Read time: 8 minutes
Welcome back to The Ruck Filter.
Everyone spent Friday arguing about whether the Federal Reserve raises rates on Wednesday. Odds went from roughly half two weeks ago to somewhere between 85% and 90% by the close. That debate will resolve itself in three days and I have nothing useful to add to it.
The number I can’t stop looking at is a difference. American producer prices accelerated to 5.4% in the year to August, up from 4.8%. Consumer prices stayed at 3.4%, and the core measure actually eased to 2.4%. That is roughly two percentage points between what companies pay and what they charge.
That spread is margin. It is sitting on income statements right now, and it has exactly two ways out.
Three sections: what the gap means and the one development this week that could close it from the other side, why the Treasury’s bond-buying programme failed its first real test, and how Europe ended up paying for its own stimulus twice.
1. Signal vs. Noise: The Two-Point Gap 📊
The Noise: Inflation came in roughly as expected, so the story is just whether the Fed moves on Wednesday.
The Alpha: The interesting number isn’t either print. It’s the distance between them. Producer inflation is accelerating while consumer inflation is flat, which means companies are absorbing input costs they haven’t passed on. Either they pass it through and consumer prices head higher, forcing more tightening, or they don’t and earnings compress. I can’t find a third option, and almost nobody is framing it this way.
The Filter: Take the two prints in order. Thursday’s producer price index rose 0.4% on the month, with the annual rate accelerating to 5.4% from 4.8%. Markets sold off. Friday’s consumer price index came in close to expectations: headline at 3.4% over twelve months, the same pace as July, with core at 2.4%, down from 2.5%. The monthly core figure was the one blemish, at 0.3% against the 0.2% economists expected.
Read together, those describe a squeeze rather than a stabilisation. Input costs are rising at 5.4% a year. Output prices are rising at 3.4%. The difference has to live somewhere, and where it lives is gross margin.
I’ve been assembling evidence for this for two months without naming it properly. In Issue #033 Walmart reported average ticket growth of 1.1% against 3.4% inflation, which told you the pass-through wasn’t happening at the largest retailer in America. In Issue #034 Nvidia guided its own gross margin down from 75% to 71–72%, with its CFO saying plainly that memory scarcity is being driven by the AI build-out itself. Those aren’t three separate stories. They’re the same story seen from the supplier, the retailer and the chip designer.
The consumer expects the pass-through, which is its own problem. Friday’s University of Michigan survey missed estimates at 47.8 — worse than August’s 51.0, which I already described as a recession-level reading — and its one-year inflation expectation came in at 4.6%, well above the actual rate. Households are braced for prices they haven’t been charged yet.
And oil is pushing the same direction. Brent touched $108 intraday this week, its highest since May, before giving back about 3% on Friday.
Now the one thing that cuts the other way, and it arrived on Thursday.
The Chinese developer DeepSeek released a model called V4.1-Flash, and the claim in it matters more than the model. The company says the new architecture reduces the KV cache — the attention state a model holds while it works — to roughly 890 bytes per token, one-quarter of what its previous Flash model kept in high-bandwidth memory and one-eighth of what it wrote to solid-state storage. It does this by reusing cached information across network layers, storing the cache at 4-bit precision, and reconstructing parts of it on demand rather than keeping the whole record. The model activates between 800 million and 1.6 billion of its 552 billion parameters during inference.
The qualification is essential and I want to be precise about it. DeepSeek did not claim its whole model or a data centre uses 75% less memory. The comparison covers the KV cache only, and it is measured against DeepSeek’s own preceding architecture. Memory for model weights and for training is untouched.
But the direction is what I’d pay attention to. April’s V4 had already cut KV-cache requirements to one-tenth of its predecessor at full context length. September’s model cut them by another three-quarters. Two step changes in five months, in the component that grows fastest — because in long-running agent workloads, the cache eventually needs more memory than the model itself.
The market response was informative precisely because it was split. Samsung fell 3.53% in Seoul on Friday and SK Hynix 2.21%, with SK Hynix ADRs and Micron sliding as much as 5% overnight in New York. Yet Micron and SanDisk closed the US session positive. Two days before the release, Goldman Sachs had turned more constructive on memory. The same week, JPMorgan initiated coverage of SK Hynix’s ADR at Overweight with a $245 target, arguing the AI memory upcycle could run more than five years.
Here’s why this belongs in a section about margins. The AI cost boomerang I described in Issue #025 — memory scarcity raising input costs across the economy — assumes that memory demand rises with AI usage. DeepSeek is the first serious evidence that the relationship might be weaker than that. Efficiency is deflationary. If the biggest single driver of AI-related input inflation gets cheaper per unit of work, the two-point gap starts closing from the cost side rather than the price side.
A measured approach:
The gap is the position, not the print. Watch gross margin guidance across this earnings season rather than headline inflation. If companies guide margins lower while holding revenue, the squeeze is real and the Fed’s problem is smaller than it looks. If they push prices instead, the Fed’s problem is larger.
