Week recap · Macro · AI · semis · consumer · May 31, 2026
We were wrong on the math. The thesis broadened.
On Tuesday we laid out an outlook with three roughly-equal PCE scenarios (2.4–2.5%, 2.6–2.7%, 2.8%+) and a non-consensus call that the S&P’s eight-week streak was statistically a setup for mean reversion, with the base rate at 79% flat-to-down for week nine.1 By Friday at 4 PM, none of those held in the way we wrote them. Core PCE printed at 3.3% Y/Y — 50 basis points above our hawkish scenario.2 The S&P closed at 7,580.06, its ninth consecutive up week, the longest streak since 2023.3 Dell put up its best single trading day on record at +33%.4 And the 10Y, which we said was the cross-asset divergence the market was missing, rallied 25 basis points from its May 20 intraweek peak of 4.7% to a 4.45% close.5
That is the kind of week that forces a real conversation about what we got wrong. We were wrong on three specific calls: the mean reversion math, the AI-multiples canary, and the cross-asset divergence. Each of those was a non-consensus read in the May 26 piece, and each of them broke on the print. Calling that out matters more than backfilling a winning trade with retroactive logic.
And yet — the structural thesis held. AI infrastructure remained the dominant trade, and the market validated it across more names than we had positioned to, not fewer. Dell’s +33%, Salesforce’s +20%-area follow-through on a $1B Agentforce ARR confirmation, Marvell raising fiscal 2028 to $16.5B (+45% above FY27) — these are not "NVDA-and-everyone-else." They are a broader supercycle, distributed across the infrastructure stack. The Friday tape was not a contradiction of our thesis. It was a refinement of it.
The pivot is mechanical, not philosophical. Rotate within the AI complex, not out of it. Trim concentrated NVDA exposure into rallies (NVDA itself sold off ~10% over the back half of the month from the May 14 peak, even as the broader complex made new highs).6 Add to the infrastructure breadth trade: networking, custom silicon, AI software with monetization tells. Keep duration light into the June 17–18 FOMC and the June 6 NFP that comes first.
This is a recap piece. It is honest about what didn't work and specific about what did. The job is the same job — sit with the truth, write it down, and adjust the book.
1. Where we are: ninth week up, 10Y back through 4.50%
The S&P 500 closed Friday at 7,580.06, up roughly 1.4% on the week and up 5% on the month. The Nasdaq finished at 26,972.62, up roughly 2.4% on the week and 8% on the month. The Dow set another record at 51,032.46 (+0.72% on the day, ~+3% on the month). Nine consecutive up weeks is the longest streak since 2023, and May 2026 closes as one of the strongest months for the broad equity index of the past two years.3
The 10-year Treasury closed Friday at 4.45% — the lowest yield in more than two weeks and down approximately 25 basis points from the May 20 intraweek peak of 4.7%. Two things happened to drive that: tentative US–Iran de-escalation language pulled the geopolitical inflation premium out of crude, and the April PCE deflator printed +0.2% month-over-month against +0.3% expected, giving the market enough cover on trajectory to ignore the still-elevated 3.3% year-over-year reading.5
Chart 1 — S&P 500 weekly closes, late March through May 29
Nine in a row to 7,580 — the longest streak since 2023, set against a 10Y that came down 11bp on the week
S&P 500 weekly closes Mar 27 — May 29 2026. Friday May 29 close: 7,580.06 (+0.22% on the day, +5% on the month). The May 22 close (gold marker) was the 8-week peak we cited as the mean-reversion setup; the May 29 close (gold halo) is the 9-week extension that broke the call. Source: TheStreet & BBN Times Friday close summaries.
