Tuesday, October 27, 2026, written at 9:50 a.m. Eastern, twenty minutes after the open. Equity marks below are early and will move. Futures carried a modest bid into the bell with the Nasdaq contract leading, crude is lower for a second session despite a missile strike on a Saudi refinery, and volatility is being sold ahead of a Federal Open Market Committee decision that nobody expects to change the policy rate. Microsoft and Alphabet report tomorrow afternoon, roughly two hours after Chair Warsh speaks. September core personal consumption expenditures land Thursday.
Top Call: We Finally Have Audited Numbers for a Frontier Lab, and They Cut Both Ways
Anthropic filed its public S-1 yesterday, seeking to list Class A stock on Nasdaq under the ticker ANT. The revenue line is the part that should end an argument. First-half 2026 revenue was about $16.3bn against roughly $1.2bn in the same period last year, second quarter revenue alone was near $11.6bn, and the company turned its first quarterly operating profit of about $974mn. Net revenue retention is reported near 900%. For two years the bear case on frontier labs has rested on the claim that enterprise adoption was a pilot phenomenon that would not compound. These are not pilot numbers.
The headline $80.7bn first-half net loss is mostly not a loss in any economic sense. Roughly $72.4bn of it comes from fair-value remeasurement of convertible notes, derivatives and simple agreements for future equity, liabilities that grew precisely because the equity underneath them appreciated. Anyone trading off that number without reading the reconciliation will be trading the wrong way.
What should actually move the complex is the obligation side. Reported totals for future compute commitments range from about $542bn to $810bn depending on what each outlet counts, with roughly $630bn the most frequently cited figure and about $441bn of that described as guaranteed. Against cumulative revenue well under $30bn, that is a contractual bill somewhere between six and eight times everything the company has ever billed. The S-1 does not make that bill safe. It makes it legible, and the market now has to price it with numbers rather than guesses.
The second filing of the last twenty-four hours is the one we would spend more time on. Bank of America, Citi and Morgan Stanley are arranging roughly $43.5bn of senior first-lien debt, about 69% of a $63bn package, with Blackstone providing the committed second lien. A newly created special purpose vehicle buys Broadcom’s XPU accelerators and networking equipment and leases the hardware to Anthropic, with deliveries beginning in 2027. Broadcom provides what is described as residual value support, which functions as a guarantee for the senior lenders. The Wall Street Journal reports that some banks considered walking away and that Hock Tan accepted higher interest rates to close it.
Read the two documents together and the structure of this cycle is now explicit. The operating company publishes audited accounts and a transparent capital structure. The financing for the hardware those accounts depend on sits in a vehicle that neither the chip vendor nor the customer consolidates, credit-enhanced by the vendor’s own balance sheet, and syndicated to banks that had to be paid up to stay in the room. We wrote yesterday that the artificial intelligence build was refinancing itself out of debt and into equity. Today’s correction to that view is narrower and more uncomfortable: at the top of the quality curve it is moving into equity, and one layer down it is moving off balance sheet entirely. Moody’s framing is the right one, that Broadcom is among the strongest large technology credits and that the risk compounds if several of these are done in quick succession. The first one is fine. The fourth one is a sector.
Broadcom is up around 2.3% near $336.71 after DA Davidson moved to Buy with a $400 target from $350. Nvidia is committing $3bn to Meta’s equity programme, which was raised overnight to $97.5bn, with AMD and Broadcom together indicating up to $5bn. Note what that means in practice. Broadcom is simultaneously guaranteeing the senior debt that buys its own chips for one customer and subscribing to the equity of another. Both legs are defensible on their own. Together they describe a supplier underwriting its own demand in two different parts of the capital structure, and that is the circularity problem we have been flagging since August, now with the paperwork to prove it.
