episode.