The AI Industry Is Quietly Turning: Token Usage Falls, Server Rents Crash, the Semiconductor Cycle Is About to Change Face
US Stocks · Options · News · Views
The signal to watch isn't in stock prices. It's in the Token curve. Industry tracking shows that Token usage for AI models peaked in mid-May and then dropped noticeably. Frontier large models still have rigid demand, that's a fact. But the issue is that a lot of tasks that can run locally, on small models, are rapidly migrating off expensive closed-source APIs. 3B-parameter-class models now run on ordinary consumer laptops; many everyday scenarios simply don't need top-spec hardware.
Behind this, the capability gap between open-source and closed-source models is closing. Even Musk has admitted the convergence of the two is coming; "distillation, quantization, replication" will keep pushing costs down another step. Microsoft has already deployed and is evaluating self-hosted DeepSeek V4 in Azure AI Foundry, not an announcement, but tests already running. When the biggest payer starts using cheaper models to deliver services, the pricing pressure on Tokens will only intensify.
One-line summary: declining high-performance Token demand plus expansion of cheaper localized models plus slowing data-center demand growth. These three forces stack and ultimately feed back to compress the capex cycle. It's a reflexive negative loop. Once the industry sees that "investing a little less is fine," the demand side of semiconductors starts absorbing price-adjustment pressure.
Server and data-center rental prices are crashing, and that curve is more honest than any roadshow deck. It's not that there's no demand. It's that the marginal return on each new dollar of investment is no longer attractive. MU, the storage chain, HBM: for two years the story was "capacity is full, raising prices is justified." Now they face a different question: "Is investing one more dollar still worth it?"
The storage complex had two near-term catalysts today to lean on. Apple's Cook cited tight memory-chip supply and signaled price increases (the market is guessing the new iPhone could rise as much as 20%, with August/September as the verdict), and Trump said Apple has agreed to work with INTC to design and manufacture chips in the U.S. But these are sell-side sentiment catalysts, not evidence of a capex cycle resuming. What really decides the medium-term path of NVDA and SOXX is the four capex heavyweights: Amazon, Google, Microsoft, and Meta. Watch who first signals "slower capex growth" on the Q2 earnings call. Not "stop spending," slower growth. That pivot signal will most likely surface this Q2 earnings season.
All four majors rose today, with Nasdaq recouping yesterday's losses, looks pretty. But it's a textbook triple-witching (quarterly options expiry) settlement tape. Total options settlement was the largest in history; volume was structurally inflated, with a huge share of prints bunched into the opening stretch and the final minute of the close. This is not a normal market statement.
ISRG printed a high-volume stabilization pattern today, but in the context of triple-witching, that signal's reliability moves to the back of the line. A pattern with a long lower wick would be more actionable; today's combo plus distorted volume argues for patience, don't rush in. NFLX isn't even a stabilization pattern, just normal price action.
The call is direct: a triple-witching high-volume bounce has little directional meaning, don't treat it as a trend signal.
The next two weeks have little to do with fundamentals; they're mostly mechanical institutional repositioning and rebalancing flows, up today, down tomorrow, back the day after, weak directional signal. Hence "chop and chaos." The S&P is pinned in the 7400 to 7600 range, going nowhere.
The hard data on the bearish side: JPMorgan estimates the month-end/quarter-end rebalancing is enormous. Pensions need to sell roughly $165 billion of equities (one of the largest in history) because aggregate positioning has run up to about 110%; they must sell stocks and buy bonds to rebalance. That creates a clear oversupply of equities into month-end.
The bullish buffer: July 1 opens a new asset-allocation cycle. Target-date funds, passive vehicles, mutual funds, and systematic strategies all re-deploy for the second half, with flows moving between sectors. On one side, month-end selling; on the other, fresh month-start money. In between, it's a tug-of-war. For short-term traders, these two weeks of long/short jumping carries almost no trend signal; leaning on fundamentals here isn't very useful.
After today's settlement, two structural signals are worth remembering. First, dealer Gamma on the S&P has flipped negative, meaning the zone from current prices down to about 7350 is a "vacuum." Systematic capital has little footprint here; dealers have less ability to suppress volatility. Mapped to the tape: either it doesn't fall, and instead grinds slowly higher along 7500/7520/7550 (with heavy overhead resistance); or, if it really does fall, volatility through that band gets amplified, and it would take a drop toward 7350 before dealer systems step back in and dampen the move.
Second, today's net flow was negative, and it came primarily from incremental put buying, not just plain covering. After quarter-end rebalancing and rolling, traders are adding downside hedging and locking in profits. Upside? They don't care, at most they lose a little premium. Downside? Protected gains. Note that they did not add to calls; that's a new, defensive shift.
SpaceX (the private stock, not the S&P index) had two items today. One: it's preparing its first investment-grade bond issuance, roughly $20 billion, primarily to refinance the $20 billion bridge loan maturing in September 2027. The company's total debt is about $29.1 billion; rolling the big maturity first makes the rationale clear. Two: the stock held $180 and was bought into the close; turnover moved back near 50%.
Over the past three days, turnover was 93, 46, 57; only yesterday's one-way down day was lower (32%). After this round of chop, the publicly traded float has essentially changed hands once: some are trapped above; some have formed a support army below $150/$160; the chip structure has returned to relatively neutral. Neutral positioning means next week's path may look different from now, and the odds of an upside wiggle are increasing.
Anyone playing here must be clear-eyed: it's speculation. If you like the value, treat it as a value trade; if you're speculating, set your take-profit and stop-loss. The worst mistake is "I came in to speculate, got trapped, and then lied to myself that it's value." On valuation, my personal view is that it's rich; fair value should be under $100, and after the December lockup partial release there's a chance it revisits below $100 within a year. That's a subjective view, not a promise.
Disclaimer: This article is personal market observation and trading review only, not investment advice. Markets carry risk; trade with caution.