Celestica Is the Print That Matters, and It Tells You What the Build Costs
Celestica is up about 16% pre-market on the quarter of the season so far. Revenue of $5.83bn rose 83% year on year and cleared a guidance range of $5.25bn to $5.55bn. Adjusted earnings of $3.30 beat guidance of $2.88 to $3.08 and more than doubled from $1.58. Compute and storage revenue grew 242% against guidance near 190% as hyperscaler accelerator racks ramped. The hardware platform solutions business did $2.3bn, up 70%, on 1.6 terabit and 800 gigabit switch programmes across several hyperscalers. Full-year revenue guidance went to $21.0bn from $20.5bn, adjusted earnings to $11.55 from $11.30, and management set a 2027 revenue target of $40bn.
| Metric | Guidance | Actual | Result |
|---|---|---|---|
| Revenue | $5.25bn to $5.55bn | $5.83bn, up 83% | Clear beat |
| Adjusted EPS | $2.88 to $3.08 | $3.30 | Clear beat |
| Compute and storage | About +190% | +242% | Large beat |
| 2026 revenue guide | $20.5bn | $21.0bn | Raised |
| 2026 free cash flow | $600mn | $460mn | Cut |
Our take is that the two guidance revisions matter more than the beat, and they point in opposite directions. A 2027 revenue target of $40bn against $21bn this year is not a forecast of growth, it is a statement that orders already exist for something close to a doubling. At the same time free cash flow guidance was cut to $460mn from $600mn, and adjusted gross margin fell 50 basis points to 11.2% on mix even as operating margin expanded to 8.5%. That combination is the honest arithmetic of this cycle in one company. Serving hyperscaler compute at scale consumes working capital faster than it generates cash, and the mix shift toward racks dilutes gross margin while operating leverage covers it up further down the statement. Celestica is being rewarded this morning because the revenue is visible and the customers are solvent. The names one tier below with the same working capital profile and weaker customers are not getting that treatment, and should not.
Cadence confirmed the same demand picture last night without the drama. Revenue of $1.633bn grew 22%, non-GAAP earnings were $2.11 against $1.93, backlog reached a record $8.2bn and the full-year outlook went to 21% growth. The stock added roughly 4%. Design software backlog is a forward read on silicon that has not been taped out, so a record there is a statement about 2027 and 2028 rather than this quarter. Amkor was the weaker half of the pair, with revenue near $1.99bn and earnings of $0.76 a touch below consensus, though gross margin expanded 150 basis points sequentially to 18.3% and fourth quarter guidance of $1.9bn to $2.0bn with margin of 18.0% to 19.0% holds the line. Advanced packaging remains the physical bottleneck behind high bandwidth memory, and the lack of upside there is worth watching.
The real test is tonight. SK hynix reports Wednesday morning in Korea, which is this evening Eastern. Consensus looks for roughly 98.2 trillion won of revenue and about 77 trillion won of operating profit, a margin near 78% and a sixth consecutive record. Two things could spoil it. Bernstein estimates Samsung generated $8.7bn of high bandwidth memory revenue in the quarter against SK hynix at $7.6bn, which would mean Samsung has taken the lead for the first time, and there is reporting that SK hynix shipments fell short for a second straight quarter as Nvidia’s Rubin platform slipped. A record quarter that comes with lost share and a delayed ramp is not the print the memory complex needs, and the unresolved $12.1bn special dividend question adds another reason for the stock to trade on something other than the headline.
Rates: The Decision Is Settled, the Statement Is Not
Tomorrow at 2:00 p.m. Eastern the committee is near certain to hold at 3.75% to 4.00%, with futures around 90% on no change and under 10% on a hike. December is the live meeting, priced near 75%, and there is no Summary of Economic Projections at this one. That removes the dot plot as a channel and concentrates all the information in the statement language and in whether Hammack, Kashkari and Logan repeat their hike dissents.
