The jobs number put a rate hike back on the table — and the physical semiconductor chain ignored it completely

Friday, 4 September 2026 · 07:00 BRT slot · compiled 16:55 BRT, after the US close

The one thing

August payrolls came in at 162,000 against a Dow Jones consensus of 53,000 — the strongest month since March — with unemployment steady at 4.1%, and June and July revised up by a combined 55,000. Short-dated Treasury yields jumped and the market moved back toward pricing a quarter-point hike at the 15–16 September FOMC. Not a pause. A hike. The index tape did what you would expect: S&P 500 −0.37%, Nasdaq −0.25%, Dow −0.53%, giving back part of Thursday’s rally.

Underneath it, the split was violent and it ran along a single line. Everything physical went up: KLA +7.6%, Marvell +7.0%, Micron +6.0% through $1,015, CoreWeave +5.9%, Lam +5.3%, AMD +5.0%, Intel +4.6%, Applied Materials +4.4%, ASML +4.3%, TSMC +2.9%, Oracle +2.8%. Everything sold as a recurring-revenue multiple went down: Adobe −6.8%, Synopsys −5.4%, Netflix −5.2%, Snowflake −5.1%, Cadence −4.0%, MongoDB −4.0%.

That is the whole session in one line: the market is paying for tonnage and selling duration. Wafer starts, DRAM bits, deposition tools and contracted GPU capacity got bid on a hawkish print; the same AI theme expressed as a software multiple got sold on the same print. Snowflake is the cleanest illustration — it closed +16.6% on Thursday on a third consecutive quarter of accelerating product revenue, then gave back 5.1% on Friday without any news. If Wednesday’s regime read was “beats don’t pay, acceleration does,” today’s is narrower and more useful: within AI, near-term physical throughput is being repriced up and duration is being repriced down, simultaneously, in the same sector, on the same day.

Sector pulse — covered names with material news

Overnight & last session

Semis & equipment — 3 named, plus a sector-wide bid

NVIDIA confirms it is buying Hugging Face for $12.93bn

NVDA · 3 Sep · nvidianews.nvidia.com and 8-K

NVIDIA agreed to acquire Hugging Face for $12,930,300,000 — reported as roughly $11.9bn to investors plus up to $1bn in employee retention — with the transaction expected to close in the first half of 2027. Hugging Face hosts about 3 million models, 1 million applications and 500,000 datasets, used by 18 million developers, researchers and creators and more than 200,000 companies. Jensen Huang wrote that the platform stays open: “NVIDIA compute will not be required to build on or deploy through Hugging Face.” NVIDIA is already the largest single contributor there, with 500-plus models and 250-plus datasets published. NVDA closed +1.8% Thursday at $228.45 and finished Friday around $230.55, +0.9% — a notable non-event given the size of the move around it.

The read: this is not a silicon acquisition and it should not be modelled as one — $12.9bn is under two weeks of NVIDIA’s current revenue run-rate (FY2Q27 revenue was $96.2bn, +106% YoY). It is a distribution acquisition. Your NVIDIA initiation frames the moat as CUDA plus the installed base; Hugging Face is the layer above CUDA where model choice actually gets made, and NVIDIA has just bought the shelf rather than the product on it. The open-platform pledge is the part to hold management to — it is the same commitment structure as the neocloud arrangements, where NVIDIA is simultaneously supplier, investor and demand guarantor. This belongs on the circular-financing map, not in the revenue model.

Memory and front-end equipment melt up while the index falls

MU · KLAC · MRVL · LRCX · ASML · 4 Sep, at 15:57 ET

On a session where the S&P fell and the Fed was repriced toward a hike, the physical semiconductor chain led the entire market: KLA +7.6%, Marvell +7.0%, Micron +6.0% through $1,015, Lam +5.3%, AMD +5.0%, Intel +4.6%, Applied Materials +4.4%, ASML +4.3%, ARM +3.8%, TSMC +2.9%. Non-covered proxies moved with them — SanDisk +7%, Astera Labs +10% — and CoreWeave, the levered edge of the same trade, closed +5.9%.

