🌐 The Meter Filter: Warsh Loses the Bond Market, Big Tech Splits & Wafers Wake Up
The Ruck Filter #030 • August 3, 2026
Read time: 8 minutes
Welcome back to The Ruck Filter.
Last week I asked which profits are actually money. This week the market answered with something sharper, and I think it’s the most important shift of the year: it stopped asking how much a company spends and started asking whether the spending shows up on a meter.
The setup was brutal. On Wednesday the Fed held rates, three of twelve voting members dissented in favour of a hike, and the bond market decided that wasn’t enough. The 30-year yield hit 5.24%, its highest in about 19 years. The Dow fell more than 1,100 points, its worst day since April 2025. Then earnings landed and split Big Tech clean down the middle. Microsoft posted its best session since 2008 and added roughly $450 billion in market value in a day - the largest one-day market-cap gain in corporate history. Amazon raised capex to about $220 billion and rose 15%. Meta raised the floor of its capex range and lost nearly 10%. Apple posted record June-quarter revenue and fell.
Every one of them grew revenue by double digits. The dividing line was something else entirely.
And in Frankfurt, largely unnoticed, a German wafer maker I’ve had on the watchlist since Issue #028 raised its sales outlook on AI demand.
1. Signal vs. Noise: The Meter Test 📊
The Noise: The market punished capex in July, then rewarded it in August. Investors are inconsistent and the AI narrative rules everything.
The Alpha: There’s nothing inconsistent about it. In two weeks the market has built a genuinely coherent test, and almost nobody has named it: spending is forgiven when there is a meter attached to it. Not a strategy, not a roadmap - a metered, contracted revenue line that turns capital into billings the customer cannot avoid. Microsoft and Amazon have one. Meta doesn’t. That single distinction explains outcomes that otherwise look random.
The Filter: Look at what actually happened, in order.
Microsoft reported Wednesday after the close. Revenue of about $90 billion beat consensus near $87.6 billion, up roughly 18% year over year. Azure grew in the low-to-mid 40s and its trailing-twelve-month revenue passed $100 billion for the first time in company history. Contracted backlog rose $51 billion in the quarter, to $678 billion. Microsoft 365 Copilot paid seats reached over 30 million, up from just over 20 million in April - roughly 50% seat growth in a quarter. Capital expenditure and finance leases jumped 69% to about $41 billion, and management left calendar-year capital spending guidance broadly unchanged near $175 billion. The stock rose about 15.5% Thursday, its best day since 2008, and closed the week up around 21% - its strongest week since October 2000.
Amazon did the opposite on the spending line and got the same verdict. It raised 2026 capital expenditure to roughly $220 billion from about $200 billion. AWS grew about 37%, its fastest since 2021. Quarterly revenue passed $200 billion for the first time. The stock gained around 15-17% on the week, its best since 2015.
Then Meta, reporting the same evening as Microsoft. Revenue of $60.8 billion, up 28%, beating consensus. And in Meta’s own release: operating cash flow of $31.86 billion, capital expenditure of $31.08 billion, free cash flow of $784 million - against $8.55 billion a year earlier, a 91% collapse. Earnings per share of $6.18 missed by roughly 14%, weighed down by a $2.4 billion charge related to legal proceedings. Meta raised the floor of its full-year capex range to $130–145 billion from $125–145 billion. Third-quarter revenue guidance of $61–64 billion put the midpoint below consensus. The stock fell roughly 9-10%.
Apple completed the pattern from the other end. Record June-quarter revenue near $109 billion, and the stock fell about 4% on softer Services growth and weakness in China. Strong business, no AI meter, no reward.
Here’s the mechanism, and it’s the part I haven’t seen written down clearly anywhere. A hyperscaler’s capex converts into a metered consumption line - Azure, AWS - where every GPU installed becomes billable compute with a contracted backlog behind it. Microsoft’s $678 billion RPO is the receipt. That makes the spending legible: investors can trace dollars in to dollars out with a visible lag. Meta’s capex converts into better ad targeting and models it uses internally. The return is real but it arrives diffused inside an existing revenue line, unattributable and unmetered. Same technology, same suppliers, same electricity. Completely different auditability.
Which reframes something I flagged in Issue #026, when Meta launched its compute business. I called it a competitive move against the neoclouds. This week suggests a second motive I underweighted: Meta needs a meter. On the call, Zuckerberg again gestured toward selling compute externally. That is not primarily about market share. It is about converting an unauditable cost into an auditable revenue line, because the market has just demonstrated what it pays for each.
A measured approach:
Apply the meter test as a screen, not a slogan. For any company raising capex, ask one question: what contracted, metered revenue line does this spending land in, and can I see it grow? If the answer is a backlog number or a consumption metric, the market will likely extend patience. If the answer is “it improves the core product,” expect the cash-flow line to be judged instead.