2. The May 26 scoreboard: what we wrote vs. what we got
On Tuesday morning we laid out a specific framework with specific calls. Here is what we said and how each one closed by Friday at 4 PM. Pass/Fail is whether the prediction matched the outcome, not whether the trade made money.
| Call | Our view (May 26) | What actually happened | Verdict |
|---|---|---|---|
| PCE scenarios | Three roughly-equal: 2.4–2.5% / 2.6–2.7% / 2.8%+ | Core PCE 3.3% Y/Y — above our hawkish-surprise scenario | WRONG (calibration) |
| 10Y direction | Hot PCE pushes 10Y through 4.65%, dollar strengthens | 10Y closed Friday at 4.45%, −25bp from Tuesday peak; dollar softer | WRONG (direction) |
| S&P 9th week | Base rate: 79% flat-to-down for week 9 after an 8-week streak | +1.4% on the week; 9th straight up week; longest since 2023 | WRONG (probability didn't matter) |
| AI infrastructure long | Stay long. Q2 guides validated the thesis. | DELL +33% Fri (record), CRM +13.8% Wk, MRVL FY28 raised to $16.5B (+45%) | RIGHT (broader than expected) |
| NVDA "sell the news" | Sell-the-news read; wait for a retest of the May 13 low zone to add | NVDA −10% from May 14 peak; rolled while index made new highs | RIGHT (rotation thesis) |
| WMT/HD bifurcation | 2001-style structural, not 2008-style cyclical; Costco confirms | Costco beat: comp +6.8%, e-commerce +13.6%, membership renewal 92.7% | RIGHT (consumer split holds) |
| AI multiples canary | If CRM/DELL/MRVL beat & trade flat-to-down, multiples are topping | All three beat — all three ripped. The canary was singing, not dying. | WRONG (premise) |
| Salesforce read | Agentforce monetization is the read on enterprise AI demand | Agentforce ARR > $1B (first quarter-end disclosure); AI+Data ARR $3.4B | RIGHT (cleanly) |
Four right, three wrong, one in-line. The three wrong calls were all from the “What Wall Street might not be thinking” section — the non-consensus reads. That is exactly where non-consensus calls should land or fail: at the asymmetries. Two of them (the mean reversion math and the cross-asset divergence) were technically the right setups and the wrong outcomes. The third (AI multiples canary) was the wrong premise.
3. PCE in detail: the level was hot, the trajectory was the story
The April PCE release dropped Friday at 8:30 AM ET. Headline PCE printed 3.8% Y/Y, up from 3.5% in March. Core PCE — the cut the Fed actually targets — printed 3.3% Y/Y, up from 3.2% in March. On a month-over-month basis, core PCE rose +0.2%, against +0.3% expected and +0.3% the prior month.2
Chart 2 — April 2026 core PCE: Y/Y (hot) vs M/M (cool)
The market traded the trajectory, not the level — M/M deceleration gave doves enough cover to push back
Y/Y core PCE picked up 10bp (3.2% → 3.3%) but M/M core PCE decelerated 10bp (0.30% → 0.20%, against 0.30% expected). The market favors the trajectory because M/M is the leading indicator on the Y/Y trend; the Fed targets the level. Both camps got something to work with, and that is precisely why the FOMC remains structurally split. Source: BEA, April 2026 PCE release; CNBC market summary.
This is the critical distinction we did not weight enough in the May 26 outlook. Our scenarios were anchored on the Y/Y level (where we missed by 50bp). The market traded the M/M trajectory (where the print was actually cooler than expected). That trajectory read is what gave the Fed’s dovish camp the framing they needed to hold ground at the upcoming meeting, and it is what let the 10Y rally 25bp from its May 20 high through Friday.
On the geopolitical side, reports of a tentative US–Iran de-escalation took the inflation-via-crude tail risk meaningfully lower in the front of the curve. WTI closed the week at $87.36, down approximately 17% for the month of May — the worst monthly performance since April 2025 — and the breakeven inflation curve (10Y nominal minus 10Y TIPS) compressed about 8–10bp on the week. The duration trade got a tailwind we didn’t price in.
