Published Sunday, August 30, 2026 · Prices are Friday, August 28 closing prices (markets closed all weekend) · @dailyanalysts
Remaining 2026 is a collision: the strongest AI fundamentals of the cycle (NVDA's 70% FY28 guide, ASML effectively sold out for 2027, Marvell's $120B Google deal) versus the first Fed hiking discussion in years (September hike odds ~56%, 10Y at 4.72%, debt past $40T) and a $1–2T AI IPO supply wave (OpenAI, Anthropic). Base case (50%): choppy S&P 7,500–8,000, not a straight line to Goldman's 8,000. Own the monopoly layer on macro-fear dips: MRVL (entry zone $210–225, target $270), ASML (adds $1,550–1,650), GOOGL ($330–347). The one level that changes everything: 10Y above 4.80% compresses AI multiples roughly 10–15%.
Friday's tape was the whole remaining-year thesis compressed into one session. Nvidia closed -4.6% at $217.55 — two days after reporting revenue of $96.2B, +106% YoY (Q2 FY27 press release), and guiding fiscal-2028 growth to ~70% versus the ~44% consensus. Marvell closed -10.3% at $216.62 two days after beating and raising its FY28 revenue target to ~$18B. SMH fell -3.5%, ASML -2.2%, TSM -2.3%, AMD -2.3%. The catalyst was not AI demand — it was Fed Chair Kevin Warsh's first Jackson Hole speech Friday morning, in which he said this summer's better inflation readings "do not tell me that underlying trends have meaningfully improved," that financial conditions are not currently restrictive, and that rates remain the Fed's "predominant tool." The 2-year yield jumped ~11bp to 4.35%, the 10-year closed at 4.72% (30-year: 5.21%, both at 2007-era highs), and CME FedWatch moved September hike odds from ~35% to ~56%, with a hike by December priced above 70%.
Layer on top: a Politico report Thursday (eight sources) that the White House is weighing a sweeping new round of Section 232 semiconductor tariffs covering chips and end products — laptops, data-center servers, gaming hardware — on a staggered rollout. And in the background, the largest equity-supply pipeline in market history: OpenAI (confidential S-1 filed, eyeing a September debut at $1T+), Anthropic (confidential S-1 filed, test-the-water meetings underway, investors projecting a float near $2T), and SpaceX already public at ~42x forward sales with an unlock calendar ahead.
Claim: The AI earnings engine is real and accelerating — but for the next four months the multiple, not earnings, decides returns, and the multiple is now hostage to the long end of the Treasury market and the IPO supply pipeline. Wall Street's year-end targets (Goldman 8,000, JPM 7,800, Morgan Stanley 8,000) embed a flat ~21x forward multiple. That only holds if the 10Y stays under ~4.80% and the Fed is one-and-done at most. With the Fed openly teeing up hikes into 3.3% core PCE, a credibility-impaired Treasury intervening in its own market, and $1–2T of new AI equity looking for buyers, the probability-weighted path is a violent 7,500–8,000 range with rotation inside it — not a straight line. Returns from here are earned by owning the toll-booth layer of AI (bought on macro-fear dips) and by respecting regime levels, not by chasing AI-momentum rips. The behavior of the tape proves the regime: on Nvidia's best day since April (Thursday, +8.7%), roughly 70% of the S&P 500 declined (Susquehanna via CNBC), and Nvidia still closed the week below its May 14 all-time high of $236.54. When the best earnings of the cycle can't break out the best stock, the constraint is macro.
