@dailyanalysts · Published Sunday, August 30, 2026 · Rest-of-year roadmap: general market, AI chips, upstream (ASML / Marvell), AI model companies, and the debt–rates–Fed–inflation knot. All equity prices are Friday, August 28, 2026 closes (market closed at publication); crypto quotes are Sunday 06:37 UTC. Live data via the Finnhub/CoinGecko handlers; primary sources read directly (linked in Appendix H).

1. What just happened — and why the rest of the year is already trading

Friday was the whole rest-of-2026 trade in a single session. Fed Chair Kevin Warsh — 100 days into the job — used his first Jackson Hole keynote to say the Fed's "predominant focus right now should be on prices," that he would "be hard pressed to describe broad financial conditions as restrictive," and that unless underlying inflation is moving to 2% "clearly and at sufficient speed… we have work to do" (the speech). September hike odds went 33% (July 31) → 57.5% (Aug 26) → ~63% post-speech (CME FedWatch via Morningstar). The 2-year yield jumped 12bp on the day; the long end barely moved.

The tape split exactly along the fault line that will define the next four months:

Wednesday, Nvidia reported $96.2B of quarterly revenue (+106% YoY) and — for the first time ever — a full-year forward outlook: fiscal 2028 revenue growth of ~70%, versus the 45–50% the street had modeled (press release; CFO Colette Kress on the call, per Schwab and CFO Dive). Two days later the stock sits 8% below its May 14 high. The market's problem with AI is no longer demand. It is the price of money and the quality of the cash flow.

2. The core argument: everyone is watching the wrong end of the curve

OPINION The market is obsessing over whether Warsh hikes on September 16. That is mostly the wrong question. A funds rate at 3.75–4.00% instead of 3.50–3.75% changes almost nothing for a company doing $96B quarters with 75% gross margins. What changes everything is the long end: the 30-year at 5.21% with a 5.31% cycle high — its highest levels since 2007 before Treasury stepped in — held down only by a Treasury that is now buying back its own long bonds (operations doubled to "at least" $4B per operation, running September 9 through November 4, per CNBC) while the Fed pointedly refuses to monetize. Bessent is trying to cap the long end; Warsh spent Friday renouncing forward guidance and telling markets to set the price. Someone is going to lose that argument, and the loser decides the S&P's multiple.

Here is the claim in one sentence: the rest of 2026 is a race between AI earnings power and AI + Treasury debt supply for the same pool of savings. Q2 earnings season printed 52% S&P EPS growth — still 26–29% after stripping the mega-caps' one-time investment gains (Schwab, from 485 reported companies; 88% beat on EPS). Against that, Moody's now counts $1.2 trillion of hyperscaler lease commitments (over $820B of it for data centers not yet built) and ~$460B of direct debt across six hyperscalers, with capex heading to $785B this year and ~$1T in 2027 (CNBC on the Moody's note). Every one of those dollars competes with Treasury issuance for the same buyers. That competition — not the September dot — sets the price of the 30-year, and the 30-year sets the multiple. If earnings keep winning the race, one hike gets absorbed and the S&P makes new highs by December. If the long end breaks 5.31%, nothing else in this piece matters: chips derate 15–25% regardless of their guides, and every scenario below shifts one notch bearish.

Numbers with consequences:

• Sept hike odds 63%, October cumulative ~88%, December: under ~10% for zero hikes, modal year-end funds rate 4.00–4.25% (two 25bp hikes). Each 25bp adds roughly 13% to floating-rate private-credit book income (that is the OBDC long, below) and compresses high-duration software multiples by roughly 10–12%.
• Core PCE 3.3%, and the six-month headline running 4.1% (Warsh's own framing in the speech): the trend is not improving. He said the quiet part out loud: summer's better prints "do not tell me that underlying trends have meaningfully improved."
• 54% of the PCE basket inflating above 3% (vs. 32% pre-pandemic). That is why one hike is a floor, not a ceiling — the distribution of inflation is still wide even as the average stalls.
• 30Y at 5.31% = the regime line. A weekly close above it and I cut AI beta across the board, cover the IWM short into weakness rather than at $282, and de-risk the entire private-credit book (HYG below $78 is the kill switch on the existing credit calls).