This morning’s data gave the hawks very little. September durable goods orders rose 1.8% against 1.7% expected, but nondefense aircraft orders jumped 28.4% and did most of the work. Excluding transportation the gain was 0.5%, and core capital goods orders, the series that actually proxies business investment, also rose just 0.5%. Shipments were up 0.9%. Year to date orders are 8.0% higher, which is a genuinely firm number, though a large share of it is aerospace backlog conversion rather than broad capital spending. Consumer confidence for October arrives at 10:00 a.m. with consensus near 80.5 after September’s 81.9, the weakest reading since 2014, alongside September new home sales. Richmond and Dallas Fed surveys follow.
The curve is doing something mildly interesting into this. The two-year is a basis point lower near 4.75% while the ten-year sits right at 5.00% and the thirty-year is unchanged at 5.30%. A small steepening ahead of a hold is consistent with a market that has fully accepted the near-term outcome and is beginning to think about term premium and supply rather than the next move. We would not build much on two basis points, but it is the opposite of the flattening we flagged yesterday, and if it persists through the $70bn five-year auction today and $44bn of sevens Thursday it becomes a position rather than noise. Thursday’s core personal consumption expenditures print, tracked near plus 0.36% month on month, remains the number that decides December.
Energy: A Refinery Was Hit and Crude Fell
Missiles struck the Ras Tanura refinery, a 550,000 barrel a day facility, and Saudi authorities say the resulting fire was contained. Saudi-led forces intercepted further Houthi missiles and drones, explosions were reported near Prince Sultan Air Base, and the Houthis claim another attack on Riyadh airport. Saudi airstrikes hit the port of Mokha. WTI is down 0.82% at $100.93 and Brent down 0.65% at $110.02.
A strike on a half-million-barrel refinery that moves crude lower is the clearest possible confirmation of the argument we have run for three weeks. The premium in this market is attached to transit through Hormuz and to the option value of escalation, not to refining capacity or to barrels already lost. Refining is not the constraint here. Getting cargoes out of the Gulf is the constraint, and nothing that happened overnight changed that calculus in either direction.
What did move the probability is diplomatic. Iran’s ambassador to South Korea restated the seven-day plan, which links reopening the Strait of Hormuz to Washington accepting Tehran’s conditions, the same roadmap the United States rejected in late September. Qatari, Saudi, Turkish and Pakistani officials have been in Tehran. President Trump continues to frame the choice publicly as a deal or finishing the job, and CBS reports that strike plans involving a Massive Ordnance Air Blast weapon are prepared if ordered. Shipping attacks in the strait are at a record. OPEC+ is expected to leave December production targets unchanged at Sunday’s meeting, and the administration is reportedly easing methane rules on oil wells, neither of which changes near-term supply.
Our read is that the two sessions of softness are the market pricing a modestly higher chance of a negotiated reopening, from a very low base, and nothing more. The midterms are a week away. The reporting still points to a campaign that resumes afterwards. We would treat $100 as a floor built on optionality rather than a level that reflects improved fundamentals, and we would note that effective spare capacity near a quarter of a million barrels a day means any actual supply loss prices violently from here.
Earnings Elsewhere: The Old Economy Cleared and Boeing Did Not
| Company | Result | Versus Street | Read |
|---|---|---|---|
| Coca-Cola | Revenue $13.4bn up 7%, adjusted EPS $0.91 | Beat on both, guidance raised | Pricing power intact |
| UPS | Adjusted EPS $1.69, revenue $22.9bn | Beat on both, full year raised | Volumes better than feared |
| PayPal | Non-GAAP EPS $1.45, revenue $8.58bn up 2.0% | EPS beat, revenue growth thin | Margin story, not growth |
| Boeing | Revenue $24.19bn up 4%, adjusted loss $0.34 | Missed both, backlog $745bn | Loss narrowed to $116mn |
| HSBC | Revenue $18.62bn, pre-tax profit $9.6bn up 32% | Small miss on both | Down 6.5%, $1bn buyback |
Coca-Cola raising organic revenue guidance to at least 5% and earnings growth to at least 11% is a better defensive print than the market has been getting from staples, and it argues that volume elasticity is still manageable at these price levels. UPS beating on both lines and raising full-year guidance is the more interesting macro signal of the two, because parcel volume is a reasonably honest read on goods demand and the consensus going in was defensive. PayPal is the mirror image, delivering earnings on cost control while revenue grew 2.0%, and a payments business growing at two percent in nominal terms is losing share to something.