4. The Dell day: structural significance of a +33% move
Dell Technologies reported fiscal Q1 2027 results after Thursday’s close: revenue $43.8B (+88% Y/Y), EPS $4.86 (+214%), AI server revenue $16.1B (+757% Y/Y), and full-year revenue guidance raised by $27B with the AI server revenue guide raised to $60B. The stock opened Friday up roughly 25% and closed +33% — its best single trading day on record.4
A +33% single-day move for a $200B market-cap name (Dell’s post-rally cap is roughly $273B; pre-rally it was around $198B) is not an earnings reaction; it is a re-rating. The implied delta on Dell’s forward multiple from that move is roughly 4–5 turns of forward P/E. The market just decided that Dell is not a low-margin server commodity company anymore — it is a strategic node in the AI infrastructure stack with structurally higher mix, structurally higher margin, and structurally more durable demand.
The complement to the Dell day was Salesforce and Marvell. CRM reported Tuesday after the close: revenue $11.13B (+13% Y/Y), Agentforce ARR crossed $1B for the first time, AI+Data ARR hit $3.4B, and FY27 revenue guide was raised to $45.9–46.2B. The stock finished the week up ~13.8%. MRVL also reported Tuesday: revenue $2.418B record (+28% Y/Y), custom silicon now 25% of data center revenue, and management raised the fiscal 2028 revenue target to $16.5B — about 45% above fiscal 2027.
Chart 3 — AI infrastructure leadership broadened — the new "supercycle" stack
Dell printed the best day on record. Marvell raised FY28 by 45%. CRM crossed $1B Agentforce ARR. NVDA itself rolled.
Weekly stock moves, May 26–29 2026. Dell’s +33% Friday move alone accounted for the bulk of the week’s gain; CRM, MRVL, and CSCO all confirmed strength. NVDA finished the week down ~5%, sliding from the May 14 ATH of $235.74 to Friday’s $211.14 close (~−10% off the peak). The supercycle leadership is broadening, not narrowing. Sources: TheStreet weekly recap; Yahoo Finance NVDA history.
5. The NVDA paradox: index up, leader rolling
NVIDIA closed Friday at $211.14, down 1.45% on the day and down approximately 5% on the week. The all-time closing high was $235.74 on May 14 — meaning the stock is now roughly 10% off the peak even as the broader index is at a new all-time high.6
This is the canonical signal of a rotation within a complex: the prior leader sells off, the prior laggards take leadership, and the index keeps printing higher because the supercycle thesis is broadening rather than breaking. We flagged this dynamic specifically in the May 26 piece (“the trade isn’t AI vs. not-AI — it’s high-multiple AI vs. lower-multiple infrastructure picks”) but we attached the wrong implication to it. We treated it as a setup for the AI complex to roll. It is actually a setup for the AI complex to broaden, with the original leader retracing while the next-tier infrastructure names get bid through their multiples.
The mechanical read for the book: NVDA is no longer the cleanest expression of long AI infrastructure at these levels. The cleaner expression is the diversified basket — networking, custom silicon, AI software with measurable monetization — with a relative-value lean toward names that just re-rated higher on their prints.
6. Where we were wrong (specifically)
Three calls broke. Each one matters because each one was a non-consensus read — the asymmetries we explicitly said the Street was mispricing. Owning the misses cleanly is the only way to keep the framework credible.
We said: 8-week up streaks resolve flat-to-down 79% of the time. Market said: nine.
The base-rate table held the historical record up — the 11-of-14 figure for week 9 after an 8-week streak is real. What we didn’t do is condition the base rate on regime. In environments where the streak coincides with a clean fundamental catalyst (here: NVDA-then-Dell back-to-back AI infrastructure validation), the conditional base rate likely flips. We treated unconditional history as a universal probability. It is a starting prior, not a prediction. Lesson: weight base rates against the fundamental tape, not against each other.
We said: 10Y at 4.56% + S&P at ATH = ERP compressed; hot PCE breaks both legs. Market said: 10Y rallied 25bp.