Watch the depreciation wave, because it’s the delayed bill. Capex becomes depreciation two to three years later, and it hits margins whether or not the revenue arrived. What to watch: year-over-year depreciation growth versus revenue growth at each hyperscaler over the next four quarters. That ratio, more than capex itself, is where this cycle’s disappointments will surface.
The combined number is now the market’s central fact. The four largest cloud providers expect to spend a combined $720–745 billion on capital projects in 2026. That is the demand backdrop for everything in the physical layer I’ve tracked since Issue #021 - and simultaneously the largest financial commitment corporate America has ever made against an unproven payback.
2. Warsh Loses an Argument With the Bond Market ⚖️
The Noise: The Fed held rates, stocks recovered by Friday, and the hawkish scare passed.
The Alpha: It didn’t pass. The 30-year Treasury yield ended the week near its highest level since 2007, after three FOMC members voted to hike. Warsh’s deliberate strategy of giving less guidance and letting markets price the path is now being tested - and this week the bond market’s answer was that it doesn’t believe the Fed is ahead of inflation. Equities recovered on earnings. The debt market did not un-say what it said.
The Filter: The mechanics. On Wednesday the FOMC left rates unchanged, but three of twelve voting members dissented in favour of a hike - an unusually wide split reflecting concern that inflation has run above the 2% target too long, with higher energy prices from renewed Iran tensions adding pressure. At the press conference Warsh tried to reassure markets that policymakers would act when necessary. The bond market wanted more than that. The 10-year climbed above 4.67% and the 30-year surged above 5.2%, hitting 5.24% - the highest in roughly 19 years. The Dow fell more than 1,100 points, or 2.2%, its worst single day since April 2025. By Friday the 10-year had topped 4.7%, its highest since January 2025.
Then the data cut the other way. Second-quarter GDP grew just 1.5%, below the 2% expected and down from 2.1% in Q1. June PCE fell 0.1% on the month, with the annual rate easing to 3.7% from May’s 4.1%; core PCE rose 0.1% on the month and 3.3% year over year. So: growth slowing, inflation cooling, and long yields at multi-decade highs anyway.
That combination is the whole story. When the long end rises on cooling inflation data, it isn’t pricing inflation expectations - it’s pricing risk premium. Investors are demanding more compensation to hold duration, either because they doubt the Fed’s resolve, or because they’re looking at supply, or both. One commentary I read this week put it bluntly: investors don’t seem enamoured with Warsh’s approach of letting bond investors do the Fed’s work for it.
I want to be careful here, because this is where sloppy analysis lives. A 5.24% 30-year with 3.3% core PCE is not automatically a crisis. It is, however, a materially different discount rate than the one embedded in equity valuations built during a decade of cheap duration - and it arrives in the same week that four companies committed to spending three-quarters of a trillion dollars on assets that depreciate. Those two facts belong in the same sentence more often than they appear in one.
The European contrast held, and it’s the cleanest argument for the allocation I’ve been making since Issue #025. The DAX rose 2.1% on the week and 2.5% in July, closing Friday at 25,629, within about ten points of its record at one stage. The ECB has already held with eurozone inflation near 2.8%. Same global AI tape, far less duration stress.
A measured approach:
Take the long end seriously as an input, not a headline. At 5.2%, the 30-year is a genuine competitor to equity risk for the first time in this cycle for patient, liability-matching capital. That doesn’t argue for abandoning equities. It argues for demanding more from them.
Watch the September meeting, but watch auctions more. The next FOMC is September 15–16. What to watch: long-dated Treasury auction demand between now and then. If yields keep rising on soft data, the message is supply and credibility, not inflation - and that’s the version that matters for valuations.
The DACH duration advantage is real and underappreciated. European rate-sensitives and domestically anchored names carry less of this repricing risk than their US equivalents. That case strengthened again this week.
3. Wafers Wake Up: Germany's Quiet Entry to the AI Supply Chain 🇩🇪
The Noise: German industrials are a China story and a fiscal story. The AI trade happens elsewhere.
The Alpha: On Thursday a Munich wafer maker raised its sales guidance specifically on AI, server and memory demand - and three brokerages upgraded it the next morning. It’s a small company with real problems, and that’s exactly why the signal matters: when AI demand starts lifting the unglamorous end of the German supply chain, the build-out has reached the base of the stack.
The Filter: I put the German chip-equipment chain in the theme bank in Issue #028 with a note to circle their Q2 reports in early August. Here’s what arrived.