The read: the interesting fact is not the size of the move, it is its sign relative to rates. A hawkish payroll print should compress every long-duration AI multiple, and it did — for software. Front-end WFE and memory went up anyway. That is the market treating 2026–27 wafer and bit demand as contracted rather than discounted, which is consistent with the order-book lengthening you documented at Broadcom on Wednesday and with TSMC’s July revenue at NT$467.6bn, +44.7% YoY. The test lands in six days: TSMC’s August monthly number on the 10th is the first hard datapoint that can confirm or break the tonnage thesis. If you want one chart from this week, it is covered semis’ daily returns plotted against the 2-year yield — the correlation has flipped sign, and that is a publishable observation in its own right.

EDA is the one semiconductor group trading like software — and it broke today

SNPS · CDNS · 4 Sep, at 15:57 ET

Synopsys closed −5.4% and Cadence −4.0%, the two worst performers in the semis book on a day everything physical rallied. There is no company-specific news: Synopsys last reported on 26 August, and it was a good quarter — FY3Q26 revenue $2.477bn, +42.4% YoY, non-GAAP EPS $3.91 and GAAP EPS $2.84, with full-year guidance raised to roughly $9.715bn of revenue and $15.07 of non-GAAP EPS at the midpoints and Q4 EDA growth guided to accelerate to double digits. That release also raised the estimated pre-tax GAAP restructuring charge to $425–500m — note that this is the cumulative plan estimate through FY2027, not a FY2026 expense; FY2026 GAAP restructuring guidance is $380–410m, and the ~10% workforce reduction figure comes from the November 2025 8-K rather than the August update. SNPS is down roughly 11% year to date (aggregator figure) against an S&P up double digits. Its Investor Day is 30 September.

The read: EDA has the strongest reported growth in your semis coverage and the weakest tape. Both facts point at the same thing — EDA is priced as a recurring-revenue software franchise, so it takes the duration hit a WFE name does not, while its 42% growth is being read as Ansys acquisition arithmetic rather than organic AI pull. That is a testable claim and you have the data to test it: your Synopsys initiation already flags that the valuation workbook doubles as the master and that its FY3Q26 column was a pre-earnings estimate. Refresh that column against the filed 26 August numbers, strip Ansys, and publish the organic EDA growth rate before the 30 September Investor Day. If organic EDA is genuinely accelerating to double digits as guided, today is the entry point and nobody has shown the clean series.

SaaS & enterprise software — 1 named

Adobe names Anil Chakravarthy CEO; David Wadhwani is leaving

ADBE · 3 Sep, after US close · news.adobe.com

Adobe announced that Anil Chakravarthy — currently president of Customer Experience Orchestration and worldwide field operations, previously CEO of Informatica for four years, at Adobe since January 2020 — becomes president and CEO and joins the board on 1 December 2026. Shantanu Narayen, CEO since 1 December 2007, becomes executive chair; at the handover that is exactly 19 years in the seat, and Adobe’s own wording is “more than 18 years.” He signalled the transition in March. Separately, David Wadhwani, president of the creativity and productivity business for nearly five years and the internal candidate most of the street expected, is leaving. ADBE closed −6.8%, the worst print in your coverage today, after finishing +2.1% on Thursday. Adobe reports FY3Q26 on 10 September, after the close.

The read: the market is not pricing the appointment, it is pricing the departure. Adobe’s AI problem has always been located in Creative Cloud — the segment Wadhwani ran, and the one exposed to generative substitution — and the board has just handed the company to the data-and-marketing-orchestration side while the creative leader walks. Your Adobe initiation already documents that FY2025 segments do not foot; a leadership change of this shape usually precedes a segment redefinition, and Adobe is a serial redefiner. Flag it now, before 10 September, so the note is dated ahead of any reporting change rather than after it. The setup is unusually rich: a new CEO named seven days before a print, with the outgoing divisional head already gone.

Hardware & devices — 1 named

Apple’s “Surprise and shine” keynote is Wednesday — the first under John Ternus

AAPL · Event 9 Sep, 13:00 ET = 14:00 BRT

Apple’s event runs 9 September at 10:00 PT / 13:00 ET (14:00 BRT) at Apple Park. John Ternus took over as CEO on 1 September 2026, so this is his first keynote. AAPL closed −2.4% at $320.28.