| Call | Entry zone | Target | Invalidation (one measurable condition) | Horizon | Conviction | Audience |
|---|---|---|---|---|---|---|
| LONG MRVL (restated — Friday's close $216.62 is inside the zone) | $210–225 | $270 | Weekly close < $195, or any disclosed cut to Google TPU volumes/timeline | 1–3 months | SPECULATIVE (one signal; size small) | All, small size |
| LONG ASML (staged adds; Friday $1,696) | $1,550–1,650 | $1,950–2,000 | Weekly close < $1,480, or management cuts 2027 Low-NA EUV order coverage | 6–12 months | HIGH (2027 orders ~fully covered + monopoly) | Core holding |
| LONG GOOGL (Friday $346.59) | $330–347 | $400 | Capex guidance cut at Q3 print, or weekly close < $300 | 6–12 months | HIGH (2+ signals: TPU external revenue + Anthropic stake NAV) | Core holding |
| NVDA: hold, don't chase (Friday $217.55; ATH $236.54) | Add $195–205; momentum re-entry only on a close > $236.54 | $280 | FY28 growth guide cut, or weekly close < $190 | 6–12 months | WATCH at current price | Core holding |
| AVGO: no pre-earnings entry (reports Wed Sept 2 AMC; Friday $368.79) | Post-print dip $330–345, only if AI revenue guide confirms | $420 by early 2027 | AI revenue guide implies < ~$16B/qtr run-rate, or confirmed TPU share loss | 3–6 months | WATCH (binary event Wednesday) | Tactical |
| BTC: buy the macro flush (spot $78.2K) | $72–74K | $92–95K | Weekly close < $68K | 1–3 months | SPECULATIVE | Crypto allocation |
| AVOID / trim IWM (restated open call) | $293–297 short/trim | $282 | Weekly close > $302 | Through the Sept 15–16 FOMC | HIGH | Tactical |
| AVOID long duration (TLT $82.88, 30Y 5.21%) | No long-dated bonds until 10Y ≥ 4.85–5.00% | Then tactical TLT long becomes interesting | — | Regime trade | WATCH | Macro |
Conflicts disclosed: (1) The IWM $282 target runs through the open $290 regime-warning level on the floating-rate-leverage complex — if IWM works, that risk condition triggers on the way; coherent as a pair, but readers follow both. (2) MRVL/ASML/GOOGL longs sit against my own cautious macro view deliberately: MRVL is small and its $195 invalidation fires before portfolio-level damage; ASML/GOOGL are 6–12-month core positions to be built on dips, sized for 15% drawdowns. (3) The prior CRM/CRWD/OKTA trim call resolved correctly on the 8/28 hawkish event; re-entry zones (CRM $215–225, CRWD $200–210) remain live but a hiking Fed argues for patience.
Nvidia. The quarter settled the demand debate: $96.2B revenue (+106% YoY, per the Q2 FY27 press release) with a $108B Q3 guide, a 70% FY28 growth guide versus ~45–50% expected, Amazon committing to 2M additional GPUs in 2027–28, and CFO Colette Kress saying current-quarter growth is driven mainly by non-hyperscaler customers (neoclouds, enterprises) — demand broadening, not narrowing. Jensen Huang's claim that customer "return on investment capital is now less than a year" is the single most important sentence for the whole complex: if true, capex is self-justifying; if it breaks, everything below breaks with it. The bear case (circular financing, receivables, margin drift) didn't disappear — it was out-voted. My read: hold, add on a flush to $195–205, and only chase above a $236.54 close. CNBC's Investing Club raised its target to $280; mine is the same, on ~28x a FY28 EPS trajectory consistent with the guide.
Marvell — the cleanest derivative of the custom-silicon shift. On August 19 Marvell disclosed a landmark custom-silicon agreement tied to Google's TPU ecosystem: up to $120B of potential revenue through fiscal 2033, with Google granted a $12.18B warrant (~59M shares) — Google is literally paid in Marvell equity to buy Marvell silicon, the tightest customer-vendor alignment structure this cycle has produced. Fiscal Q2 then beat (~$2.7B revenue, +37–38%) and the FY28 revenue target was raised to ~$18B — and the stock fell 10.3% anyway, in a post-NVDA expectations reset plus the tariff report. Sell-side targets now cluster at $300 (Craig-Hallum, Rosenblatt, Needham), $315 (B. Riley), $265 (Susquehanna); the stock's 2026 record is $316. My $270 target requires only a return to the pre-earnings trend, not new highs. The load-bearing input is Google TPU volume — if that is cut, the thesis is gone, which is why the invalidation is binary.
Broadcom — Wednesday is the next test. AVGO reports fiscal Q3 after the close on Wednesday, September 2: guidance is ~$29.4B revenue (consensus ~$29.43B, EPS ~$3.24), after a Q2 of $22.2B (+48%, record) with AI revenue of ~$10.8B. The Marvell precedent — beat, raise, -10% — says even a good print can get sold in this tape, and Broadcom additionally trades under the shadow of the Marvell-Google deal (the market read it as competitive pressure on AVGO's TPU-adjacent custom franchise; AVGO analysts pushed back). Don't hold the binary; buy the overreaction. If AVGO prints in-line-or-better and still sells into the $330–345 zone, that is the entry. A guide below a ~$16B/quarter AI run-rate invalidates.
ASML — the highest-conviction upstream holding for the remaining year. The July 15 quarter (beat, second FY26 outlook raise of the year) delivered the number that matters: CFO Roger Dassen said ASML is "close to being fully covered with Low-NA EUV orders for 2027" and is adding ~30% more Low-NA capacity next year (~56 systems planned for 2027; 60+ total EUV shipments in 2026). Memory makers are buying EUV for HBM4 architectures, not inventory — memory orders were up 71% YoY in Q4 2025, and the HBM shortage is now expected to persist through at least 2027. TSMC's deferral of High-NA to ~2027 is the known knock; it delays, not denies, and Low-NA demand more than fills the gap. Tariff risk cuts both ways (tools are exempt from chip tariffs but a capex slowdown would hit 2028 orders), and China demand remains the wild card. At $1,696 the stock is ~35% above where Morgan Stanley's January target ($1,624) sat — don't chase; stage adds at $1,550–1,650, which is exactly where a tariff-or-Fed-driven 5–8% semi correction would put it.
