Summary

Friday’s payrolls report showed a 162K gain with the prior month revised higher, but seasonality dominated the number: it proves neither an overheated economy nor, even less so, inflation. Most of the increase came from leisure and hospitality—August coincided with the World Cup, with bar employment up roughly 59K—and local-government education during the back-to-school season; hourly earnings growth at 3.1% beat expectations, but has shown little meaningful pickup since 2026 began. Inflation is currently being driven more by input costs such as tariffs and oil.
尚浩’s core conclusion: even a rate hike need not trigger a blow-up in risk assets. This round of input-driven inflation cannot be directly solved by rate hikes; markets are roughly pricing 2–2.5 hikes in 2027–2028, while 尚浩 goes further and argues that 3 hikes are already priced. The question is whether hikes will continue as they did in 2022, and whether they will be disorderly; neither is visible at present.
The key risk variable is rate volatility, not the level of rates. 6m10y swap vol surged during the rapid-hiking cycle, but today rates are moving steadily higher without vol rising, and the pace of hikes is nowhere near enough to flatten the yield curve. Rates will reset valuations and multiples, but the real drawdowns come from interest-rate vol or a loss of control at the long end that triggers a liquidity shock.
尚浩 believes AI is the first genuine productivity gain in decades, meaning the global growth model reliant on money printing will have to reverse course. If the Fed stops controlling the long end, long-term rates should gradually approach nominal GDP, potentially reaching 6%–6.5%; if demand for the long end weakens, they could instead fall to 4.8%–4.7%, as Goldman Sachs forecasts. The upside side of the K-shaped economy cares little whether rates are 5%, 6%, 7% or even 8%; the Fed wants to hand the long end over to the market.
The essence of the multiple reset is that the market now needs higher growth. Growth of 15% may previously have supported a 20x P/E, whereas 20% growth may now be required for the same multiple; companies without sufficient growth should therefore be avoided. 尚浩 sees Nvidia’s roughly 70% growth and roughly 70% gross margin at just 12x P/E as irrational, and believes a $10T market cap is ultimately only a matter of time.
Compared with the US, 尚浩 believes the bigger risk may lie in Japan. Japan’s debt burden is high, and rates have risen from zero to 3%; if they reach 3%–3.5%, falling government-bond prices would generate losses for financial institutions. High inflation expectations also make Japan less willing to print money. US banks have already absorbed much of their duration risk and large Treasury losses during the previous cycle.
Positioning and vol suggest that the deleveraging of AI momentum trades is nearing completion. Prime Book continues to buy TMT momentum, and negative news around Anthropic’s 65B ARR caused little damage; implied vol for the AI basket has fallen to 65, leaving vol-control strategies close to fully invested. The 65B figure is below the prior linear extrapolation of more than 70B; 尚浩 believes this may reflect the exclusion of tokens distilled by Chinese users and the standardization of accounting ahead of an IPO, rather than a deterioration in July demand. Unlike in 2000, today’s hyperscalers are trillion-dollar companies, and their CapEx is driven primarily by demand from their own enterprise customers, not simply by Anthropic or OpenAI.

Deep dive

1. Payrolls beat at 162K, but seasonality dominated and does not prove overheating

尚浩’s breakdown: the 162K payroll gain, alongside an upward revision to the prior month, would historically be considered a very strong report, but the increase was concentrated in leisure and hospitality and local-government education. August coincided with the World Cup, lifting bar employment by roughly 59K; the back-to-school season also boosted local-government education jobs. “Although the number is positive, more than half of the jobs themselves carry seasonal factors,” so “there is no way to prove the economy is overheating.”
Inflation consists of three components: services, rents and input-driven goods. 尚浩 believes the third bucket is now dominant—follow-on effects from tariffs, oil and other inputs. Hourly earnings growth at 3.1% beat expectations, but has shown little meaningful pickup since 2026 began; consumer credit is recovering, but remains well below 2021–2022 levels. The employment report therefore says more about resilient fundamentals than persistent inflation.
Manufacturing added 16K jobs, including roughly 6K in metals and machinery and roughly 8K in nonresidential activity; construction added 22K. 尚浩 links those gains to data-center construction, noting that the regions with stronger economies in the Beige Book are also largely those building data centers. Finance and infotech contracted, reflecting AI’s replacement of some labor in those sectors.