Don’t over-read DeepSeek in either direction. A narrow claim about one component, measured against the same company’s prior model, is not a refutation of the memory cycle. It does mean unit efficiency belongs in every demand forecast alongside token growth, and I don’t think most of them have it. What to watch: Micron reports on September 30, guided to roughly $50 billion in revenue with an 86% gross margin. Its calendar 2026 high-bandwidth memory supply is already contracted on price and volume, so the question isn’t this year. It’s whether the company repeats the forward estimate it moved from 2028 into 2027.
In Issue #032 I wrote that memory had converted price risk into counterparty risk. This week added a third category I didn’t name: demand-per-unit risk. Contracted volumes protect 2026. They say nothing about what gets contracted for 2028.
2. The Buyback That Couldn’t 🧰
The Noise: Yields are up because of oil and inflation data. Normal repricing.
The Alpha: Something more specific happened. The Treasury’s expanded bond-buying programme — the one I wrote about at length in Issue #033 — started operating on September 9. In its first week it could not buy its full allotment, and markets treated that as a reason to sell rather than a reason to buy. The toolkit got tested, and it did not hold.
The Filter: The mechanics, since the detail is the story. Treasury announced in August that it would at least double the size of its liquidity-support buybacks in the 10-to-30-year sector, effective September 9 and running to November 4. At the time I wrote that the announcement itself produced a ten basis point drop in the thirty-year that reversed within twenty-four hours, and that the market now knew a bid existed and could be enlarged.
This week the programme actually ran. In its operation, Treasury purchased $5.187 billion of securities against a maximum authorised amount of $6 billion. It did not fill its own order. Market commentary described a sharp sell-off in Treasuries driven by higher oil prices and a disappointing buyback operation — meaning an intervention designed to support the long end was cited among the reasons for selling it.
Where that left prices by Friday’s close is the part worth committing to memory. The thirty-year yield approached 5.38%, within two basis points of the June 2007 high of 5.40%. The ten-year traded at 4.975%, its highest intraday level since October 2023. The two-year reached about 4.65%, a two-year high. And the iShares 20+ Year Treasury Bond ETF fell to its lowest price since 13 May 2004 — a twenty-two-year low.
Now put the other instrument next to it. In Issue #034 I argued from Warsh’s Jackson Hole speech that credibility, not balance-sheet operations, is what lowers long-term yields. That approach is now being deployed at a scale I did not anticipate. The ECB hiked unanimously on Thursday. The Fed is priced at 85% to 90% for Wednesday. Four sources told Reuters to expect a 25 basis point move from the Bank of Japan next week, with a signal of faster tightening to follow. And JPMorgan published a forecast this week that eight of nine developed-market central banks will raise rates by year-end — the Fed, the Bank of Japan, four in Europe, Australia and New Zealand.
So both tools are running simultaneously. Purchases at the long end, tightening at the short end, in a coordinated way that has no recent precedent. And long yields are at multi-decade highs anyway.
In Issue #034 I set the test in one sentence: if long yields rise despite all of this, the central banks have paid with growth and bought nothing. I don’t think one week settles that, and I want to be careful not to declare a verdict on a single operation. But the first evidence is in, and it is not encouraging.
A measured approach:
Watch the buyback operations themselves, not the announcements. They are published, they are scheduled through November 4, and their execution rate is now a live indicator of dealer willingness to sell into the bid. A second under-filled operation would say something the announcement never could.
A twenty-two-year low in long-duration Treasuries is a portfolio fact, not a headline. Any allocation built on bonds diversifying equity risk has been tested for two months and found wanting. What to watch: whether the thirty-year takes out 5.40%, which would be the highest since before the financial crisis.
Equity leadership is narrowing while this happens. The Nasdaq-100 fell 0.52% on the week while the average stock fell roughly three times as hard. I flagged the same pattern in Issue #024 and again in Issue #030. It is a thin margin for error in a market repricing its discount rate.
3. Europe Pays Twice 🇪🇺
The Noise: The ECB hiked a quarter point as expected. Priced in, move on.
The Alpha: Look at how Lagarde justified it. The ECB upgraded its growth forecast, and among the reasons given was manufacturing supported by defence and infrastructure spending. That is the German fiscal programme I have tracked since Issue #023 — now cited by the central bank as evidence of the resilience that warrants tighter policy. The stimulus is generating the strength that justifies the rates that raise the stimulus’s own cost. I haven’t seen anyone else point at that loop.
The Filter: The decision first. The Governing Council raised all three key rates by 25 basis points on Thursday, taking the deposit rate to 2.50% from 2.25%, effective 16 September, with the main refinancing rate at 2.65%. It was unanimous. Lagarde called it a no-brainer and said the move was robust against all three scenarios the ECB had mapped for the region.
Two things in the projections matter more than the rate itself.
First, the level. Economists generally place the neutral range at 1.75% to 2.50%, which means the ECB has just arrived at its upper boundary. Any further increase moves policy into restrictive territory for the first time in this cycle. Lagarde said explicitly that the neutral band would not inform the Council’s decisions — but markets have drawn their own conclusion, pricing a 75.1% probability of another increase on 29 October and 73.3% for 17 December. A Deutsche Bank survey of clients found no consensus at all: more than a third expect a peak at 2.75%, while one in four expect the ECB to stop here.
