The S&P 500 finished Friday at 7,811.54, close to topping the all-time high it set three days earlier, after a week in which the Treasury sold roughly $119 billion of 3-year, 10-year and 30-year paper at yields that haven't been seen in a generation and found a buyer for every dollar. The scare story of the week was the bond market. The better story is that the bond market got its price and cleared. What's left is an inflation print on Wednesday and a question about AI revenue that the equity tape started asking out loud on Thursday.

The Long End Found Its Buyers

Start with the auctions, because they were the test. The 10-year on Wednesday (Oct 7) cleared at 5.3%, the highest auction yield since 2000, and the people who showed up were not the usual shaky hands. Indirect bidders, which include global central banks, took 80.3% against an average near 72%, and dealers were left with just 2.5% of the notes, well below their usual 9%.

Thursday's 30-year was less glamorous but passed. It priced at 5.618%, below where the market expected at the bid deadline, and the bid-to-cover of 2.54 beat the recent average of 2.41. The 10-year, after touching a fresh 24-year high of 5.35% earlier in the session, settled at 5.24% on Friday.

That matters for how we read the whole tape. For a month the worry has been a failed auction, the kind of tail that turns a bad day into a regime. Two sales in a row with real-money sponsorship at 5.3% says the long end has a clearing price, and that price is high but findable. Supply panic is off the table for now, which leaves the Fed and the inflation data as the only live variables.

The Fed, for its part, was unambiguous on Wednesday. The September minutes showed all 19 officials backing the hike to 3.75%-4.00%, and most saw another increase as likely appropriate by year end. What caught our eye was the reasoning: participants cited, among other things, a desire to prevent strong AI demand from driving inflation up further. A central bank that names the capex boom as an inflation input has put the buildout on its watch list, whatever the communications team says.

The market is paying the toll collector and doubting the traffic. Thursday's reported OpenAI revenue reset is the first serious argument for the doubt.

Here is the read nobody on the morning shows will give you. On Thursday (Oct 8), tech slid on reports that OpenAI's annualized revenue was roughly $20 billion lower than previously estimated, yet the chip complex ended the week at records. Nvidia's market cap is now close to $6 trillion. Investors are bidding up the supplier of compute while quietly marking down the revenue of its largest end customer. Those two beliefs can coexist for a while, since the chips are contracted and the backlog is real, but they can't coexist forever at 5.3% on the 10-year.

If a crack comes, in our view it shows up first in the financing layer, among the leveraged renters of compute, before it touches a chip order. SemiAnalysis counts more than 300 neoclouds now renting GPUs, up 55% in under 11 months. That's a lot of balance sheets underwritten when money was cheaper, and the one thing the week made clear is that money isn't cheap now.

What Moved, and What It Means by Pillar

AI chips. The compute layer had a better week than its customers. Nvidia set a record on Monday after Foxconn reported third-quarter revenue up 47%, driven by AI servers, and printed an intraday high of $243.37 on Tuesday. The more interesting detail is the multiple: at Monday's close it traded near 17 times forward earnings, a price that belongs to a boring company, not the most valuable one on earth.

This is a price discovery story, not a thesis change. The contracted, supply-constrained compute thesis is intact, and the guide that already excludes China datacenter revenue means export headlines have limited power over the numbers. Thursday night's TSMC report is the cleanest test of whether the order books are noticing anything.

Platforms and enterprise AI. The OpenAI revision lands on the hyperscalers, who are the other side of the chip trade. One strategist on CNBC's tape described it as the market starting to push back on the wave of AI-linked debt and equity issuance. The capex owners have the cash flow to wait for monetization, but the multiple they are granted for waiting depends on evidence, and this week the evidence got slightly worse.

Power and grid. No single headline here, but the mechanism deserves a sentence because the week was all about it. Higher discount rates hit firm power twice, once on the multiple and once on the cost of financing the plants and wires that datacenter contracts require. The physical bottleneck thesis is unchanged. The valuation tolerance for it is what the 10-year is adjusting.

Energy and hard assets. Brent swung back above $105 on Thursday as Iranian attacks on tankers in Hormuz picked up and Gulf Coast operators prepared for a storm, then gave back ground after President Trump said the US wouldn't attack Iran until after the November midterms. It finished the week near $104. A political calendar is now setting an oil risk premium, which is a fragile thing to rely on.