3. The Fed, inflation, and the debt overhang

What is priced. Current target: 3.50–3.75%. CME FedWatch (Aug 28, 13:45 ET, via Morningstar): September 16 hike ~63%; October 28 cumulative ~88%; by the December 9 meeting a hike is effectively fully priced, with the most likely year-end landing zone 4.00–4.25%. The committee is more hawkish than the chair: the July 29 hold came with three dissents in favor of a hike, Cleveland's Hammack said days later it is "now the time to act," and the June SEP lifted the 2026 core PCE forecast to 3.3% from 2.7% with roughly half the dots seeing at least one hike. Warsh gave no forward guidance and promised none — "committed to a discipline, not to a decision," and a "quieter Fed." Translation: every CPI print between now and December is an FOMC meeting. That is a volatility regime, not a levels regime.

Inflation's shape. July: headline PCE +0.2%/3.7% YoY, core +0.2%/3.3% YoY (CNBC). Goods fell 0.1% on the month; services rose 0.3%, with financial services +1.2% and housing +0.3%. Michigan 1-year expectations eased to 4.0%, but 5-year sits at 3.3% — exactly the core rate. The war complicates the exit: the Strait of Hormuz is running ~13 ships/day (Kpler, via Fortune), Brent printed ~$87 mid-August, and WTI ended the week at $83.30 (−4.3%) only because of the Venezuela 65-billion-barrel arrangement and the sanctions escalation. Energy supply risk plus a 3%+ core means no cuts this year; the entire 2026 easing narrative died with the war.

The debt side. Treasury's doubled buyback (Sept 9–Nov 4) bought one good day — 10Y −5.7bp to 4.647%, 30Y −9bp — and the effect faded within 24 hours. El-Erian called it "small in both absolute terms and relative to net issuance," "a broader deployment of 'yield curve control'." Evercore's Krishna Guha: it "changes almost nothing in terms of the fundamentals — in particular the unchanged need to finance the tidal wave of hyperscaler debt in addition to very large government deficits." Rogoff argues the interventions are squandering credibility without a deficit plan; Bessent has since promised "a new fiscal plan." I will believe the plan when I see the refunding calendar. Behind all of it sits the $41.1 trillion debt ceiling with a projected mid-2027 X-date; Democrats lead the generic ballot by ~6 points ten weeks out from November 3, and TD Securities expects the ceiling fight "to go down to the wire" once a divided Congress owns it (CNBC). Bills maturing around the X-date are already trading at a premium. The midterm year pattern (weak Q3, strong Q4 post-election; divided Congress historically better for stocks since 1950 per JPMorgan) holds — provided the long end cooperates.

OPINION My call on the Fed: one hike — September if August CPI (due ~Friday, September 11, five days before the FOMC) confirms at ≥0.2% core, October if the data blips soft — then a hawkish hold in December with dots implying a 2027 terminal around 4.0–4.5%. El-Erian says 60% September odds are "too high" (anchored expectations, AI productivity, housing strain). He may be right about the level. But Warsh has spent his first 100 days making inflation credibility the entire identity of his chairmanship — after Friday's speech, the reputational cost of not delivering a hike exceeds the cost of delivering one. Expect the hike; the trade is in what it does to the long end and to small caps, not in the dot itself.