2. Even a real hike is no big deal: if it’s neither persistent nor disorderly, it’s not a crash

Waller’s message on Thursday was to wait for Friday’s CPI before making a decision; Warsh was relatively more hawkish, arguing that inflation was not yet fully under control. 尚浩 therefore puts the odds of a hike at roughly 50/50, if anything leaning toward a hike.
But wage growth has not repeated the roughly 7% surge of 2021–2022, and consumer credit has not expanded on a large scale. Input-driven inflation cannot be directly solved with rate hikes: the impact of oil prices or tariffs does not disappear simply because demand is suppressed. As 尚浩 put it: “What does oil falling to $200 have to do with your hiking rates? Nothing.”
Markets are roughly pricing 2–2.5 hikes in 2027–2028; 尚浩 also believes 3 hikes are already priced, along with Warsh’s more hawkish stance and the improvement in economic data. His expectation for Friday’s CPI is that it “won’t be too bad”—not particularly high, given that wages and consumer spending have not accelerated meaningfully.
He distinguishes between persistent and disorderly hikes. Disorderly hikes would mean inflation is very high and the central bank is forced to raise rates rapidly, flattening or even inverting the yield curve. 6m10y swap vol surged during the rapid-hiking cycle of 2021–2022; rates are now moving steadily higher without vol rising, and there is no sign of a pace of hikes sufficient to flatten the curve.
The level of rates will reset asset valuations and multiples, but multiples have already adjusted over the past several weeks. The real catalyst for a major selloff would be interest-rate vol or a loss of control at the long end, triggering a liquidity shock and deleveraging. Higher rates therefore cannot be equated directly with a market crash; 尚浩 also points out that capital markets performed reasonably well after rates stabilized from 2023 onward.

3. AI is the first real productivity gain in decades; the long end must reprice

尚浩’s framework is that, after 2008, the West lacked both population growth and productivity growth but still needed to sustain roughly 2% growth, so it relied on money printing and liquidity. Japan’s YCC and the US’s QE were both, at core, ways of keeping rates below nominal GDP and using the spread to sustain growth. He describes the process as “like an addiction: the more you use, the more you need.”
AI has delivered the first genuine productivity gain in decades. If the Fed stops controlling the long end, long-term rates should gradually approach nominal GDP, potentially reaching 6%–6.5%; under a K-shaped economy, however, 尚浩 is unsure whether rates will remain at current levels or fall back to 4.8%–4.7% as Goldman Sachs forecasts, driven by weaker demand for the long end.
The upside side of the K-shaped economy is highly productive and cares little whether rates are 5%, 6%, 7% or even 8%; rising rates primarily hurt the downside side. 贝森特 is more likely to manage duration to prevent excessive volatility at the long end than to determine where long-term rates ultimately settle. The Treasury will also issue 10-year notes this week.
尚浩 believes the Fed is handing the long end over to the market. Once it does, long-term rates will ultimately track GDP growth rather than simply the number of short-end hikes. As long as rates rise steadily with growth rather than running out of control, the market need not crash.

4. Multiple reset: the market needs higher growth, and Nvidia at 12x P/E makes no sense

A multiple reset means that the same valuation multiple now requires higher growth: 15% growth may previously have supported a 20x P/E, while 20% growth may now be required for the same 20x P/E. Companies without sufficient growth will therefore face multiple compression.
尚浩 notes that companies growing 90% are not unusual in semiconductors. Nvidia is growing at roughly 70% with a roughly 70% gross margin, yet trades at only about 12x P/E; he does not believe that valuation can remain depressed indefinitely and thinks a $10T market cap is ultimately only a matter of time.
His investment conclusion is to avoid companies without enough growth and look for assets that can match the new growth and technology cycle. The US’s roughly 2.5% real growth in the past partly relied on money printing; AI could lift the economy itself to 2.5%, 3% or even 4%, meaning investments must at least outrun the new growth rate.

