Thursday, September 3, 2026, written at 9:05 a.m. Eastern, after the Challenger release and before jobless claims and the cash open, so every US number below is a future or a pre-market print and can look different by ten o’clock. Four companies from four different layers of the same supply chain reported inside twenty-four hours. All four beat. All four raised. One of them was sold.
Top Call: The Whole Stack Reported and Only the Discount Rate Is Still in Dispute
Broadcom delivered fiscal third quarter revenue of $29.59 billion, up 86% year on year, with adjusted earnings of $3.32 against $3.24 expected and AI semiconductor revenue of $16.7 billion, up 221% year on year and 54% sequentially. Hock Tan then did something no one else in this cycle has done: he put dated numbers on the next two years, guiding fiscal 2026 AI revenue to $58 billion, fiscal 2027 to $115 billion and fiscal 2028 to $230 billion. He also named Anthropic as the company’s largest custom silicon customer from 2027 onward, with roughly one gigawatt of deployment this year, five gigawatts next year and more than ten gigawatts incrementally by 2028.
The stock is down about 3.5% pre-market at roughly $354. The reason is that October quarter revenue was guided to $34.8 billion against a $35.03 billion consensus. That gap is $230 million on a $35 billion base, or seven tenths of one percent, and Barclays reads it as supply constrained rather than demand constrained.
Hold those two facts next to each other, because together they are the whole market. A company being valued on a 2028 revenue line was marked down for missing a 2026 quarterly line by less than one percent. That is not a market disputing the AI story. It is a market that has stopped paying for the story and started charging for the wait, which is the same thing the ten-year at 4.81% has been saying for three days. Dell settled the demand question on Wednesday, Broadcom has now put a two-year quantity on it, Ciena confirmed it in the transport layer this morning and Snowflake confirmed it in consumption. What remains in dispute is not whether the revenue arrives. It is what that revenue is worth discounted at five percent and change.
The Overnight Scoreboard
| Company | Guide | Tape | |
|---|---|---|---|
| Broadcom FQ3 | Rev $29.59B (+86%), EPS $3.32 vs $3.24, AI $16.7B (+221%) | Q4 $34.8B vs $35.03B; AI $115B FY27, $230B FY28 | -3.5% pre-market |
| Snowflake FQ2 | Rev $1.55B (+35%) vs $1.48B, EPS $0.62 vs $0.45 | FY product rev to $6.07B, op margin 14.5% from 13.5% | +23% to about $374 |
| Ciena FQ3 | Rev $1.67B (+37%), EPS $2.11 vs $1.72, optical $1.19B (+46%) | FY26 rev raised to $6.42B, up 35% | +4.3% pre-market |
| Dell FQ2 (Wed) | Rev $47.0B (+58%), EPS $7.04 vs $4.90, backlog $95B | FY27 raised $25B to $192B | +15.8%, -1.3% today |
Four beats, four raises, and roughly nineteen percentage points of dispersion between the best and worst reaction. The tape is no longer trading the sector. It is trading each print against its own expectations bar, and those bars are set by how far the stock has already run.
Broadcom: Where the Value Actually Sits
The line that matters is not the quarter. It is the customer disclosure. Six core custom silicon customers now include Anthropic, Google, Meta and OpenAI, and Anthropic overtakes the incumbents in 2027. That is the second time in three weeks that Anthropic has surfaced as the marginal buyer of compute at scale, after its own revenue disclosure in mid-August, and it reshapes the concentration risk rather than removing it. Broadcom’s forward revenue is now underwritten by a small group of private companies whose own funding depends on capital markets that are actively repricing duration.
TD Cowen kept its buy rating and flagged precisely that: customer concentration and financing. This is the more honest bear case than the guide miss. Gross margin at 76.3% deserves the same attention, because custom silicon carries a lower margin than networking, so a mix shift toward $230 billion of XPU revenue is also a mix shift away from the margin profile that earned Broadcom its multiple. Growth of this magnitude with a falling margin is still an excellent business. It is simply a different business from the one the market has owned since 2023, and at 30 times forward earnings the market has not yet decided what to pay for the new version.
