If they cannot, and the cheque receivers keep surging while the cheque writers remain pinned down, investors will increasingly stop calling it rotation. They will start treating the divergence as a warning that the AI buildout may be generating plenty of revenue upstream while the return on the spend becomes harder to find downstream.
Takeaways
- The week’s most important signal was not that stocks fell on a weekly basis. It was that they fell despite lower oil, easing Treasury yields, contained inflation and a blockbuster memory-chip earnings report.
- The S&P is being dragged around by the sheer weight of mega-cap technology, even as the equal-weight market and eight of eleven sectors continue to push higher.
- Investors are beginning to distinguish between the companies collecting the AI scarcity premium and those financing the infrastructure race.
- The shorter-term risk is more mechanical than existential. rebalancing, month-end books being squared, and quarter-end reallocation flows
The AI Invoice Starts Changing Hands
US stocks drifted higher into Friday, yet the was still heading for a weekly loss of more than 1.5%. On its face, that should not have happened.
fell roughly 10% as the Strait of Hormuz reopened and markets began looking beyond the immediate energy shock. May core PCE came in broadly as expected. eased by more than 10 basis points toward 4.37%, normally a welcome tailwind for equity valuations. delivered the kind of quarter that should have reinforced every bullish AI narrative already sitting inside the market.
And still, the spent much of the week pressing against 20 while the Mag 7 fell between 3% and 8%.
That is the question hanging over the weekend. Was this simply a dip in the names that had done the most work, or was the market beginning to question the one big trade that has carried so much of the index?
The reassuring answer is that the broader market has not rolled over. The continues to outperform the cap-weighted index this year and reached fresh highs, while eight of the eleven major sectors traded higher even as the headline index sagged. The problem is simple arithmetic. When more than 30% of an index falls more than 5%, the other 490 stocks can be having a perfectly respectable week and still fail to keep the benchmark afloat.
This is the flip side of one big market. Mega-cap technology became so large, so successful and so universally owned that it stopped being just another sector. It became the index’s centre of gravity. When those shares wobble, everything around them looks more unstable than it really is.
But there is a less comfortable explanation beneath the surface. The market is beginning to separate the AI ecosystem into those collecting the scarcity premium and those paying for it.
Micron’s strength is telling. Memory, HBM, networking and selected infrastructure suppliers sit on the right side of a bottleneck that everyone else is racing to secure. Scarcity is transferring pricing power toward the owners of the parts, and their earnings are beginning to reflect it. They are the cheque receivers: companies able to monetise every new capacity commitment without having to prove the final use case themselves.
The hyperscalers, device makers and enterprise users are the cheque writers. They are funding the compute, memory, power, cooling, networking and data-centre buildout while investors become more demanding about when the payoff arrives. Their strategic logic is clear enough. No board wants to discover it economized on GPUs or power contracts just as AI became the next operating system for corporate America.
But the market is no longer rewarding the spend merely because it is large.
raising prices across parts of its hardware range and facing similar pressures in gaming are small but revealing details. They show that AI costs are no longer sitting quietly in a capex line waiting to be amortized. They are beginning to move through the value chain, pressuring margins, product pricing, and eventually the consumer.
That is why this week’s divergence matters. AI demand remains strong. The issue is whether the companies writing the biggest cheques will capture enough of the eventual economics to justify the bill.
The suppliers are selling the tools required to keep the race moving. The hyperscalers are trying to build the platforms that eventually turn those tools into durable revenue streams. In the first phase, the suppliers tend to look cleaner because their pricing power is immediate. The platform owners carry the more difficult task of proving that the infrastructure spend becomes revenue, margins and free cash flow rather than merely a more expensive ticket to stay competitive.
That is where the reflexive risk sits. If suppliers continue to outperform while the hyperscalers lag, investors will begin treating the gap as a judgment on the economics of the spend. Underperformance brings scrutiny, scrutiny raises pressure for capital discipline, and every new data-centre announcement starts being measured against a harder question: when does this convert into shareholder return?
No one needs to abandon AI for that loop to matter. It only takes one major player to slow a project, stretch a timetable or place more emphasis on spending discipline before investors begin asking who else may follow.
The open-weight model debate adds another longer-dated wrinkle. If enterprises can run simpler inference tasks locally or through lower-cost models, premium cloud capacity may increasingly be reserved for the hardest workloads. That could broaden demand for hardware, servers and storage, but it would make the hyperscalers’ economic prize less automatic.
For now, the shorter-term risk is more mechanical than existential. Russell rebalancing, month-end books being squared, and quarter-end allocation flows are arriving just as investors are trying to decide whether this is healthy broadening or something more unsettling. That can make the next few sessions look far messier than the underlying fundamentals deserve, particularly when the largest names in the index are also the most widely owned.
But the cleanest market signal remains the cloud and hyperscaler complex. Those shares are now the pressure gauge for the broader AI trade because they sit at the point where ambition meets the bill. If they can regain their footing while memory, HBM and infrastructure suppliers continue to perform, this week will likely prove to have been a necessary release valve after an extraordinary run.