The setup was real (ERP was thin at Tuesday’s close). The catalyst direction was correct (PCE Y/Y was hot). The trade still didn’t work because we underweighted two non-PCE drivers: (1) the US–Iran de-escalation pulled the geopolitical inflation premium out of crude, and (2) M/M PCE printed below expectations, giving the duration trade a trajectory tailwind. We modeled a one-variable reaction function. The market is multi-variable and it kept the second-order risks honest. Lesson: cross-asset divergences need a clean catalyst without offsetting tailwinds elsewhere in the system.
We said: if CRM/DELL/MRVL beat and trade flat-to-down, the AI complex is topping. Market said: all three ripped.
This was the most fundamental miss because the premise was wrong. We extrapolated from a single data point (NVDA −1.77% on the print) into a system-wide claim about AI multiples. One name’s reaction is not a system — it is positioning. The Dell, CRM, and Marvell prints proved the rest of the system has room to re-rate higher even as the largest constituent in the complex digests its move. Lesson: don’t generalize a positioning unwind in one name into a multiple-compression thesis across the complex.
Three explicit losses on the non-consensus reads. The asymmetries we identified weren’t actually asymmetric — they were the consensus, dressed differently. That is a useful piece of information about our process going forward.
7. Where the thesis still holds
Three calls broke; three held cleanly. The ones that held are the ones that anchor the book going into June.
AI infrastructure long — right, just broader
The core structural thesis (the AI infrastructure supercycle is multi-year, accelerating, and broadening across the stack) was validated by the Q1 prints. The breadth is the new piece. The May 18 framework that the supercycle had to pass three tests, and the May 26 framework that PCE was the next gate — both held in spirit. The thesis was right and the trade is the same trade. The names doing the work changed.
WMT/HD bifurcation — right, Costco confirmed it
Costco reported Thursday after the close: total net sales $69.15B (+11.6%), comparable sales +9.8% (+6.6% adjusted for gas inflation and FX), e-commerce comp +21.5%, U.S./Canada membership renewal rate at 92.2%, worldwide renewal at 89.7%. That print, set against Walmart’s +4.1% U.S. comp the prior Thursday, bookends the high-quality consumer story we wrote up. The trade-down narrative is intact at the entry level; the upper-middle-income consumer is still showing up. The bifurcation widens by income band, not by overall consumer stress. That validates the “2001-style structural, not 2008-style cyclical” non-consensus framing — even as the other non-consensus reads broke.
NVDA-specific rotation — right, just in the opposite direction
We said NVDA would not be the cleanest expression of long AI infrastructure at the May 26 levels and suggested waiting for a retest of the May 13 support zone to add. The stock instead rolled from the May 14 peak of $235.74 down to Friday’s $211.14 close, a clean ~10% retracement that has held the May 13 support range. The direction was right; the level we were waiting for was the May low itself, not a specific dollar number. This is the kind of miss that is positive for the book because the structural call held and we positioned to add into weakness rather than chase strength.
8. The pivot: rotate within the AI complex, not out of it
The pivot is mechanical and small. We are not repositioning the structural thesis. We are repositioning the names that express it.
- Trim concentrated NVDA exposure into rallies, not into weakness. The May 14 peak to $211 was a healthy 10% retracement; we are not chasing it lower. If NVDA bounces back into the $220–225 range without a new catalyst, that is the trim zone for over-weight positions, redirecting proceeds into the broader infrastructure basket. Structural long stays on at neutral weight.
- Build the AI infrastructure breadth basket: DELL, MRVL, AVGO, ANET, VRT. The five names that benefit asymmetrically from the “broadening supercycle” thesis — server (DELL), custom silicon (MRVL/AVGO), networking (ANET), power/thermal (VRT). After Friday’s Dell move, the entry-window on this basket is tighter than it was a week ago, but the structural runway is now multi-quarter rather than position-trade.
- Add AI software with monetization tells — CRM is the cleanest one. Agentforce ARR crossing $1B with disclosed token throughput up 152% sequentially is the kind of monetization data point that lets the multiple expand. CRM was 33% YTD-down going into the print and has now snapped back ~14% in a week. Position before the next print, not into it.