4. AI chips: Nvidia's guide changed the year — the balance sheet will decide it

NVDA, $217.55. The arithmetic first, because it is the most under-appreciated fact in the market right now: FY27 is tracking to roughly $400B+ ($177.8B in H1 + the $108B Q3 guide + a seasonally larger Q4), and management guided FY28 at +70% — roughly $680B — versus a street that was at +45–50% before the print. Zero China data-center compute is assumed in the outlook. Gross margin 75.0%. OPINION My arithmetic on those guided inputs: at 74–75% gross margins and operating expenses growing far below revenue, FY28 net-margin power of 55–60% implies EPS of roughly $15±; at $217.55 the $5.4T market cap is ~14–15x that year's earnings — cheaper than the S&P 500's ~21x, with triple the growth. Consensus targets moved toward ~$300 after the print (pre-print average was ~$297 per IG; BofA at $350). The stock is not expensive. It is heavily levered to the long end, which is why it fell 4.6% the day after the best guide of the year.

Now the part consensus has not modeled — the quality of the cash flow, from the release's own statements of cash flows and balance sheet:

• Receivables grew 3.5x faster than revenue: accounts receivable +$24.6B in one quarter to $63.1B (+64% QoQ) against +18% QoQ revenue. That is customers (increasingly financed ones) stretching payment terms on $40k+ rack orders.
• Cash conversion is diluting: Q2 operating cash flow $24.1B versus $59.7B of net income (0.40x); H1 OCF $74.4B vs. $118.0B (0.63x). Earnings are real; the cash is arriving late.
• The circularity is measurable now: marketable equity securities jumped $12.9B → $42.8B and non-marketable investments $22.3B → $51.2B — ~$94B of ecosystem stakes; long-term debt went $7.5B → $32.4B on $24.9B of new issuance; $2.9B went into Groq; and the quarter's headline announcement was compute-financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman and KKR "to mobilize over $500 billion of third-party capital." NVIDIA increasingly sells to counterparties it invests in, lends to, or guarantees.

None of this is fraud-shaped; all of it is 1999-shaped. The tell to watch at the November Q3 print: OCF/net income. A second consecutive quarter below ~0.5x means the growth is being bought with balance sheet, and I trim on that alone regardless of price. Until then, the washout into a live FOMC is a gift: buy NVDA at $198–212, target $248, invalidation a weekly close below $190. Two independent signals: the FY28 guide shock (+70% vs. +45–50%) and a selloff driven entirely by macro repricing rather than any demand datapoint — every piece of upstream evidence (ASML's bookings, Marvell's guide, Broadcom's backlog) points the other way.

AVGO, $368.79 — Wednesday September 2 is the first domino. Broadcom reports fiscal Q3 after the close: consensus ~$29.4B revenue (+84% YoY), EPS ~$3.22–3.24, 67% operating margin. The only number that matters is the fiscal 2027 AI-semiconductor target — management kept it at "over $100 billion" in June (the market wanted ~$130B; Goldman models $133B) and the stock fell 12.6% the next day from its $495 all-time high. It remains ~23% below that high with AI backlog over $30B against $10.8B shipped last quarter and street targets averaging ~$508 (range $390–630). Levels: a raise that lifts FY27 AI guidance clearly above $100B reclaims $400 and re-rates the whole custom-silicon complex (read-through to MRVL); unchanged phrasing risks the June pattern toward $350 then $320. WATCH — no position into the print; the options market will charge too much for the binary.

AMD, $465.58. OPINION The weakest risk/reward of the three U.S. designers: up ~120% on the year, ARK is trimming, and I have no fresh catalyst I can price. Pass. The AI-chip exposure is better expressed through NVDA's valuation gap and the upstream names below.