That oil level feeds directly into the frame the fund has been testing all autumn. The 2-year sits around 4.75% against headline CPI of 3.4% in August, so short real rates are positive and the bills are paying you to wait. In that world, assets that depend on repression, meaning a Fed forced to hold rates below inflation, are being asked to prove themselves on other grounds. We take that tension seriously and haven't resolved it.

Next Week: CPI, Banks, and TSMC

It's a compressed week with the cash Treasury market closed Monday for Columbus Day, so Monday trades with the bond market's price discovery on mute. Don't over-read whatever equities do that day.

Tuesday (Oct 13). JPMorgan, Goldman Sachs, Citigroup and Wells Fargo report. For JPMorgan, analysts expect earnings up 16.4% to $5.90 a share on revenue of $51.3 billion. The group has been weak, with the S&P 500 banks index down 9% over the past month, so the question is whether net interest income and capital markets are strong enough to survive a tough tape. Deposit costs and consumer credit commentary matter more than the headline beat.

Wednesday (Oct 14). September CPI at 8:30 a.m. ET, the last one before the Fed meets. Forecasts point to headline inflation rising from 3.4% to between 3.6% and 3.7%, with core near 2.4%. Bank of America and Morgan Stanley report the same morning, so the sector will be graded and repriced within the same hour.

The setup into the print is asymmetric. Markets give October hike odds of just 17% but December odds of 84% (other sources show roughly 69-78%, which tells you how fluid the pricing is). A hot core number pulls December toward certainty. A soft one doesn't cancel the hike, since the minutes made the committee's intent plain, but it would buy time.

Thursday (Oct 15). TSMC reports third-quarter results in the early hours ET, alongside PPI. Retail sales, jobless claims and the Philadelphia Fed survey all land at 8:30 as well. Aggregate expectations are loud: third-quarter S&P 500 earnings are expected to grow more than 30%, a bar that leaves little room for ordinary results.

Levels. The S&P's record close is the 7,818.93 from Tuesday, with a 52-week intraday high of 7,844.52. The 50-day average is near 7,697, which is the first support that matters. On rates, the week's high near 5.35% on the 10-year is the line. Pimco's Dan Ivascyn told the FT a move toward 6% is feasible, though he also argued that elevated yields should draw buyers and limit how far it runs.

Where Our Thinking Stands, by Pillar

AI chips and hardware: conviction high. The week supports the view that contracted, supply-constrained compute gets paid whatever the downstream arithmetic looks like. What would change our mind is a TSMC outlook that hedges on capex, or evidence that the neocloud and AI-debt layer can't refinance, because that is where a demand shortfall would first become an order shortfall.

Platforms and AI software: conviction moderate, and a little lower than a week ago. The hyperscalers can fund the buildout from operating cash flow, but the OpenAI revision weakens the argument that monetization is arriving on schedule. We want to see enterprise and cloud revenue growth hold through the coming prints before treating this as noise.

Power and grid: conviction in the bottleneck, caution on the multiple. Firm power and the wires are the physical constraint on every datacenter plan, and nothing this week altered that. A 10-year that holds above this week's high would keep the valuation pressure on regardless of the fundamentals.

Financials and the rate factor: watching. Next week's bank prints are a read on how the rate regime is working through deposits, lending and capital markets, which is more informative than any single stock. A sector that falls 9% in a month on a hawkish Fed and then beats is telling you the fear was priced ahead of the fundamentals.

Hard assets and the repression frame: unresolved. The thesis has a clear falsifier, and it's a Fed that hikes and says it isn't done. Positive real short rates and a firmer dollar are the conditions under which these assets struggle, and the minutes did nothing to change that. A soft CPI with a patient Fed would reopen the case. Another hot print would close it for a while.

Energy as macro input. Brent near $104 is feeding the inflation numbers the Fed is reacting to, and the political promise of no strike before November is a delay, not a resolution. Watch Hormuz tanker traffic and freight rates, because those move before the price does.

This document is for informational purposes only and does not constitute investment advice, an offer, or a solicitation to buy or sell any security. Nova One Capital, the Nova X Convergence Fund and affiliated persons may hold or trade securities discussed. Market data is sourced from public reports as of the close on October 9, 2026 and is subject to revision. Past performance is not indicative of future results. Published October 10, 2026.