Ciena Confirms the Light Layer
Ciena reported before the bell with adjusted earnings of $2.11 against $1.72 expected on revenue of $1.67 billion, up 37%. Optical networking alone reached $1.19 billion from $815.5 million a year ago, a gain of 46%. Adjusted gross margin rose to 46.4% from 41.9% and adjusted operating margin more than doubled to 22.5% from 10.7%. Full-year revenue guidance moved up to roughly $6.42 billion, growth of about 35%.
The margin expansion is the interesting part. Optical transport has been a commodity business for two decades, and Ciena just added four and a half points of gross margin while growing 37%. Pricing power in the interconnect layer means the bottleneck has moved out of the accelerator itself and into moving data between racks and between buildings. It also corroborates Broadcom’s networking commentary from the opposite end of the same wire. When two independent companies at different points of one physical link report the same acceleration, that is a supply chain signal rather than a company signal, and it is the cleanest new information in the overnight.
The Macro Went the Other Way Overnight
Every asset that had been punished for three days reversed. The ten-year eased to 4.784% after touching 4.814% on Tuesday, its highest since late 2023. The dollar index slipped to 99.545. Gold, which had fallen for five consecutive sessions, rallied roughly 2%, with spot back above $4,400 and the front future at $4,540.60. Bitcoin added about 1.9% to $78,548. The VIX fell to 15.03. Crude handed back part of a three-day rally as Brent slipped toward $95 from $97.
The proximate cause is Trump, who said of the renewed Iran campaign, “I don’t think too long,” and asserted that the United States controls the Strait of Hormuz after overnight strikes on radar and mine-laying capability along Iran’s southern coast. That was the second round of strikes in three days. Shipments through the strait had recovered to roughly half of pre-war volumes during the August lull, and the market is now pricing a short campaign rather than a closed waterway. That is a reasonable base case and a fragile one, since the same assumption was made and broken twice already this year.
The gold reversal contradicts something this note argued yesterday. Bullion had fallen through a hawkish Fed chair, a shooting war and a global bond rout, and I wrote that nothing of the debasement thesis survived that sequence. One session later it is up two percent on a weaker dollar and softer yields. The correct reading is narrower than either version: gold has been trading the real rate and the dollar all week, not the geopolitics. When yields backed off, it went straight back up. That is a rates instrument behaving exactly like one.
Japan Is the Other Half of the Rout
The yen jumped more than 1% to 156.34 per dollar, its strongest in a month, and thirty-year JGB yields fell as much as ten basis points. Board member Takata said the bank should move “nimbly” against rising inflation and hinted at a faster pace than its recent semiannual cadence. Overnight index swaps now imply a full 25 basis point hike at the September 18 meeting and roughly 44 basis points of tightening by year end.
This matters well beyond Tokyo. A yen that strengthens on a credible hike is the mechanism by which Japanese capital comes home, and Japanese institutions remain among the largest foreign holders of long-dated US and European government paper. Monday’s synchronised global steepening and today’s synchronised relief both run through that same funding channel. If the Bank of Japan delivers on September 18 and the yen holds this level, pressure on the US long end resumes regardless of what the Federal Reserve does two days earlier. The bond market has spent this week treating the Fed as the variable; Japan is the one with the larger unhedged position behind it.
A Labour Market That Neither Hires Nor Fires
Challenger counted 52,881 announced job cuts in August, down 38% year on year and the quietest August since 2022, though up 58% from July’s 33,429. Restructuring accounted for 16,173 of them. Artificial intelligence was blamed for only 3,462, the lowest monthly total since late last year, which is a useful corrective to a narrative running well ahead of the data. Year to date announcements total 529,914, some 41% below the equivalent 2025 stretch.
