Two inflation reports in two days, both cooler than feared. The market did what it does with that combination — and then couldn’t quite hold the round number it had just tagged.
The S&P 500 closed at a record 7,798.99, up 0.65%, after crossing 7,800 for the first time in its history intraday and printing an all-time high of 7,816.79 before easing back. It surpassed the 7,757.64 close set on August 7. The Nasdaq Composite rallied 0.81% to 26,803.03, led by the chip complex, while the Dow Jones Industrial Average managed just 0.13% to 53,839.99 as Cisco’s collapse capped its gains. The VIX settled at 14.44.
The catalyst was July’s producer price index, which came in unchanged month-over-month against forecasts for a 0.2% rise, with the annual rate decelerating to 4.7% from June’s 5.5%. Stacked on Wednesday’s in-line CPI, it pushed the odds of a September rate hike down from roughly 39% to 30%.
But the report carried a detail worth understanding, because it complicates the clean disinflation story — and because it is about to distort the Fed’s preferred inflation gauge in a way the Fed itself will likely look through.
Index Performance
Rates, Dollar, Commodities
The Portfolio Management Quirk
Beneath the friendly headline, core PPI excluding food, energy, and trade services accelerated to roughly 0.35–0.4% month-over-month, the fastest since May and about four times June’s pace. The single largest driver was portfolio management fees, which surged 6.5% in July — the biggest jump in over a year — and are now up 22.5% year-over-year. Transportation and warehousing services fell 1.8%, and airfares declined, partly offsetting.
Here is why that 6.5% matters, and why it shouldn’t. Portfolio management fees are charged as a set percentage of assets under management, which means the PPI reading tracks the level of the stock market almost mechanically. Investors didn’t agree to pay a higher rate for advice in July; their portfolios simply became more valuable, so the fees measured in dollars rose. As Fifth Third’s chief US economist Bill Adams put it, this price category tracks closely with stock market indexes — and the market is way up over the last twelve months.
The complication is that portfolio management flows directly into core PCE, the Fed’s preferred inflation measure, under current methodology. Analysts accordingly raised their estimates for the July core PCE release on August 26. But the distortion is well understood: Adams noted that the upward contribution will be revised down on September 30, when the BEA makes annual revisions that render the calculation less sensitive to the level of the stock market. A hawkish core PCE print driven by rising asset prices is unlikely to change many minds at the FOMC.
Adams’s summary is the fair one: the report doesn’t change the big picture — inflation is still too high, core is running below headline, and the picture for both improved in July. The CPI and PPI reports together keep a narrow path open for a September hold, though August’s data will arrive before that decision and gets the final word.
Oil Falls — but for a Reason Worth Noting
Crude slid roughly 2–3%, with WTI settling at $81.25 and Brent at $87.07, adding to the disinflationary mood and helping fuel the rally. The headline reason, however, was not a supply development: the International Energy Agency cut its 2026 global demand forecast by 1.6 million barrels per day.
That distinction matters. Falling oil driven by abundant supply is unambiguously good news — cheaper energy, cooler inflation, healthier margins. Falling oil driven by weaker demand is the oil market pricing a slower global economy. Thursday’s tape took the disinflation and largely ignored the demand signal. With retail sales due Friday, that asymmetry is worth watching: the same cheaper oil that lifted stocks could start reading as a growth warning if consumption data softens.
Separately, the Trump administration stated it retains “total control” over the Strait of Hormuz, disputing the private vessel-tracking data that has shown depressed shipping traffic — a continuation of the data dispute running since earlier this week.
Labor Data and the Fed Chorus
Initial jobless claims rose 9,000 to 209,000 for the week ended August 8 — still near historic lows — while continuing claims fell 22,000 to 1,777,000. The labor market is cooling at the margin without any sign of rupture.
The Fed speakers split predictably. Richmond’s Tom Barkin said the signs of easing inflation provide sufficient grounds to hold rates steady, while cautioning that renewed tightening remains possible if some price pressures prove entrenched — though notably, Barkin is not a voter this year. Cleveland’s Beth Hammack, who is, reaffirmed her hawkish stance, expressing doubt that the recent disinflation will persist and repeating her call for a hike now.
Despite three consecutive data points cutting the same way — July payrolls at -23,000, CPI at 3.4%, PPI at 0.0% — Fed watchers still broadly expect at least one hike by year-end. Jackson Hole at the end of August is the next scheduled test of whether the committee reads the data as the market does.
The Long End Isn’t Convinced
For all the front-end relief, the Treasury’s 30-year auction told a different story. The $25 billion sale priced at 5.216%, tailing the 5.212% when-issued yield by 0.4 basis points and marking the highest auction yield since 2001. The bid-to-cover ratio came in at 2.392 versus the 2.429 average of the prior six auctions, and indirect bidders — the proxy for foreign demand — took 66.8% against a 67% average.
None of those numbers is alarming in isolation, but together they describe soft demand at a record yield. Cooling inflation data has pulled short-term rate expectations lower; it has not made investors eager to lend the government money for thirty years. With hyperscaler bond supply still building, the long end remains this market’s most persistent unresolved risk.
SanDisk’s Investor Day: HBF Steals the Show
SanDisk surged double digits after an investor day that delivered both financial confidence and a genuinely interesting technology story, dragging Micron up about 5.6% and lifting the broader memory complex with it.