5. The real risk may be Japan, not the US

AI’s means of production and semiconductors benefit not only the US, but also China, Japan and South Korea. Europe is relatively weaker, but 尚浩 still believes it could enter a rate-hiking cycle. Compared with the US, he sees the greater risk potentially lying in Japan.
Japan’s debt burden is high, and its 10-year yield has risen from near zero to roughly 3%. If inflation expectations continue to push rates to 3%–3.5%, falling government-bond prices could inflict major losses on financial institutions. Japanese inflation expectations are high, as reflected in inflation swaps, making Japan less willing to solve the problem by printing money.
US megabanks lived through the 2008 crisis and have continued to move duration risk off their balance sheets. As yields rose from just above 1% to 4.8%, large Treasury losses were already recognized; 尚浩 believes the resulting hole has gradually been repaired from 2024 to the present. Regional banks including SVB also went through a loss cycle in 2023.
Japan’s debt is mostly domestic, so in theory it can also print money, but doing so is harder in a cycle where growth and inflation are rising together. If Japan is unwilling to print, government-bond yields rise and prices fall, potentially leaving financial institutions with large volumes of bad debt. That is why 尚浩 believes Japan may deserve more attention than the US when assessing rate risk.

6. Positioning, vol and Anthropic’s 65B: momentum is proving hard to knock down

Prime Book continues to buy TMT momentum, and the large-scale deleveraging ended in July. Aggregate gross and net margin remain relatively low, net position is close to its level at the start of the year, and total leverage may be approaching March levels.
After the negative news around Anthropic’s 65B ARR, AI momentum stocks were not hit particularly hard. 尚浩 believes this shows that the remaining downside force is limited, similar to the point in August when Korean leverage could no longer delever meaningfully; it does not, however, allow for a precise call on the price bottom.
Rate vol has not risen, equity vol is falling, and single-stock vol is falling faster. The implied vol of an AI basket comprising roughly 20–30 AI-related stocks has fallen to 65, while volatility across the constituent companies has roughly halved; under a normal vol-control strategy, positioning is already close to fully invested.
Anthropic’s 65B is below the more-than-70B figure implied by the earlier linear extrapolation of AI Labs data. 尚浩 offers two possible explanations: first, tokens generated by Chinese users distilling models were excluded because they should not count as genuine demand; second, accounting was standardized for an IPO, with sales revenue shares deducted from Amazon-channel sales because they should be booked as costs. The figure therefore does not mean that July demand actually deteriorated.
He also expects Anthropic’s token growth to slow inevitably because pricing is too high and gross margin could reach 90%. The more encouraging point is that tokens sold by Microsoft are not cheaper than OpenAI’s, so this is not simply a price war; by integrating customers through its own platform, Microsoft may even be able to sell at a higher price.

7. This is not another 2000

尚浩 believes the market may repeat the rate-hiking process, but the outcome will not be identical. In 2000, there was almost no capital or demand outside the system, and Amazon could sell only a limited range of products such as books; today, there is more spendable capital and demand extends beyond coding and research within the AI ecosystem.
After reviewing data from Microsoft, Google and Amazon, the hyperscalers, he is more confident. These companies lived through 2000 and are now trillion-dollar or even multi-trillion-dollar companies, with relatively responsible guidance.
Their CapEx is not determined simply by demand from Anthropic or OpenAI, but by the enterprise customers they control and their own judgment about AI demand. 尚浩 therefore does not see the current situation as 2000, when expectations at small technology companies ran out of control, companies trampled one another and the entire market collapsed.

$TBBB: Tiendas 3B is Mexico's Aldi. Is it too late to buy? | Fruit Tree Capital