5. Upstream: ASML and Marvell — the two cleanest reads on 2027

ASML, $1,696.16. Read the primary source, because it is the strongest guidance chain in the entire market: 2026 revenue guidance has been raised twice in six months — €34–39B (January) → €36–40B (April) → €43–45B (July 15) — with Q3 guided to €11–12B at 55–57% gross margin, a ~28% sequential jump. The CEO's own words: order intake "remained extremely strong," and ASML is planning +30% capacity on ~65 low-NA EUV systems for 2027 (investigating another +30% for 2028) and +30% on ~130 DUV immersion tools — visibility management explicitly ties to AI logic and memory demand through 2027–28 (the release). The second demand leg is memory: SK hynix's CEO says the shortage persists through 2030 with a $4B Indiana fab and HBM4E by 2029. At 57x trailing earnings within 3% of the June 30 high ($1,741), you are paying full price for known-good numbers — so the trade is the September dip, not the breakout: buy zone $1,560–1,650, target $1,900, invalidation a weekly close below $1,480. Honest caveat: if the dip never comes, the call simply doesn't fill — the cost is opportunity, not loss. On the tariff scare: the sweeping semiconductor tariff under consideration (below) taxes chips and servers imported into the US; ASML's tools overwhelmingly ship to fabs in Taiwan and Korea, and US fabs — TSMC's $265B Arizona complex — still need its machines. The January Section 232 action covered equipment but with carve-outs. ASML's tariff exposure is retaliation and quota games, not demand.

MRVL, $216.62. The market just punished the only company in the complex that raised twice in one call. Q2 revenue $2.739B (+37% YoY), data center $2.17B (+46%), non-GAAP EPS $0.94 (+40%); FY27 raised to ~$12B (+45%); FY28 raised to ~$18B (+50%) with data center growing >60% and custom silicon more than doubling (full call transcript). The Google warrant disclosed in the ~Aug 19 8-K is worth ~$12.2B tied to up to $120B of procurement over roughly six years — JPMorgan's Harlan Sur did the math live on the call: ~$18.5B/year at full milestone run-rate. Management says FY28 guidance already includes it and the real step-up lands in FY29 and beyond (prior FY29 custom target was $10–11B; now "greatly increased," quantification promised at the October 6 Investor Day). Why did it fall 10.3% on a beat-and-raise? Margin mix — Q3 non-GAAP gross margin guided down to 57.5–58.5% as lower-margin custom ramps — plus a post-Nvidia expectations reset. That is a sentiment problem, not a demand problem. The existing call is now live and filled at $216.62 inside the $210–225 entry zone: target $270, invalidation a weekly close below $195 or any disclosed cut to Google TPU volume/timeline. Street targets ($300–315 at Craig-Hallum, Needham/Rosenblatt, B. Riley) sit above mine; I stay below street because the margin-mix compression is real. Catalyst path: AVGO's print Sept 2 (read-through), Investor Day Oct 6 (quantification). Size small — one signal, sentiment-driven, and the FY29 back-loading gives the market months to get bored.

TSM, $417.52. The direct tariff target (90%+ of cutting-edge supply sits in Taiwan; the January Taiwan deal permits only partial duty-free imports). Friday's −2.3% is the tariff discount. Its pricing power will pass costs through, but between a live tariff process and a geopolitical premium I prefer the upstream expression through ASML and the design layer through MRVL — no standalone call. OPINION

The tariff overhang — the single most mispriced Q4 risk. Per Politico and CNBC, the administration is weighing a sweeping semiconductor tariff that would, for the first time, potentially drop the January data-center exemptions — hitting servers, laptops, and the imports the AI buildout physically depends on, with duty-free quotas tied to US investment pledges. One industry official: "The math literally just does not work… Those are chips we physically can't buy here." If broad tariffs with no exemptions land before the late-October hyperscaler prints, 2027 capex guides get cut and this entire roadmap shifts a notch bearish. Probability-weight it: the tech lobbying blitz is enormous and the phase-in language suggests slow-walking past the election — but it is the load-bearing tail risk in every chip name here.

6. AI model companies: the $2 trillion stress test

The divergence nobody is underwriting properly: Anthropic's run-rate hit $65B at the end of July — 7x a year ago, Q2 preliminary revenue $11.5B, up more than 140% sequentially — while OpenAI's run-rate is $40B on Q2 revenue of $6.7B, up just 18% QoQ (Axios, CNBC). In five months the enterprise-first model overtook the consumer-first model. Even Warsh cited it from the podium: token sales at the two leading labs exceed $100B annualized, +500% YoY. Efficiency is the next battleground, and Anthropic's enterprise mix is currently winning it (a federal judge also just blocked the Pentagon's attempt to blacklist Anthropic — that risk is off the table for now).

