AI Complex — Cross-Layer Outlook (Ai Complex Sector)

Executive Summary & Action Plan

Verdict

NEUTRAL (selective) — after re-underwriting both remaining value candidates with adversarial verification, exactly one name in the 37-stock AI complex still clears our bar for new capital: Adobe (ADBE). The July 21 rotation into silicon momentum pushed two-thirds of the universe 25–80% above verified fair value, and it demoted the other candidate, Salesforce (CRM), to the sidelines.

This report is struck deliberately before tonight's Alphabet and ServiceNow prints — the first of an eight-week earnings gauntlet that decides whether the rotation was flow or regime.

Price $227.16 | Market cap $90.3B | Target $290.00 | Upside +27.7% IWANNAVY Fair Value (re-underwritten 2026-07-22) · price as of 2026-07-21 close · Street consensus $273.28 (Hold, n=40, reference) · Morgan Stanley cut to Underweight $240 on 7/21 — our mark differs because the same cash-flow read was being double-counted against a $336 DCF, and the residual discount is governance timing, not demand

Abstract

We re-priced the entire four-layer AI coverage universe at the July 21 close and re-underwrote the only two BUY-rated names, each mark rebuilt from scratch and then adversarially audited. The audits cut in opposite directions: Salesforce fell to $192 and HOLD because its prior mark quietly underwrote multiple expansion into a documented de-rating, while Adobe rose from the analyst's own $273 to $290 BUY because the cut below $300 double-counted one cash-flow read against a $336 DCF and anchored on the street mean. The July 21 session that ripped silicon 8–18% and sold software was one Morgan Stanley rotation note colliding with a short-covering squeeze that received two genuine catalysts — NVIDIA's 9.3% Nebius stake and a Cerebras–CrowdStrike inference deal — not a demand break; hyperscaler capex is still being raised. The stance is deliberately narrow: one cheap, governance-discounted cash machine on the long side, discipline everywhere else, and dated falsifiers on both legs. Principal risks are seat cannibalization, frontier-lab release shocks, and a contracted memory cycle extending the squeeze.

Forecast Path

The complex-index path prices continued multiple compression against intact 2026 earnings; bands are dispersion, not scenario endpoints. Probabilities are derived from estimate spread and gap asymmetry, judgmental, not calibrated.

Action Plan

One position: build Adobe (ADBE) in thirds — a quarter at market, add at the $219 support shelf, complete at $205, stop below the 52-week low at $188.50. The tactical caveat is real: with the CEO and CFO seats both vacant and no print until September 10, the near-term path is sideways-to-down despite the value, which is why the ladder leans on the lower rungs. Everything else in the complex is inherited discipline — nothing in the silicon momentum cohort (Arm, CoreWeave, Micron, Intel, Astera, Cerebras) survives its gap to fair value, and the mega-cap prints this week are events to observe, not to front-run.

• Risk/Reward (ADBE) 1.63:1 | Prob-weighted 12M return +15.4% (Bull 27% × +56% + Base 31% × +28% + Bear 42% × −20%) | Confidence Medium | Key monitor: AI-first ARR trajectory at the 9/10 print
• Salesforce (CRM) fails the entry gate: at $170.06 against a $192 fair value and a $145 technical stop, reward-to-risk is 0.88:1 — below the 1.5:1 house minimum. HOLD, revisit below $160 or after the 9/2 Agentforce print.

---

The Complex, Ranked

Every covered name against its IWANNAVY fair value at the 2026-07-21 close. Marks dated 07-22 are freshly re-underwritten; 07-21 marks are yesterday's adversarially-verified silicon set; 07-12 marks are inherited from the software sweep and re-priced.

The structure is unambiguous: every name with meaningful upside is the application layer the market just reset, and everything 40%+ rich is the silicon momentum the market just chased. The complex's value migrated down the price stack while money rotated up it.

---

What Happened

Monday's session (7/21) was two verifiable desk events colliding with a squeeze, not a regime change. On the software side, one Morgan Stanley note reset the sector: Salesforce cut to Equal-Weight with the target slashed $287 → $185, Adobe cut to Underweight $365 → $240 with explicit rotation language ("cleaner growth and AI monetization elsewhere"), and Workday downgraded the same day. On the silicon side, two genuine catalysts landed on the most-shorted cohort: NVIDIA disclosed a 9.3% passive stake in Nebius (filed 7/21, NBIS +18.8%) and Cerebras won a CrowdStrike inference partnership to power Falcon AI Detection & Response (CBRS +17.9%). The magnitude — double-digit moves in recent IPOs and memory — carries the classic signature of short-covering into an oversold tape, four days after the Kimi K3 efficiency shock that the market has now re-read through the Jevons lens: cheaper inference means more tokens, which is accretive to infrastructure even as it pressures the app layer's AI premium.

What it was not is an earnings event. Nothing in the drawdown-to-rebound sequence since June 22 has come from guidance: TSMC raised both revenue and capex mid-selloff, Samsung printed the largest quarterly operating profit in technology history, and 2026 hyperscaler capex aggregates to $750–780B, up 77%, still being raised. The fundamental verdicts start arriving tonight.

The Gauntlet — What to Watch Now

Eight weeks of scheduled evidence, with the two heaviest days in the next seven sessions. Dates confirmed unless marked (E).