- Keep the WMT/HD bifurcation pair on. Costco confirmed the high-end leg; HD’s soft +0.4% comp confirmed the discretionary leg. Pair-trade math has at least another quarter to run before the spread compresses.
- Light duration into June 6 NFP and June 17–18 FOMC. The 10Y at 4.45% has carry, but the June 17–18 meeting still has the procedural-hike risk we flagged. We are not net short duration; we are sized below benchmark.
- If the June 6 NFP prints below 100k with rising unemployment, add duration and add Russell. That is the symmetric pivot to scenario 3 from the May 26 piece — a labor-side dovish surprise that lets the Fed’s dovish camp take the meeting framing.
9. Looking into June: NFP, CPI, then the FOMC
The June macro tape is dense and the FOMC at the end of it is the next real test of whether the hawks have a procedural path or whether the dovish camp’s framing wins. Five major prints between now and June 18:
- Jun 2 (Mon): ISM Manufacturing — the production-side cooling signal we've been tracking.
- Jun 6 (Fri): May Nonfarm Payrolls — the labor read. Sub-100k with unemployment ticking up is the dovish setup.
- Jun 11 (Wed): May CPI — the dress rehearsal for the FOMC. Headline already above 3%; the question is whether shelter is finally cooling.
- Jun 12 (Thu): May PPI — the input-cost read on the deflation pipeline.
- Jun 17–18 (Tue-Wed): FOMC meeting. The question is not the rate decision — it’s the statement. Does the “easing bias” language stay or come out? Three votes is the bar.
The conditional pathway: if NFP prints sub-100k and CPI prints below consensus, the dovish camp gets a quiet meeting and the supercycle gets a tenth (and likely an eleventh) up week. If NFP holds 150k+ and CPI surprises higher, the hawkish three-vote bloc has a real shot at removing the easing-bias language — and that is when the index finally gets the catalyst for the pullback the base-rate math missed this week.
10. Outlook & framework into the FOMC
The forward window we have conviction on is the next 90 days — through the June 17–18 FOMC, the May labor and inflation prints between now and then, and the start of the Q2 earnings cycle in late July. Beyond that is dart-throwing. Here is the framework we are positioned for: a near-term S&P scenario set and a 10Y range. Single-name framework lives in the standalone pieces — see the MU after UBS and MU June outlook for the memory cycle; the Dell pitch is forthcoming.
Chart 4 — S&P 500 scenarios into the Q2 earnings cycle
Base case 7,500–7,800 (50%). Bull 8,000+ (25%). Bear 7,000–7,200 (25%).
Three roughly-equal-conviction scenarios anchored on the June 17–18 FOMC outcome and the early-Q2 earnings tape. Base case requires the easing-bias language to hold; bull case requires it plus a dovish surprise; bear case requires the hawkish trio to win the procedural fight and remove easing-bias.
10Y range
4.25–4.65% through June. Geopolitical de-escalation gives the rate side ~20bp of room to rally; the inflation level gives the rate side ~10bp of risk to back up. We size below benchmark on duration into the FOMC.
11. What would falter our thesis
Three risks we are actually watching — the ones with the highest probability and the highest stakes against the structural thesis. If any one of them trips, we adjust position sizing. If two trip simultaneously, we re-underwrite the book.
May NFP above 250k with average hourly earnings above 4% Y/Y.
A re-acceleration in the labor market with sticky wage growth is the cleanest signal that the disinflation trajectory we are leaning on at PCE M/M is breaking. If that print lands on June 6, the Fed’s hawkish trio gets the substance they have been arguing the procedural fight for. The 10Y goes through 4.65% immediately; the S&P loses 3–4% in two sessions; the rotation into AI breadth pauses.
Our response: Cut AI infrastructure gross by 25%; add Russell put spreads; extend duration light only when 10Y backs up through 4.75%.
META, MSFT, AMZN, or GOOGL pre-announces a CapEx reduction at a mid-year update.