The event: Anthropic is targeting an October IPO at a reported ~$2 trillion valuation — which would be the largest public offering ever — with investors expecting a $100–120B run-rate by year-end; Morgan Stanley, Goldman and JPMorgan are running it (FT via Fortune; Axios). OpenAI: $852B valuation, confidentially filed, "will be a public company in 2027" or sooner per CFO Friar, $122B raised in March — but with three senior executive departures in eight weeks and a Musk trial verdict pending, it is the messier file. SoftBank is reportedly seeking another $10B loan just to fund its OpenAI stake. The ecosystem is vertically integrating at speed: SpaceX (public, volatile) bought Cursor for $60B and OpenAI is cutting Cursor's model access November 12.

OPINION How I'd play the IPO (WATCH, with discipline): this is the market's first public pricing of frontier-AI terminal value, and the information is worth more than the position. At $2T on July's $65B run-rate it is ~31x trailing; on the projected $100–120B year-end run-rate it is ~17–20x forward. Take primary allocation only if the roadshop shows Q3 revenue continuing to compound at ≥50% QoQ (i.e., ≥ ~$17–20B for the quarter); decline a deal priced above ~$700–800B if sequential growth is decelerating toward OpenAI's +18% pattern — that is the comp that says 21x for a decelerating consumer platform is the ceiling, not the floor. Two signals matter more than the trade itself: a clean $2T print that holds re-rates every proxy (GOOGL, MSFT, META, NVDA); a postponement or a sub-$1T clearing price is the first public mark-down of frontier terminal value — in that case cut the MRVL and ASML targets by roughly a third. That is the highest-information event of Q4.

Public proxies — no standalone entries, and here is why. Friday's rotation is the model: through a hike window, own the cash-flow owners, rent the hardware growth. GOOGL has the strongest stack position (TPUs via Broadcom and Marvell, an Anthropic stake, and an $84.75B equity raise in June to fund buildout — dilution now, structural cost advantage later); META carries the $18B child-safety settlement headline but printed a strong quarter; MSFT's rerating is "probably over" near-term per the downgrade crowd — hold, don't chase. After the first hike lands, rotate the tilt back toward hardware beta. OPINION

7. Crypto: positioned wrong for a hiking Fed

BTC $78,188 (Sunday 06:37 UTC). The August short squeeze — $62k to $81k, liquidating nearly $13B of shorts per Glassnode — ran directly into Warsh: −3.2% Friday with $488M of liquidations, closing below the $81k trigger. Fear & Greed at 69 ("Greed") three days after a 24% squeeze is the definition of crowded longs. The May playbook is the warning: April CPI at 3.8% produced a record 10-day ETF outflow streak. To be fair, the futures structure is cleaner than it was — Schwab's liquidation-fuel work shows long fuel near ~$2.5B within 10% of spot, and the $6.4B Deribit expiry cleared Friday's stale positioning. Levels: $81k reclaim = decoupling resumed (chase toward $90k); $68–70k = the squeeze fully unwound (max-pain zone); $62k = bear market resumed. ETH $2,457 (Aug 21 high $2,547, best since February); SOL $104.95, +12% on the week — the disinflation vote passed (issuance floor reached ~2029, ~18.9M fewer SOL), Bitwise's staking ETF first to $1B. JPMorgan cut its 12-month BTC target to $112k in March. No trade. A hiking Fed into Greed-positioned crypto is a fade-first tape.