---

Top Pick: Adobe (ADBE)

The cheapest quality name in the software map: 9.4x EV/EBITDA, a 10.2% trailing FCF yield with SBC fully expensed, GAAP operating margins in the mid-30s, and a buyback retiring roughly 10% of the float a year — sold on a factor rotation and a governance vacuum, not on demand. The last print was a beat-and-raise. The discount has two named causes: the dual CEO/CFO vacancy following the Narayen transition (the search is dragging, and Morgan Stanley and UBS both capitulated on 7/21 as it dragged), and the freemium pricing pivot the bears read as ARPU dilution and we read as funnel expansion pending proof.

The mark carries an honest dispute band, disclosed rather than smoothed: Morgan Stanley $240, our own analyst's re-underwrite $273, the verified house mark $290, the standalone DCF $336. The verifier's correction is methodological — the $273 came from two collinear cash multiples outvoting the DCF two-to-one and landing on the street mean to within 0.1%, an anchoring tell — and the $290 restrike is the independence-corrected midpoint less a genuine no-catalyst haircut for the eight-week wait to the September 10 print.

The scenario cone prices the governance timing honestly — the bear weight is the largest single weight, and the case is still positive because the downside is short and the upside is long.

• Base assumptions (3): AI-first ARR exits 2026 above a ~$700–800M run-rate; Creative Cloud net-adds stay positive through the freemium pivot; the CEO seat is filled by year-end without a kitchen-sink guide reset.
• This is wrong if: generative design reaches mass-market good-enough and Creative Cloud net-adds turn negative; the CFO transition produces a guidance-framework reset; or open-weight commoditization collapses AI credit pricing before conversion proves out. p(thesis wrong) = 45%.

The Demotion: Salesforce (CRM)

Both re-underwrites were audited; Salesforce's did not survive. The refreshed $210 depended on 14x forward EBITDA against the stock's own 11.5x, and an 8.2% target FCF yield against the 10.2% it trades at — roughly 20% of multiple expansion assumed at the exact moment the tape and the sell side are compressing it. Struck at no-expansion, both multiple legs collapse to spot. The corrected mark is $192: the reverse-DCF crux is intact (the market still prices sub-2% perpetual growth for an 11% grower with Agentforce compounding at +205%), but conviction requires the multiple to expand, and nothing on the calendar forces that before September 2. At +12.9% upside with a 0.88:1 reward-to-risk to the technical stop, it is a HOLD with a positive-asymmetry print ahead: Morgan Stanley just reset the bar to $185, so an Agentforce reacceleration on 9/2 rejects the downgrade from a lowered baseline.

What Would Make the Rotation Right

The steel-man deserves its own section because it is strong. The seat-cannibalization case is not hypothetical — Klarna publicly cut Salesforce seats after deploying agents, Benioff froze engineering hires on AI productivity, and Agentforce's own consumption pricing concedes the seat model's mortality. The Jevons case cuts cleanly in silicon's favor: the same efficiency shock that compresses the app layer's AI premium expands token volume for compute owners. Memory is a contracted three-player oligopoly mid-squeeze, and our silicon fair values were struck at the hawkish bottom-tick of a $3.3T drawdown — if the July 28–29 FOMC softens the rate path, those marks are too low, not too high. And the sharpest internal tell: our own framework put NVIDIA and Synopsys at fair value, which means the honest version of "avoid silicon" is "avoid the extended silicon" — the momentum cohort, not the cash generators. The independent risk audit puts the probability the house stance is wrong at 43%, driven mostly by the silicon-avoid leg fighting a live memory cycle. That number is why the long book is one name and not four.

---

Catalysts & Risks

Falsification. The long leg is wrong if CRM's Agentforce decelerates below ~150% YoY or discloses seat contraction on 9/2, if ADBE's AI-first ARR stalls below a ~$700M run-rate or Creative Cloud net-adds go negative on 9/10, or if consensus FCF revisions for both turn negative through the prints. The avoid leg is wrong if the SOX takes out its June 22 high on rising forward estimates, if the three memory prints (7/29, 7/30, ~9/22) beat-and-raise with HBM pricing stepping up, or if the hardware-over-software divergence widens and holds for six-plus weeks instead of mean-reverting — flow reverses in days; regimes persist. Two trips in either column force the corresponding leg down.

---

References

• Morgan Stanley software reset — CRM to Equal-Weight $185, ADBE to Underweight $240 (via Yahoo Finance quote pages, 2026-07-21) · ADBE
• NVIDIA 9.3% passive stake in Nebius (filed 2026-07-21) · Cerebras–CrowdStrike inference partnership (2026-07-22)
• Alphabet Q2 consensus and 7/22 date · Salesforce 9/2 print · Adobe 9/10 print · FOMC 2026 calendar
• Salesforce Q1 FY27 — Agentforce ARR $1.2B +205% (2026-05-27) · Adobe Q2 FY26 — AI ARR tripling (2026-06-12)
• Moonshot Kimi K3 (Reuters, 2026-07-17) · The $3.3T chip drawdown (Yahoo, 2026-07-17) · Warsh hawkish debut (CNBC, 2026-06-23)
• Companion house reports: AI Compute Silicon sector (2026-07-21) · AI Value Chain software (2026-07-12) · Cerebras company (2026-07-20) · Korea AI memory (2026-07-13)