The tariff overhang is real but slow. The reported Section 232 expansion (chips + end products: servers, laptops, consoles) is "early phases," staggered, and months from final form — and the January precedent (25% on certain AI chips with a data-center exemption) shows this administration exempts what it wants built. A tariff that raises server costs slows the marginal data-center project at the margin; it does not touch the sold-out 2027 EUV book or NVDA's FY28 backlog. Treat tariff headlines as entry-creation events for the monopoly layer, not as a thesis change — unless the data-center exemption is formally removed, which would raise hyperscaler capex ~5–10% and genuinely slow the build (that is a falsification condition, see §8).
Memory is the quiet second derivative. HBM/DRAM tightness through 2027 is lifting prices across NAND and networking DRAM too — compute is ~$380B of 2026 hyperscaler capex (FactSet), roughly double 2025, partly because of memory price inflation. That is revenue for SK Hynix/Samsung/Micron and margin pressure for the hyperscalers — one more reason to own the receiving side of capex over the spending side.
Anthropic raised $65B at a $965B valuation on May 28, disclosed a $65B annualized revenue run rate in July (from $47B in May, $30B early this year, $10B in 2025 — the fastest revenue ramp in software history), filed a confidential S-1 on June 1, and is now holding test-the-water meetings where investors talk about a ~$2T float. Notably, its own prospectus will list the AI/data-center backlash as a risk factor — 7 in 10 Americans oppose local data-center construction (Gallup), ~$130B of projects were blocked or delayed in Q1 2026 alone (Data Center Watch), and the NRSC calls data centers a midterm "sleeper issue." A judge blocked the Pentagon's Anthropic blacklist on August 28, removing one overhang.
OpenAI filed confidentially May 22 (confirmed June 8, Goldman/Morgan Stanley running it), is targeting September at $1T+, and carries the tape's biggest known unknown: reported leaked financials showing a ~$39B loss, a Musk trial over the for-profit conversion in progress, and Friday's news that it is cutting Cursor's model access (November 12) after SpaceX's $60B Cursor acquisition — the AI stack is now vertically warring with itself. SpaceX (SPCX) already raised $85.7B in June and trades at ~42x forward sales with a heavy unlock calendar.
My view (opinion): The model companies' revenue ramps are real — Anthropic's especially — but the remaining-year market impact is supply, not fundamentals. Between OpenAI, Anthropic, follow-ons, and hyperscaler equity raises (Alphabet's record $84.75B in June; Oracle planning ~$40B for FY27), public markets must absorb well over $150B of fresh AI-linked equity into a 21x tape with a hiking Fed. That is a valuation ceiling mechanism: every dollar into the IPOs comes out of somewhere, most likely the crowded AI leaders. First consequence: expect multiple compression in listed AI leaders around the first mega-pricing — before any fundamentals break. Second consequence: hyperscaler stakes in the labs (Microsoft–OpenAI, Google/Amazon–Anthropic) get marked to public NAVs, which quietly supports GOOGL and AMZN — GOOGL is the cleanest listed way to own the Anthropic/IPO window (plus Gemini momentum, TPU now an external revenue line with Anthropic and Meta as customers, and the only hyperscaler outperforming the S&P over the last 12 months). Third consequence: if the first mega-IPO prices poorly, the private-mark-to-public-mark arbitrage unwinds and the whole complex de-rates at once — that is the single biggest left-tail for Q4.
The macro stack for the remaining year, in order of importance:
Inflation: headline PCE 3.7% YoY (July, above consensus), core 3.3%, CPI 3.4% — with tariff pass-through (Canada retaliation, chips) and a six-month-old Iran war still distorting energy (Brent fell 5%+ this week on the Venezuela deal and Gulf export recovery — a genuine disinflationary wildcard if it holds). Inflation is stuck ~1.3pp above target with the Fed already at 3.75%. The market's remaining-year debate is not "cuts vs. hold" — it is "hold vs. hike," and positioning is still catching up to that.
Breadth is the tell to monitor weekly: tech drove ~80% of the S&P's market-cap gain this year; on NVDA's best day 70% of the index fell. Concentration that extreme means index-level "dips" are mostly 5–8 stock events — which is precisely why this piece holds both "avoid IWM" and "buy selected AI leaders on dips": in this tape they are the same trade expressed twice.