The read: if the reported split launch happens — Pro and foldable now, base model in spring — it changes the shape of the December quarter, not just its size: higher ASP mix, lower unit count, and a revenue line no longer comparable year over year at the segment level. Your Apple initiation is online-only and derives from CONSTELLATION, so the fiscal 1Q27 bridge is straightforward to build both ways, and building it before Wednesday is cheap insurance. The genuinely underwritten question is what a first-time CEO does with disclosure. Apple has spent a decade removing detail, not adding it; a new CEO’s first two prints are the only realistic window in which that reverses, and it is worth writing down what you would want unbundled before he decides.

Space & mobility — 1 named

Cybercab launched, then round-tripped: +5.4% Thursday, −5.9% Friday

TSLA · Event 3 Sep · shares 4 Sep close

Tesla held its Cybercab launch event on Thursday. Shares closed +5.4% at $376.37 into and through it, then −5.9% Friday at $354.03 — a complete round trip in two sessions. The complaint is uniform: no deployment schedule, no production ramp, no regulatory pathway. Future Fund’s Gary Black called the event “largely a bust”; Deepwater’s Gene Munster expects roughly 300 Cybercabs added in Austin over the next month. That is a very wide spread of interpretation over the same 24 hours.

The read: Tesla’s equity is valued on things that never appear in the financial statements, which is exactly where your TSLA initiation’s XBRL traps bite — the depreciation series in company-facts is roughly 18% below the cash-flow line, so any EBITDA built from XBRL alone is wrong, and the per-share series has two split seams. Anyone modelling a robotaxi fleet off public data is almost certainly doing it on a broken depreciation base, and a robotaxi model is entirely a depreciation model — fleet cost, useful life, utilisation. A short, sourced note on “why the public depreciation series cannot support a Cybercab unit-economics model” would land while the debate is live and would cost you a day.

Full-year capex guidance — as currently guided

Where each name stands

On the calendar

Story worth owning

“$725bn” is not a number. Nobody has built a like-for-like hyperscaler capex table — and the definitions moved this quarter.

Every aggregator this week is summing hyperscaler capex to a headline in the $690–770bn range for 2026. Verifying the nine figures in the chart above took a full sweep of transcripts and releases, and the result is that the sum is not meaningful. Microsoft’s number moved to ~$175bn from a ~$190bn expectation without a dollar of spending changing — useful lives went from 15 to 25 years, which pushed future data-centre leases out of finance-lease treatment and therefore out of capex. Oracle guides a net outlay of ~$70bn and says plainly that reported capex will be $20–25bn higher. Intel guides gross and gives no net, while taking roughly $12.2bn of SCIP partner contributions in one half. Meta includes finance-lease principal; Alphabet excludes it. Amazon’s “cash capex” is net of asset-sale proceeds and incentives. Samsung publishes no capex line at all — only >₩110tn of capex plus R&D — and gets dropped into comparison tables anyway. Microsoft has given no FY2027 total, only a quarterly floor, and that gap gets filled with estimates presented as guidance.

The publishable piece is a capex definitions register: one row per company, showing the guided figure, the fiscal period, gross or net, finance leases in or out, the useful-life assumption, what offsets are netted, and what changed in the last four quarters — then a single reconciled series on one consistent basis, with the bridge shown. This is exactly the shape of work your studies already do: the Broadcom note on the quietly dropped Adjusted EBITDA line, the Intel finding that capex appears twice in the cash-flow statement so the XBRL fact understates FY2025 by 17%, the Uber note on a headline figure that changed meaning seven times in nine years. Nobody independent has done it for the capex number the entire AI-financing debate rests on.

The useful-life change alone is worth its own chart. Extending depreciable life from 15 to 25 years lowers reported depreciation for years without touching cash, which flatters precisely the operating margins that the “AI capex is destroying free cash flow” argument is measured against — and it simultaneously shrinks the reported capex number that the same argument counts. Every one of these definitional choices moves the reported picture in the same direction. That is the note.