Siltronic, the Munich silicon wafer producer, reported on July 30. Second-quarter sales of €321.6 million against analyst expectations near €316 million, up about 5% sequentially from €307 million. EBITDA of €69.4 million versus €66 million expected, with the margin improving to 21.6% from 21.2%. Most importantly, the company raised its full-year sales guidance: it now expects a decline in the low-to-mid single-digit percentage range rather than the mid-single-digit decline previously guided, thanks to higher wafer volumes. Management attributed the strength to 300-millimetre wafers for AI, server and memory applications, while the 200-millimetre market for power semiconductors and automotive electronics stayed weak and is only expected to recover in the second half.
Now the honest part, because this is not a clean success story. First-half sales fell 6.9% year over year to €628.1 million, and the company posted a first-half net loss of €130.0 million, against an €18.8 million profit a year earlier. Gross margin remains deeply negative, improving to about -4.5% from -8.5% in Q1. The loss is driven by depreciation and financing costs from the new Singapore fab. Siltronic also completed a €273 million capital increase during the quarter and re-entered the MDAX. The stock rose about 2.4% on the day to around €69.60, and on Friday gained a further 2.9% after three research houses upgraded it.
Aixtron, the deposition equipment maker, rose 5.3% on Friday to lead the MDAX, helped by an Oddo BHF upgrade to Outperform. Infineon led the DAX, up 3.7%.
Here’s why I think this matters more than a two-day move. Siltronic is the most brutally cyclical, price-pressured, capital-intensive rung of the semiconductor ladder - and it’s the one that only benefits when volume actually gets built, not when someone announces a plan. A guidance raise there is confirmation from the least promotional part of the chain. It also connects directly to the capital-cycle argument from Issue #027: Siltronic is spending heavily on a new Singapore fab, taking the depreciation hit now, in the hope of volumes later. That is the capital cycle happening in miniature, in euros, on a German balance sheet.
The counter-case is equally instructive. The 200-millimetre weakness is power semiconductors and automotive - the exact end markets tied to European industrial demand and the German car sector I wrote about last week. One company, two end markets, two entirely different cycles. That’s the truck-versus-car divergence from Issue #029 showing up inside a single income statement.
A measured approach:
Treat the guidance raise as a demand datapoint, not an investment case. A company with negative gross margin and a €130 million half-year loss is a turnaround dependent on pricing and fab ramp, not a compounder. The signal I take is about AI demand reaching the wafer level. What to watch: whether gross margin crosses back into positive territory, and whether 200mm recovers in the second half as management expects.
The German semiconductor chain deserves a permanent slot on the DACH watchlist. Siltronic (wafers), Aixtron (deposition), Süss MicroTec (advanced packaging), Infineon (power). These are the European claims on the $720–745 billion the hyperscalers just committed. Sized as cyclicals, not held as compounders.
Note the funding pattern. A €273 million capital increase into an AI-driven demand story is precisely the mechanism I described in Issue #027 - the shortage financing its own capacity. It’s now happening in the Munich mid-cap tier, not only in Seoul and Boise.
Outro: The Meter and the Bill
Three things happened this week that I’d connect. The market decided it will fund almost any amount of AI spending, provided the spending arrives on a meter someone else pays. The bond market decided it wants more compensation to fund anything at all, at yields not seen since 2007. And at the bottom of the stack, a loss-making German wafer maker raised its outlook because AI chips need silicon before they need a narrative.
Put those together and you get the shape of the next two years. The build-out continues - $720–745 billion of committed hyperscaler capex is not a rumour, it’s a budget. But it now runs through a filter with two screens: can you meter the return, and can you fund the duration at 5%? Companies that clear both get patience. Companies that clear one get volatility. Companies that clear neither get repriced, whatever their revenue growth looks like.
That’s a harder market than the one we had in 2025. It’s also a fairer one.
The Takeaway: When two companies spend the same billions and one gets rewarded while the other loses a tenth of its value, are you holding the meter - or the bill?
Daniel Ruck
Editor, The Ruck Filter
Filter Sources this week
Meta Platforms | Q2 2026 results press release, July 29, 2026
CNBC | Fed decision, yield moves, Meta earnings and weekly market wrap, July 29–August 1, 2026
Benzinga | Big Tech weekly scorecard (Microsoft, Amazon), July 31, 2026
Zacks / Yahoo Finance | US index closes, Q2 GDP and June PCE data, July 30–31, 2026
Digital Applied | AI capex scorecard and Microsoft disclosure attribution note, July 31, 2026
TechTimes, EBC & TradingKey | Meta cash-flow analysis and Big Tech reaction data, July 29–31, 2026
eMorningCoffee | FOMC commentary and 30-year yield analysis, August 1, 2026
Alain Guillot | weekly recap and combined hyperscaler capex figures, July 31, 2026
Siltronic AG | half-year 2026 statement (EQS), July 30, 2026
finanzen.net & Investing.com | Siltronic Q2 detail and call transcript, July 30, 2026
dpa-AFX (via onvista, ARIVA) | Frankfurt close, Aixtron, Siltronic upgrades, Infineon, July 31, 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.