Crypto: BTC $78.2K, ETF inflow streak broken Friday (-$201.8M), Fear & Greed 69 ("Greed") into a hiking Fed — a fragile combination. Dominance 59.5% says the debasement bid (30Y at 5.21%, $40T debt) is concentrated in BTC, not alts. Range-trade $72–85K until the FOMC resolves; a September hike likely flushes to the low $70Ks, which is the buy. ETH $2,457 — no edge versus BTC here.
Contents: A. Data snapshot · B. Scenario & multiple math · C. Primary-source extracts · D. Tariff mechanics · E. Second- and third-order effects · F. Historical analog · G. Catalyst calendar · H. Linked sources
| Asset | Level | Friday move | Note |
|---|---|---|---|
| SPY / QQQ | $769.35 / $716.43 | -0.23% / -0.65% | S&P ≈ 7,690; Nasdaq-100 weaker |
| SMH | $553.11 | -3.47% | Rejected at June channel resistance; below 50-day ($581.71) |
| NVDA | $217.55 | -4.58% | +8.7% Thu post-earnings; ATH $236.54 (May 14) |
| MRVL / AVGO / ASML / TSM / AMD | $216.62 / $368.79 / $1,696.16 / $417.52 / $465.58 | -10.3% / -0.7% / -2.2% / -2.3% / -2.3% | MRVL: record was $316 earlier in 2026 |
| MSFT / GOOGL / META / AMZN / ORCL | $513.53 / $346.59 / $578.02 / $266.43 / $150.85 | +1.7% / +1.7% / +1.2% / +4.0% / -0.7% | GOOGL only hyperscaler beating SPX over 12M (FactSet) |
| Fed funds / 2Y / 10Y / 30Y | 3.75% / 4.35% / 4.72% / 5.21% | 2Y +11bp, 10Y +4bp Friday | 10Y & 30Y at 2007-era highs; 10Y TIPS 2.41% |
| PCE (Jul) / core / CPI | 3.7% / 3.3% / 3.4% YoY | — | Headline beat consensus by 0.1pp |
| BTC / ETH / total crypto cap | $78,188 / $2,457 / $2.63T | +0.96% / +0.98% (24h) | BTC dominance 59.5%; F&G 69 "Greed" |
| TLT | $82.88 | -0.30% | Duration keeps losing |
Inputs: 2026 S&P EPS $340 (Goldman, +24% YoY); current forward P/E ~21x (88th percentile); index ≈ 7,690.
Multiple sensitivity (my arithmetic, estimate): a 50bp rise in the 10Y costs ~0.7–1.0 P/E turns at these valuation levels. Each full turn on $340 EPS ≈ $340 ≈ 4.4% of index value. 10Y 4.72% → 5.25% ≈ ~1.5 turns ≈ -6–7% index; high-duration AI names at 1.5–2x beta ≈ -10–15%.
Scenario outputs: Base 50% — 21x–22x held, EPS revisions flat-to-up → 7,500–8,000. Bull 25% — 22x on upward-revised ~$350 (Citi's number) → 7,700–8,200+. Bear 25% — 19.5–20x on $340 → 6,630–6,800 math; I assign the modal landing 6,900–7,200 because corporate buybacks and the Treasury put slow (not stop) the descent.
Load-bearing input: 2026 EPS $340. If tariffs and margin compression take it to $320, then even a flat 21x yields 6,720 — below my bear-case landing. That is the single input that flips the whole framework; Q3 earnings season (October) is when we find out.
Reported (Politico, Aug 27, eight sources; confirmed in outline by CNBC): a new Section 232 round on semiconductors and products containing them — laptops, data-center servers, gaming hardware — staggered rollout, early-phase, months from final. Context: January 2026 already imposed 25% on certain AI chips with a data-center carve-out; the 2025 "≈100% with domestic-build exemption" threat never fully materialized. Impact chain if the data-center exemption is removed: server BOM +8–15% → hyperscaler effective capex +5–10% → either margins absorb it (EPS risk, see Appendix B) or projects slip (receivers' revenue risk). Until formal language exists, treat as volatility source, not thesis.
The cleanest rhyme is 1999–2000: a Fed hiking into a tech-supply wave (record IPO issuance) with index concentration at extremes — the multiple broke before earnings did, and the toll-booth names (then: networking equipment) held longest before following. The 2025 counter-analog: Powell's dovish Jackson Hole hint triggered an aggressive rally; Warsh's 2026 speech did the opposite — same podium, opposite regime. Treat Jackson Hole signals as regime-defining for 3–6 months; they have been, both years.
Analysis and opinions are the author's own. Not investment advice. All prices are Friday, August 28, 2026 closing prices; markets are closed through Monday. Companion piece from a parallel session with additional primary-source depth: The Rest of 2026: One Hike, a $2 Trillion IPO, and the Bond Market's Veto.