Second, the reasoning. The ECB raised its growth forecasts, to 0.9% for 2026 and 1.4% for 2027, citing greater-than-expected resilience. Lagarde attributed broad-based second-quarter growth to manufacturing supported by defence and infrastructure spending, recovering consumer confidence, and AI-related momentum in digital services and exports. Unemployment held at 6.4% in July. Inflation projections were revised higher for 2027 and 2028, with the measure excluding energy and food seen at 2.5% this year, 2.6% next and 2.3% in 2028. Lagarde warned that the Middle East conflict and developments in Russia’s war on Ukraine will keep headline inflation well above target for an extended period.
That is a central bank tightening into strength, which is always more comfortable than tightening into weakness. It also completes a circle I’d rather it didn’t.
Germany’s fiscal programme was designed when money was nearly free. In Issue #033 I wrote that Berlin sold thirty-year Bunds in August at 3.783%, the highest yield on that maturity in sixteen years, with German reporting putting the potential federal interest burden at €80 billion annually by 2030. Now the growth that programme generates is being used, in part, to justify a higher policy rate — which raises the cost of the borrowing that funds it, and the cost of credit for every domestic business delivering into it. Europe is paying for this stimulus on the issuance side and on the policy side at once.
The equity market felt it immediately. The DAX fell 1.66% on Wednesday to 25,576, losing the 26,000 level, then another 0.84% on Thursday to 25,361 after the decision. Friday brought a 0.70% stabilisation to around 25,539, leaving the index down roughly 3% on the week and running toward its 100-day line at 25,203.
A measured approach:
The October meeting is the one that matters, not the one that just happened. A move to 2.75% takes European policy restrictive while Lagarde’s own risk assessment puts growth to the downside. What to watch: whether the energy shock shows any sign of feeding into domestic prices between now and 29 October, because that is the stated condition for the next step.
Within Europe, the split I described last week has widened rather than narrowed. Businesses whose demand and financing are both set domestically now face a higher policy rate on top of a repriced curve. Exporters into the global technology supply chain face neither.
Treat backlog-heavy fiscal beneficiaries with more care than the order books suggest. A pipeline priced at one cost of capital and delivered at another is a margin problem disguised as a growth story. I’d want to see the margin commentary before I trusted the backlog number.
Outro: Three Gaps
This was a week of spreads rather than levels. Two points between what companies pay and what they charge. A missing eight hundred million between what the Treasury was authorised to buy and what it managed to buy. And in Frankfurt, a growing distance between the reason given for a rate increase and who ends up paying for it.
None of those are the kind of thing that leads a bulletin. All three tell you more than the headline numbers they sit inside. The Fed will move or it won’t on Wednesday, and either way the gap between producer and consumer prices will still be there on Thursday morning, sitting quietly on somebody’s income statement.
The Takeaway: Producers are paying 5.4% and charging 3.4%. In your portfolio, who is absorbing that difference — and for how much longer can they?
Daniel Ruck
Editor, The Ruck Filter
Filter Sources this week
US Bureau of Labor Statistics | August CPI and PPI reports, September 10–11, 2026
Investrade | daily and weekly market reviews, Treasury buyback execution and yield levels, September 11, 2026
Lance Roberts, Bull Bear Report | CPI, PPI and index breadth, September 12, 2026
University of Michigan | preliminary September consumer sentiment and inflation expectations, September 11, 2026
DeepSeek | V4.1-Flash release, September 10, 2026 (via ZeroHedge, Stocktwits/Yahoo Finance, BigGo Finance, Insider Monkey)
TipRanks | memory sector reaction and analyst positioning, September 11, 2026
European Central Bank | monetary policy decision and macroeconomic projections, September 10, 2026
CNBC | ECB decision, Lagarde press conference and Deutsche Bank client survey, September 10, 2026
Bloomberg | Lagarde “no brainer” remarks and market pricing, September 10, 2026
Morningstar | ECB neutral-range analysis and October/December probabilities, September 10, 2026
EBC Financial Group | ECB projections detail and euro reaction, September 11, 2026
icrypex Global Market Report | JPMorgan central bank forecast and Reuters BoJ sourcing, September 11, 2026
dpa-AFX (via ARIVA, finanzen.net) | DAX daily closes and technical levels, September 9–11, 2026
wallstreet-online | Brent intraday levels and chip sector reaction, September 11, 2026
Disclaimer: The Ruck Filter is for informational purposes only and does not constitute financial, investment, or tax advice. The information provided is based on data available at the time of writing and is subject to change. Investing in financial markets involves risks, including the potential loss of principal. Every reader is solely responsible for their own trading and investment decisions. Please conduct your own due diligence or consult with a licensed professional before making any financial commitments. Companies with potential conflicts of interest are excluded from coverage as a matter of editorial policy.