Boeing is the one to sit with. Revenue of $24.19bn missed a $24.39bn consensus and the adjusted loss of $0.34 was worse than the $0.21 expected, yet the net loss narrowed to $116mn from $5.34bn a year ago and the backlog reached a record $745bn. That is a company whose order book is no longer the problem and whose delivery rate still is. Note also that nondefense aircraft orders rose 28.4% in this morning’s durable goods report. The demand exists. Converting it remains the thesis, and another quarter has passed without the conversion improving enough to matter.
HSBC falling 6.5% on a modest revenue and pre-tax miss, despite 32% profit growth and a $1bn buyback, is a useful read on how little patience the market has for European bank disappointment at current valuations. It dragged the FTSE 100 down about 1% while the rest of Europe held firm.
Asia and Europe: Seoul and Tokyo Bought the Filing, China Did Not
Asia split cleanly along artificial intelligence exposure. The Kospi rose about 3.1% to 5,638.34, having been up more than 4% intraday, and the Nikkei gained 1.85% to 63,365, both led by technology on the Anthropic filing and the Broadcom financing. Taiwan added 0.12% to 46,096. The other direction was just as clear, with the Hang Seng down 0.73% at 23,041.32 and the CSI 300 off 1.32% at 4,049.55. Chinese equities have no clean way to express a Nasdaq listing by an American frontier lab, and the Fifth Plenum that opened yesterday is a party governance and personnel meeting rather than a policy one, so there is no domestic catalyst to offset. The yen remains soft near 157.76 into Friday’s Bank of Japan decision.
Europe opened lower on HSBC and has recovered through the morning. The Stoxx 600 is marginally higher at 641.04, the DAX up 0.25% at 25,473.86, and the FTSE 100 is the clear laggard at around minus 1% on banks. A gilt auction of GBP 4.5bn of 2032 paper cleared at an average yield of 5.026% with a 3.51 times cover, which is adequate rather than strong. Bunds are firmer with the ten-year at 3.351%, and the dollar is up 0.26% at 102.22 with the euro at 1.1142 and sterling at 1.3320.
Also On The Tape
- Israel votes today in a Knesset election, with exit polls at 4:00 p.m. Eastern and a final tally expected by October 30. The last Channel 12 poll had Netanyahu’s bloc at 50 seats against 56 for the opposition, with neither near the 61 needed. A deadlock is the base case, and given that Israeli policy is an input to the single largest risk premium in commodities right now, this is not a local story
- Gold is down about 0.5% near $4,009 after briefly trading below $4,000, with silver off 1.4% around $53. Gold slipping while the dollar firms and crude falls is internally consistent for once, and reads as a hawkish Fed being priced rather than an inflation hedge being bought
- Bitcoin is $80,088, up 0.66%, and has now spent several sessions pinned either side of $80,000 with no evident sensitivity to either the artificial intelligence tape or the rates tape
- Copper is up 0.8% with three-month metal near $14,400 a tonne, which continues to be the most straightforward long in the electrification and data centre build and the least discussed
- Tesla is up around 1% after Monday’s 5% gain on the Barclays upgrade. The upgrade thesis rests partly on a possible SpaceX combination that has not been announced, and we would keep separating the industrial logic from the valuation argument
- Visa, Robinhood, Seagate and NXP Semiconductors report after the close. Seagate is the one we care about, because hard disk pricing and nearline capacity demand is the least glamorous and most direct read on whether storage is still the tightest link in the artificial intelligence supply chain
- Russia and Ukraine saw a partial de-escalation around Kherson as a ceasefire began in Oleshky, offset by strikes on a Black Sea cargo ship, a Kyiv drone facility and transport infrastructure in Kharkiv
- Intel reports Thursday after the close and is trading near a three-month low going in