There is a slightly absurd feature buried inside July’s report that traders should probably understand before the number hits.
Macro Daily: The Last PCE Print Where Wall Street Can Inflate Inflation
There is a slightly absurd feature buried inside July’s PCE report that traders should probably understand before the number hits: a strong stock market can mechanically make measured inflation look hotter. Goldman expects to rise 0.23% in July, a touch above the 0.22% print, and the main reason is not rent, wages or anything households would immediately recognize as inflation. It is portfolio management fees.
Under the current methodology, portfolio management service prices borrow heavily from data, and those data are closely tied to equity-market performance because many asset managers charge fees as a percentage of assets under management. Stocks go up, asset values rise, fee revenues rise, and the statistical machinery can interpret part of that increase as inflation. Goldman estimates portfolio management services alone could contribute about 8 basis points to July core PCE, driven by an expected 4% rise in that category. The wrinkle is that July is the first month of the quarter, so the calculation reflects not only June equity performance but the strength of the market across Q2. In other words, April and May are still rattling around inside July’s inflation number.
That would normally be an interesting statistical curiosity. This time it matters more because July is the final month before the BEA changes the methodology. From September 30 onward, the initial estimate for portfolio management inflation will rely more heavily on wage growth in the sector rather than data linked to equity returns. Eventually, the numbers will be revised using Quarterly Services Survey revenue data, bringing the stock-market relationship back into the series, but with a much longer lag. The same July number could therefore initially print hot, be revised lower in September, and then revised higher again in December when the revenue data arrive.
For traders, this is a useful reminder not to treat every tenth in PCE as if it came from the same part of the economy. A 0.23% core print driven partly by booming equity portfolios is very different from a 0.23% print driven by persistent housing, wages or broad service inflation. The Fed knows that distinction as well. Officials have previously shown some skepticism toward the outsized influence market-sensitive portfolio fees can have on PCE and have at times leaned on market-based PCE measures that strip this category out.
That is why Goldman’s Fed conclusion is more important than the headline number. Their view is that most voting FOMC members would regard the July inflation mix as acceptable and would still want to see August and before deciding what to do in September. Goldman expects both core CPI and core PCE to settle around 0.2% in August, while methodological changes across portfolio management, legal services and software should mechanically shave around 0.2 percentage point from year-on-year core PCE inflation next month.
The market implication is fairly clean. July PCE may look a shade firmer than CPI, but the composition matters enormously, and this particular source of heat is about to be statistically rewritten anyway. Unless August inflation breaks materially higher, Goldman still sees the Fed staying on hold in September and ultimately through year-end.
So the old methodology goes out with a bang, but traders should be careful not to mistake the noise of Wall Street marking itself higher for a genuine reopening of the inflation fire.
Korea Daily: The Memory Trade Comes Roaring Back
Korea has gone from liquidation pit to momentum trade in barely two weeks. The jumped as much as 4.8% Thursday, taking the rebound from its July 30 low to roughly 22% and putting the index back into technical bull-market territory. and are again doing most of the heavy lifting, which tells you exactly what has changed: the market has rediscovered the AI memory bottleneck after spending July treating anything attached to semiconductors as a source of liquidity.
The speed of the reversal is important. July’s 22% collapse was Korea’s worst month since the global financial crisis, but a meaningful part of that move was not a sudden destruction of the memory cycle. It was leverage coming out of a market that had simply become too crowded. Leveraged single-stock products, margin financing and retail enthusiasm had created an unstable structure, and once the unwind started, price became the reason to sell rather than the consequence of selling. Government restrictions on leveraged ETFs and a reduction in margin debt have since helped clean up some of that excess, allowing fundamentals to regain control of the steering wheel.
And those fundamentals have not disappeared. Big Tech is still spending enormous amounts of money on AI infrastructure, while the physical supply response in high-end memory remains constrained. That is the important part of the Korean story. AI applications are broadening from model training into agents, inference and increasingly physical applications, all of which consume memory. You can debate how long hyperscalers will maintain today’s capex velocity, but you cannot instantly manufacture the advanced memory capacity required to feed it. Samsung and SK Hynix therefore remain sitting on one of the more valuable bottlenecks in the entire AI buildout.