The supercycle thesis depends on hyperscaler CapEx being structurally upward-revising, not flat or down. If any of the four announces a CapEx cut at a mid-year update (likely between June 15 and the Q2 print in late July), the AI infrastructure breadth basket loses its anchor. Dell, MRVL, AVGO, ANET are all derivatives of that CapEx line.
Our response: Net out the AI infrastructure basket; rotate to AI software where the demand-side monetization is cleaner (CRM, ORCL, SNOW). NVDA pair-trade against the basket (long NVDA, short basket) becomes the expression.
A high-profile private credit fund halts redemptions, or a regional bank discloses material CRE losses.
This is the off-radar tail risk that doesn’t show up in the equity tape until it does. Private credit AUM has doubled since 2020; the cycle hasn’t fully tested mark-to-market discipline. A KRE-style move (regional banks index down 15%+ in a week) is the symptom; the cause is likely commercial real estate marks or a fund-level liquidity event.
Our response: Cut gross exposure across the board (not just sectors); raise cash to 15–20% of book; add Treasury duration as the safe-haven trade. Wait for the policy response before re-engaging.
12. The one thing Wall Street might not be thinking about (yet)
One high-conviction forward call with explicit falsification by date.
META, MSFT, or AMZN raises 2026 CapEx guidance at a mid-quarter analyst event between June 10 and June 25 — not cuts it.
The Street is anchored on the “AI CapEx normalization” narrative. Dell, CRM, and Marvell printed the opposite on the supply side: $60B AI server orders at Dell, $1B+ Agentforce ARR at CRM, $16.5B FY28 target at Marvell. That kind of supply-side acceleration is not consistent with hyperscaler demand softening — it implies the demand side is tracking ahead of plan. The catalyst: any of META, MSFT, or AMZN hosts a mid-quarter analyst event between June 10 and June 25 with a CapEx revision higher. June is precisely the window when hyperscalers pre-announce ahead of Q2 earnings.
Trade: long the AI infrastructure breadth basket (DELL, MRVL, AVGO, ANET, VRT) into June 10. Trim post-announcement on the gap-up. Falsification: no CapEx raise from any of the four by June 25 = call dies. We do not double down.
Sources & footnotes
- May 26 outlook: "Eight weeks up. Now PCE has to confirm." Non-consensus #5 base-rate table cited 11-of-14 historical instances of 8-week streaks resolving flat-to-down in week 9. Full piece. ↩
- April 2026 PCE release, BEA: core PCE +0.2% M/M (vs +0.3% expected, +0.3% prior); core PCE +3.3% Y/Y (vs +3.3% consensus, +3.2% prior); headline PCE +3.8% Y/Y. CNBC coverage. ↩
- S&P 500 close May 29, 2026 at 7,580.06; +0.22% on the day; ninth consecutive up week (longest since 2023); +5% on the month. Nasdaq close 26,972.62 (+8% on the month); Dow 51,032.46 (record). TheStreet and BBN Times Friday close summaries. ↩
- Dell Technologies fiscal Q1 2027 earnings, reported May 28 after the close: Q1 revenue $43.8B (+88% Y/Y), EPS $4.86 (+214%), AI server revenue $16.1B (+757% Y/Y); FY27 revenue guidance raised by $27B, AI server guide raised to $60B. Friday May 29 close +33% (best single trading day on record). Sources: Dell 8-K filing; CNBC, Seeking Alpha, Yahoo Finance coverage. ↩
- 10-year Treasury yield close May 29, 2026 at 4.45%; intraweek peak of 4.7% on May 20; ~25bp decline driven by US–Iran de-escalation reports and soft M/M PCE. WTI closed at $87.36 (-17% for May). TradingEconomics/Fed H.15 summary. ↩
- NVIDIA Corp (NVDA) close May 29, 2026 at $211.14, −1.45% on the day, ~−5% on the week, ~−10% from the May 14 all-time closing high of $235.74. Yahoo Finance historical data. ↩
Nothing on this page is investment advice. Forward-looking statements are scenarios, not promises. See disclaimer.