The AI Trade Splits Three Ways as Money Rotates Out of Tech Into a 4.7% Ten-Year

Today’s tape doesn’t say “AI top”—it says the AI bull case is fragmenting. Demand is demonstrably real, but the value is migrating from the frontier labs toward infrastructure, memory, and data owners, and that migration is unfolding just as the 10-year spikes to 4.71% and capital rotates out of QQQ into energy, health, and financials. This is single-name rotation, not macro risk-off — for now.

The AI argument has stopped being one debate and become three

The old fight was demand: real or a bubble? On that, the bulls landed the day’s cleanest punch. Wayne Liang points to the

$500B+

NVDA ... – SK Group infrastructure partnership—factory buildout plus next-gen memory co-development — and dares the bears to explain half a trillion in committed capital chasing ‘demand that supposedly isn’t real.’ The tape backs the price side of his case: NVDA closed at $206.84, above its 20-day, with a fresh MACD buy signal (histogram +0.76) and still green MTD/YTD despite the selling. Fundstrat frames hyperscaler capex as rational return-on-capital allocation and expects the broad market to make new July highs; the All-In panel calls Google’s spend a buy signal, citing a 32% historical ROIC and naming Alphabet the best public AI stock to own. Luke Gromen is the loudest voice on the other side, and his objection is structural, not directional: this buildout leans on ~$1T of repayable debt rather than the self-funding FCF of the dot-com era; tech is ~90% of GDP growth; and the US is running a 6% deficit ‘in the midst of a bubble’ — so a shock now hits a far more levered, concentrated system than in 2000. Wayne Liang explicitly rejects the Burry 2008 subprime analogy; Gromen’s point is subtler and harder to wave away. But the genuinely new thread is neither demand nor leverage—it’s where the value accrues. The All-In panel argues foundation models are commoditizing faster than anyone expected: Chamath says the business model is no longer in the model but in the application layer above and the infrastructure below. Kimi K3 is cited at rough parity with Opus/GPT-5-class models; Cursor and Thinking Machines both built competitive products by post-training on Chinese open-source weights, and the panel’s bet is that owners of proprietary training data capture the rents. That reads bullish for Google/cloud and bearish for the closed-lab premium—Anthropic gets accused of regulatory-capture lobbying while its revenue growth stalls and it eats a $1.5B settlement for training Claude on 7M pirated books.

Memory is the pinch point—and where our own call is hurting

The cleanest place to watch the demand-vs-margin tension resolve is memory. Fundstrat concedes the near-term picture is ‘very messy’: rising memory prices are squeezing AI-company margins even as the analyst stays bullish long-term. Daniel Newman argues Micron’s earnings look more durable than a pure datacenter play because of its industrial and automotive exposure to physical AI, and Wayne Liang expects Burry’s MU short to age as badly as the rest of his list. The options market disagrees about the certainty: MU’s ATM implied vol is a staggering 102%, with a 1.79 put/call — this is not a name the tape thinks is settled. Straight talk on our own record: on July 23 we made MU a high-conviction dip-buy and NVDA a medium-conviction buy, and on July 22 NVDA was our high-conviction pick. Those outcomes aren’t scored yet, but they’re aging unevenly. NVDA has held — above the 20-day MACD buy signal, still green. MU has not; it closed $920.95, below both its 20-day ($975.76) and 50-day ($955.28), RSI 44, 10-day momentum −6%, MACD deeply negative. The ‘buy the laggard on the DRAM-shortage narrative’ framing is underwater short-term, and the risk we flagged then—that rising long-end rates and someone in the knowledge base trimming MU could invalidate it — is exactly what’s biting. The memory bull case may still be right on a 6-month horizon; it is wrong this week.

Rotation under the index, with rates doing the pushing

The index-level story is a genuine rotation, and the data confirms Mark Newton’s read. QQQ closed at $684.23, below both June and July lows (Sun Liao and Newton both flag it), RSI 33, MACD firmly negative, momentum −5.7%—with a punishing 2.72 put/call and a 10.2 put-call skew. Underneath, the money went exactly where Newton said: Energy +11.3% over the past month, Health Care +6.0%, and Financials +4.8%—the leaders—while XLK is −3.9% (−4.7% vs SPY) and Consumer Discretionary −4.9%. Financials are near a 52-week high; Newton’s calls on Healthcare, Utilities, Industrials, DJT transports and Aero/Defense all line up with the sector tape. The engine is rated. The 10-year jumped to 4.71% (+30bp), the 30-year to 5.17%, and the 2-year to 4.37% — a real 10-year yield of 2.43%. Eliant Capital’s throwaway line—’the 10Y is back to 4.7%, we can’t attack this weekend’ — is literally the number, and it is the single best explanation for why long-duration tech is being sold while cash-flow-rich financials and hard-asset energy lead. Crude is up 27% on the month, which both feeds the energy leadership and revives Wick’s point that the market has quietly priced out Strait-of-Hormuz supply risk. Is this healthy or the start of something? The dispersion read sides with Fundstrat’s ‘consolidation, not top’: VIX 18.6, index realized vol just 10% against 36% single name, correlation 0.24—a rotation regime, not macro risk-off. But Cem Karsan warns of a volatility shock around month-end and notes momentum names like Circle have support only into the August 19 expiration, and PharmD_KS flags sentiment as ‘pretty horrible’—the kind of washout that precedes a bounce. The bull and bear here aren’t describing different data; they’re disagreeing on whether a 4.7% 10-year lets a tech-led index reclaim new highs.

Crypto: Wick’s bear versus the clarity bulls

The crypto claims split just as sharply. Wick is running a full-throated bear case: he pegs the current downdraft as a bear that’s already carved BTC to a

$60-70k range, points to Saylor ‘failing to hold a simple $5k range at support to peg the NAV’ as evidence MicroStrategy’s mechanism is breaking, and spends most of his posts demolishing DCA—arguing it increases timing risk, that buyers enter at euphoria and ride 60-80% drawdowns, and that averaging down is a rule to never break. He and Wick’s own ‘super cycle is a stupid name… often signals a crypto market top’ land as a coherent late-cycle warning. Against that, Tom Lee/Fundstrat push the regulatory-clarity catalyst: $13T. Charles Schwab publicly urging the Senate to pass the Crypto Clarity Act, and a senator warning that failure cedes ‘jobs, investment, economic competitiveness.’ That’s a structural tailwind, not a price call — and it doesn’t refute Wick’s point about where we sit in the cycle. Lyn Alden stays deliberately above the fray, noting she likes that Bitcoin is hard to change. The two sides aren’t really arguing the same question: Lee is talking about the multi-year institutional on-ramp, and Wick is talking about the next drawdown. Both can be right in sequence.