If they cannot, and the cheque receivers keep surging while the cheque writers remain pinned down, investors will increasingly stop calling it rotation. They will start treating the divergence as a warning that the AI buildout may be generating plenty of revenue upstream while the return on the spend becomes harder to find downstream.
The China Washout: Cheap Is Not Yet the Same as Cleared
Chinese technology has become the market’s abandoned room. The lights are still on, the furniture is still there, and the businesses themselves have not suddenly vanished into the night. But foreign liquidity is leaving, volatility is rising and the largest platform names are being sold as though every portfolio manager has found the same emergency exit at the same time.
That is the real shape of the China washout.
Around $11.1bn has come out of EM equities, while foreign investors have been heavy sellers of Asian technology into month-end and quarter-end. In that sort of market, and do not escape because they are large. They become the easiest source of cash because they are large enough to sell without asking too many questions.
When the room fills with smoke, investors do not always leave through the best exit. They leave through the biggest one.
The result has been an ugly combination of weaker sentiment, rising implied volatility and a price move that is increasingly feeding on its own momentum. Goldman sales-desk colour points to CTA deleveraging, ETF rebalancing and index-driven selling doing more of the short-term damage than any dramatic deterioration in the underlying business outlook.
That matters because once systematic flows take over, price starts manufacturing its own bad news. A lower market triggers de-risking. De-risking creates more selling. Volatility rises. Options markets begin charging more for downside protection, and suddenly the same trade that was already weak becomes even harder for institutions to hold.
China is not trading like Korea did during the recent leverage-driven snapback and collapse. Korea had that strange spot-up, volatility-up profile of a market being shaken by forced positioning and retail leverage. China is showing the more traditional risk-off pattern: equities falling, implied volatility rising and investors paying up for protection as though the floor is still somewhere below them.
That is why the washout feels so relentless. It is a liquidity event wearing the clothes of a fundamental story.
Yet this is also where the valuation paradox begins to matter.
Chinese internet stocks are trading around 14x forward earnings, against roughly 25x for U.S. internet peers. That gap is not irrational. It reflects geopolitical risk, regulatory scars, softer domestic demand and a market that has repeatedly punished investors for trying to buy every apparent bargain too early.
But there comes a point when the discount stops being merely caution and starts becoming a kind of institutional muscle memory.
Many of these companies still generate substantial cash flow. They still carry strong balance sheets. They still own dominant platforms, large ecosystems and businesses that would command a radically different valuation if they were listed anywhere else. The market is pricing the house as though the roof may leak forever, even while the foundations remain intact.
Alibaba is the cleanest example of that tension.
Its core commerce business remains under pressure as platforms spend heavily on merchant incentives, rebates and competition. Cloud growth is improving, but the domestic pricing war means monetisation is harder won than headline revenue acceleration suggests. At the same time, management is trying to fund AI infrastructure, maintain cloud relevance and still return capital through buybacks and dividends.
It is an awkward place to sit. Alibaba has to spend like a growth company while being valued like a mature company with no room for disappointment.
The constructive case is straightforward enough. If cloud growth keeps improving and earnings begin to recover, the current valuation may already be discounting far too much pessimism. But the market is not prepared to pay for that optionality yet. It wants proof. It wants margins to stabilise. It wants domestic demand to stop feeling like a soft floor. And above all, it wants evidence that the heavy AI spend is not simply becoming another cost centre in an already crowded domestic technology market.
That is why the technical picture is now starting to attract attention.
Alibaba’s RSI near 16 is not a buy signal. It is not a magic floor, and it does not mean the stock cannot get cheaper before it gets better. Oversold markets can stay oversold when foreign money is still leaving and institutional accounts are still hiding behind the door.
But an RSI that stretched is evidence of something else: exhaustion.
At some point, the market runs out of fresh bearish superlatives. The story does not necessarily improve, but the price has already absorbed so much bad news that it becomes harder to find the next marginal seller. That is when the contrarian starts looking up from the screen, not because the market is safe, but because the asymmetry is beginning to change.
For now, the tactical dashboard remains simple. Watch foreign flows. Watch whether volatility can stop rising even when spot is weak. Watch whether Alibaba can absorb a bad headline without immediately printing another low. And watch for any sign that , domestic consumption or platform-policy support gives institutions a reason to step back into the room.
China tech is now cheap enough to deserve attention, but not yet clean enough to deserve blind conviction.
The washout may be nearing the point where the sellers are exhausted. The market still needs to show that someone with real money is willing to take the other side.
Running Update
It has been a remarkable injury recovery. Funny what happens when you listen to the medical experts and stretch a little more. Who would have thought that consistent stretching might help a hip flexor? Sarcasm aside, the difference has been enormous.
Running pain-free again is a big win, although the return to peak fitness has been slower than the injury recovery itself. That is probably the part runners never want to hear: the body may stop complaining before the engine is ready to run at its old level.
Still, with 17 weeks to the Luang Prabang Half Marathon( October 24) , the path is clear enough. Keep the weekly mileage building as planned, keep the pace progression intact, and I should finish comfortably within my normal half-marathon range.
The bigger question is whether I can take 21 minutes off my last time.
Unlikely? Probably.
But never say never.