8. The trade board

Existing calls carried (levels are entry specs, not live quotes): OBDC long (PC-HIKE-1) and FSK short (PC-HIKE-2) — the direct rate-impact pair, unchanged; ARES short (PC-HIKE-3) small; VIX hedge (EQ-VIXHEDGE-1) live through the September 4 jobs print; PYPL watch (no trigger yet). Bookkeeping: the CRM/CRWD re-entry zones ($215–225 / $200–210) are expired — both stocks rallied hard on their August prints and the levels are dead; removing them rather than carrying stale marks. Disclosed conflict: the IWM target ($282) runs through $290 — the level that trips the private-credit warning board (PC-HIKE-4) and forces de-risking of OBDC-type books. That is coherent (same trade, different durations), but if IWM breaks $290 I cut the credit book at the same time I bank the equity hedge.

9. Three scenarios for December 31

Load-bearing assumption, named: the Fed hikes once and stops, because Q4 growth slows as the AI-capex impulse annualizes and the consumer (Michigan 51.7, confidence 89.4, savings depleted) buckles under 4%+ front-end rates — note the tension: Chicago PMI at 47.1 (contraction) against Atlanta's GDPNow at +4.6% for Q3; the slowdown is coming, the timing is not. If the assumption is wrong and they hike twice with more signaled, the base case becomes the bear case: subtract 10–15% from every chip target and treat the 30Y as the only chart that matters. The second load-bearing input is the $905B→$1T 2027 hyperscaler capex consensus: its financing is the most fragile link in the chain (Appendix D).

10. What would prove me wrong

• Jobs, Friday Sept 4: payrolls <100k with average hourly earnings ≤0.2% MoM → hike odds collapse below 35% → the IWM short, the NVDA dip-buy thesis, and the hawkish framing all break. Shift to the bull case immediately.
• CPI, Friday Sept 11: core ≥0.3% MoM → bear case triggers; expect the hike plus a December one priced; test of 5.31% on the 30Y within weeks.
• 30Y weekly close above 5.31% → term-premium regime change; cut AI beta regardless of anything the Fed says or does.
• AVGO, Sept 2: FY27 AI guidance kept at ">$100B" with bookings momentum slowing → custom-silicon complex derates; cut MRVL target to ~$230 and the NVDA thesis loses its strongest corroborating datapoint.
• Anthropic IPO: postponed past November, or Q3 revenue < ~$15B → frontier-terminal-value markdown; cut MRVL/ASML targets ~one-third.
• Any hyperscaler Q3 print (late Oct): a capex cut, or a disclosure event around SPV/lease financing → the $905B 2027 consensus dies and the base case with it.
• HYG weekly close < $78 → per the standing regime board: exit everything, including OBDC. This is the credit-complex kill switch.

11. What to do — the general suggestion

OPINION Now through September 16: run underweight beta. The IWM short/trim is in its zone right now; the VIX hedge stays live through the jobs print; inside tech, own platforms over hardware (Friday's tape is the template). September is historically the worst month for the S&P and this one arrives with a live hike, a tariff headline risk, and narrowing breadth (equal-weight S&P below its 20-day, Russell below its 50-day, only 49% of Nasdaq names above their 200-day). Let the market come to your limits: NVDA at $198–212, ASML at $1,560–1,650.

After the first hike lands — rotate the hedge off and add hardware/upstream beta. When earnings are accelerating at 25%+, the first hike of a cycle has marked local equity lows (1998–99, 2004); when earnings are decelerating (2022), it didn't. This cycle has the earnings. That is why I buy the dips, not the break.

Outside tech, where the mandate asks for it: XLE is the one sector that rallied on the hawkish day — WTI $83 with Hormuz at 13 ships/day is a persistent supply premium (the Venezuela arrangement is the offset, so size accordingly). Front-end T-bills at 4.3%+ are a position, not a parking lot: they yield more if the hike lands. Gold at $4,550–4,600 already prices the inflation story; no edge, pass.

On the credit side of the same rate trade: keep the OBDC long / FSK short pair — the floating-rate accretion (~13% NII per 100bp) versus the stressed book — with the disclosed conflict above, and keep the Anthropic allocation discipline as written. Crypto: flat until $81k.