On the financials, management guided to gross margins holding at 80% through 2030, with fiscal revenue growing in the mid-to-high teens annually from fiscal 2028 through 2030. It disclosed long-term agreements with eight customers — three of them US hyperscalers — whose volumes will absorb roughly half of fiscal 2027 storage output and two-thirds of fiscal 2028. On capital returns, the company committed to distributing 100% of free cash flow to shareholders, anchored near-term by a $15.5 billion buyback.
The technology headline was HBF — High Bandwidth Flash — with SanDisk confirming it has completed its first prototype and is working to deliver initial samples to customers next year. Management’s framing was striking: four HBF-equipped GPUs could match the performance of eight HBM-equipped GPUs.
That claim deserves unpacking, because it is real but narrower than it sounds. HBF applies HBM’s architecture — dies stacked vertically, connected by through-silicon vias, seated beside the GPU on the interposer — to NAND flash rather than DRAM. The payoff is capacity: a 16-die HBF stack delivers 512GB against 36–64GB for an HBM4 stack, roughly 8 to 16 times the capacity at similar cost, or about a tenth the cost per gigabyte. Read bandwidth for the first generation runs near 1.6 TB/s, faster than HBM3E but slower than HBM4’s 2.0–3.3 TB/s.
The “half the GPUs” math follows from capacity, not compute. Large models — particularly trillion-parameter mixture-of-experts architectures — must be split across multiple accelerators simply because the weights don’t fit in available memory, and every split adds interconnect latency and cost. Give each GPU ten times the local memory and you need fewer of them to hold the same model. But halving the GPU count also halves the available compute, so the argument holds only for workloads bottlenecked by memory capacity rather than by arithmetic throughput.
The physics carries real constraints, too. NAND is, in the words of Objective Analysis’s Jim Handy, atrociously slow for writes though reads can be coaxed to go fast — which is why HBF targets model weights, which are read repeatedly and written rarely, while the write-intensive and latency-sensitive KV cache stays on HBM. SanDisk’s own simulation on Llama 3.1’s 405-billion-parameter model showed reading pretrained weights landed within 2.2% of hypothetically unlimited-capacity HBM, a specific and carefully scoped result. Accordingly, SanDisk and SK Hynix have published HBF through the Open Compute Project as an open standard explicitly designed to coexist with HBM, not replace it. First memory samples are expected in the second half of 2026, with inference devices built around HBF following in early 2027.
For memory investors, the significance is straightforward: if HBF works as specified, it expands the addressable market for NAND into territory currently reserved for DRAM — and it does so at a moment when HBM capacity is the binding constraint on AI inference economics.
Cisco’s 30-Year Demand Signal Isn’t Enough
Cisco fell 8.40%, its worst single-day decline in six months, after AI data center revenue guidance failed to clear investors’ elevated bar. The irony was sharp: CEO Chuck Robbins told the Wall Street Journal that the company had not experienced this level of demand across its entire portfolio in three decades, with billions of dollars of orders arriving from megacap technology firms building out AI systems.
Thirty-year-high demand, and the stock dropped more than 8%. The explanation is positioning — Cisco had rallied more than 60% year-to-date heading into the print, and expectations had run far ahead of even excellent results. As a Dow component, its decline single-handedly capped the blue-chip index’s advance. Cerebras Systems extended its slide on continued disappointment over its hardware segment.
AI Financing, Software’s Comeback, and Pershing’s Bet
AMD issued $4.75 billion in corporate bonds across four tranches with maturities from three to ten years, earmarked for working capital with possible debt repayment — the latest chipmaker to tap credit markets, following Intel’s upsized equity raise and Alphabet’s bond plan.
Software staged a powerful rebound, and the catalyst was a deal. Workday soared 17.78% on reports that private equity firm Silver Lake is in acquisition talks. The read-through lifted the entire sector, with the strongest gains concentrated in recently beaten-down names — Salesforce, Adobe, Datadog, and Intuit among them. When private capital starts hunting in a sector, the market reprices everything nearby for the possibility of being next.
Netflix rose 5.4% after Bill Ackman’s Pershing Square disclosed a new stake — a vote of confidence in a stock that had been sliding on viewership concerns and the sector’s neglect amid the semiconductor frenzy. Meta and Micron also featured among the day’s notable gainers.
After the close, Applied Materials fell in extended trading despite issuing guidance above Wall Street forecasts — the season’s signature pattern, appearing yet again, and a notable one given the stock’s roughly 190% gain over the past year.
Bottom Line
Thursday extended the disinflation rally and pushed the S&P 500 to a record above 7,800 for the first time, and the front-end math is genuinely improving: September hike odds have fallen from 50% to 30% in three sessions on the back of a weak payroll print, an in-line CPI, and a soft PPI headline. SanDisk’s investor day added a credible new growth vector for memory, and Workday’s takeover interest reminded investors that software’s valuations had fallen far enough to attract buyers with checkbooks.
The cautions are in the details rather than the headlines. Core PPI accelerated, even if the cause was an accounting artifact tied to rising asset prices. The 30-year auction cleared at the highest yield since 2001 with below-average demand. Oil fell on an IEA demand downgrade rather than a supply glut. And Cisco demonstrated once more that in this market, three decades of record demand can still be a selling event if expectations got there first. The path to a September hold is open — but as Bill Adams noted, August’s data comes first, and it gets the last word.
Leave a Reply