Set that beside yesterday’s ADP print of 38,000, the weakest since January. Firing has nearly stopped and hiring has nearly stopped with it. That configuration describes a frozen labour market rather than a deteriorating one, and it has two consequences. First, it widens the distribution around tomorrow’s payrolls considerably, with consensus itself split between roughly 60,000 on one caliber and closer to 90,000 on another, a spread that is unusual this late in the week. Second, it is a poor argument for a rate hike, because a September move is aimed at an inflation problem the labour market is not currently generating. Roughly two thirds odds on that hike look high to me against this evidence, and Waller speaks at 8:30 this morning on exactly that question.
Asia and Europe
Tokyo closed marginally higher, the Nikkei up 0.21% to 64,461.68 while the broader Topix gained 0.91%, the index held back by chip names reacting to Broadcom after hours. Advantest fell 0.8% and Fujikura 1.79%, while SoftBank added 2.9% and Tokyo Electron 0.8%. Nitori jumped 6.62% and Fast Retailing fell 3.58%. Korea was the standout, the Kospi climbing 1.4% to 6,656.81 on Samsung and SK Hynix. Hong Kong edged 0.1% higher, Shanghai gained 0.4%, Taiwan added 0.6% and Australia 0.5%.
The Topix outperforming the Nikkei by seven tenths of a point is worth a line of its own. That is the domestic, value-weighted, bank-heavy index beating the exporter-heavy price-weighted one on a day the yen strengthened over a percent. Japanese banks are the direct beneficiaries of a September hike, and Japanese equities are quietly rotating in exactly the way a market that believes the Bank of Japan would rotate.
Europe opened mixed to slightly softer, the DAX around 26,331, the FTSE near 10,833 and the CAC 40 at about 8,675, with the Stoxx 600 little changed. The pattern of the past fortnight holds: Europe is a bond market with an equity market attached, and with Bunds and gilts finally pausing, European equities simply stopped falling rather than rallying.
Key Events Today and Ahead
| When | Event | Why it matters |
|---|---|---|
| Thu 8:30am | Jobless claims; Waller on inflation | 205K expected vs 203K; Waller is the hike swing vote |
| Thu 8:30am | July trade balance | -$90.2B expected from -$73.3B, a large widening |
| Thu 10:00am | ISM services | 54.1 to 54.5 expected; services is where inflation lives |
| Thu, close | Zscaler, Lululemon, DocuSign, Samsara | Zscaler’s FY27 ARR guide is the line that carries information |
| Fri 8:30am | August non-farm payrolls | Consensus split 60K to 90K by caliber, the week’s real number |
| Sep 15-16 | FOMC | About 66% priced for a hike |
| Sep 18 | Bank of Japan | 25bp now fully priced, yen at 156.34 |
| Sep 24 | US-China summit in Washington | Tariffs and Taiwan both on the agenda |
Accountability
Yesterday’s note called Broadcom’s fiscal 2027 AI target “the only line that moves the stock.” The target landed at $115 billion, comfortably above the “in excess of $100 billion” previously guided, with a 2028 figure of $230 billion added on top. The stock fell anyway on a quarterly guide that missed by less than one percent. The call identified the right line and drew the wrong inference from it, and the lesson is one this note has now learned twice in three days: in a market repricing duration, a longer dated number is worth less than a nearer one, which is exactly backwards from how these companies are valued in most sell-side models.
The gold call from Tuesday and Wednesday is the second correction. I treated five consecutive down sessions through a war, a hawkish chair and a bond rout as evidence that the debasement bid was gone. It was evidence that gold trades the real rate, which then fell and took gold two percent higher within a session. The narrow claim holds and the broad one did not.
Wednesday’s read on the tape has held up. That note argued the market was no longer disputing AI numbers but repricing what they are worth against a rising risk-free rate. Broadcom this morning is that argument in its purest available form.