There is also a second leg developing underneath the trade: shareholder returns. Expectations that Samsung and SK Hynix will announce additional capital-return measures are giving investors another reason to reconsider the sector after the washout. Lower valuations plus improving cash generation is a much more attractive cocktail than the one investors were drinking near the June highs.
But I would not mistake a 22% rebound for a clean return to the old regime. The Kospi remains roughly 24% below its late-June peak, and foreign investors have still pulled more than $100 billion from Korean equities this year. Some overseas money is beginning to creep back after the valuation reset, but this rally has yet to demonstrate that foreigners are prepared to chase Korea aggressively again.
That is the next marker I would watch.
Korea has already shown us what happens when retail leverage meets an irresistible AI narrative. The first rally became too crowded, and the July liquidation cleared the table violently. This second move looks healthier because leverage has been reduced and the underlying memory story remains intact. But for the Kospi to graduate from rebound to something more durable, it probably needs foreign capital to stop treating every rally as an exit ramp.
For now, the memory trade is alive again. The difference is that after July, everyone at the table has been reminded where the trapdoor is.
Stock Market Daily: Cerebras Shows Where the AI Trade Gets Harder
Cerebras () just delivered a useful reminder that the AI boom can still be very real while individual companies disappoint investors badly. Shares fell sharply after earnings, not because the business is collapsing, but because the market had already priced in a much steeper revenue ramp. Third-quarter sales guidance of about $215 million was slightly above the average estimate, yet that was not enough for investors who had spent the post-IPO period treating every AI infrastructure name as if demand would arrive in a perfectly straight line.
The more interesting part of the report is where the revenue is actually coming from. Cerebras was pitched to the market as a challenger to Nvidia () with a radically different chip architecture, but its cloud-computing business has now become the larger growth engine. Core cloud revenue nearly quadrupled to $127.7 million, while hardware sales were much more uneven. On a GAAP basis, hardware revenue fell 23% from a year earlier. That does not necessarily mean the chip proposition is failing. It tells us something more important about the current AI buildout: infrastructure demand is enormous, but converting that demand into hardware revenue is still dependent on customer timing, data-center readiness and physical deployment capacity.
That distinction matters because the market has become accustomed to rewarding AI exposure first and asking operational questions later. Cerebras is now running into the second stage of the trade, where investors want to see not only demand but conversion. A customer can want the compute, sign an agreement and still lack the physical data-center capacity to install the systems immediately. That creates exactly the sort of revenue lumpiness management is describing. It is also why a company can report 74% year-on-year sales growth, raise full-year guidance to $880 million to $890 million and still see its stock get hit.
This is where the read-through for the broader AI complex gets more interesting. AMD () and Intel () have recently posted much stronger headline growth, while Nvidia remains the benchmark against which every aspiring AI hardware company is measured. Cerebras therefore faces a difficult middle ground. It has differentiated technology, major customers and strong cloud growth, but it is still trying to prove that its architecture can move from niche adoption into a broader commercial ecosystem. That is a much tougher hurdle than simply showing that AI spending is booming.
The market reaction also says something about expectations. Cerebras shares had already risen more than 40% since the May IPO. Once a stock gets rerated that quickly, “better than expected” is no longer enough if the whisper number sits higher. That is the problem with late-stage enthusiasm in any hot thematic trade: the earnings bar rises faster than the reported numbers.
The AI capex cycle still looks strong. Cerebras itself is guiding to more than triple revenue in 2027, and cloud demand remains exceptionally healthy. But this report is another warning that the market is beginning to separate the winners from the companies that merely sit near the theme.
That is healthy for the sector, even if it is painful for the stock.
The first phase of the AI trade was about buying exposure.
The next phase is about proving execution.