Sources

1. Bureau of Labor Statistics — Employment Situation, August 2026, released 4 September 2026 (payrolls +162,000; unemployment 4.1%; June revised to +31,000 and July to +21,000).
2. CNBC — “U.S. payrolls rose 162,000 in August, much more than expected; unemployment rate at 4.1%”, 4 September 2026, 08:31 ET (Dow Jones consensus 53,000; hike-odds commentary).
3. TheStreet — Stock Market Today live blog, 4 September 2026.
4. CNBC — “Stocks making the biggest moves premarket: AMC, USA Rare Earth, Lululemon, Adobe & more”, 4 September 2026, 07:43 ET.
5. CNBC — Stock market on 3 September 2026 (S&P 500 close 7,747.71, +1.06%; Nasdaq 26,584.06, +1.4%; Dow 53,686.11, +1.18%).
6. NVIDIA Newsroom — Hugging Face acquisition, 3 September 2026.
7. TechCrunch — “Nvidia confirms it will buy Hugging Face for $12.9 billion”, 3 September 2026.
8. Adobe — “Anil Chakravarthy to Become President and CEO and Shantanu Narayen to Become Executive Chair on December 1, 2026”, 3 September 2026.
9. CNBC — “Adobe names Anil Chakravarthy as CEO, replacing Shantanu Narayen”, 3 September 2026 (Wadhwani departure).
10. Synopsys — FY3Q26 results, 26 August 2026.
11. Broadcom — FY3Q26 results and quarterly dividend, 2 September 2026.
12. TSMC — 2026 monthly revenue (July: NT$467,580m, +44.7% YoY), fetched 4 September 2026.
13. CoreWeave — 2Q26 earnings release (SEC), 11 August 2026 (revenue $2.575bn; revenue backlog ~$104bn at 30 June, +246% YoY, excluding >$25bn added in early Q3).
14. CoreWeave — closes $2.6bn DDTL 5.5 facility, 10 August 2026.
15. Microsoft — FY2026 Q4 earnings call and CFO commentary, 29 July 2026 (calendar-2026 capex ~$175bn; lease and useful-life change; “over $50 billion” for FY1Q27).
16. Meta — Q2 2026 results and CY2026 capex guidance $130–145bn, 29 July 2026.
17. Oracle — Q4 FY2026 earnings call transcript, 10 June 2026 (FY2026 $48bn net / ~$8bn effects; FY2027 ~$70bn net with reported “higher by” $20–25bn).
18. Oracle — record Q4 and FY2026 results, 10 June 2026 (RPO $638bn, +363%).
19. Oracle — FY1Q27 earnings date, 10 September 2026, after the close.
20. Intel — Q2 2026 results, 23 July 2026 (2026 capex >$20bn gross).
21. Micron — FY4Q26 results date, 30 September 2026, announced 26 August 2026.
22. Adobe — FY3Q26 earnings date, 10 September 2026, announced 31 August 2026.
23. MacRumors — “Apple Event Announced for September 9: ‘Surprise and Shine'”, 26 August 2026. Product expectations in that coverage are reporting, not Apple disclosure.
24. CNBC — “Marvell’s stock pops 10% on AI chip deal that lets Google buy up to $12.2 billion in shares”, 19 August 2026.
25. 24/7 Wall St. — “Tesla Falls 3% as Cybercab Launch Leaves Deployment Questions Unanswered”, 4 September 2026.
26. Prices for all 50 covered names and the three indices: Yahoo Finance chart API, queried 4 September 2026 at 15:57 ET / 16:57 BRT, three minutes before the close.

Figures are as reported by the companies and the sources listed above, fetched on 4 September 2026, and have not been independently verified against SEC filings, 6-Ks or audited accounts. This edition was compiled after the US close; price moves are near-closing reads taken at 15:57 ET and may differ marginally from official closing prints. Capex figures sit on incompatible definitions and must not be summed — see the chart notes. Where sources disagreed, both readings are shown rather than one being chosen silently, and unconfirmed reporting is labelled as such. This is a research scan for internal use at CoreValue Research. It is not investment advice, not a recommendation to buy or sell any security, and takes no account of any reader’s circumstances.