Cadence Design Systems

Refresh delta (since 2026-06-23)

What changed

1. Q2 2026 printed strong and the guide was raised hard. Revenue $1,584.5M, +24.2% YoY; GAAP diluted EPS $1.33; non-GAAP EPS $2.11; record backlog $8.1B; FY2026 guide lifted from $6.125–6.225B / $7.85–7.95 non-GAAP EPS to $6.260–6.340B / $8.05–8.15 — what CEO Devgan called "the highest raise we ever had".

2. An open-weight model designed a chip without them, and the tape noticed. On 2026-07-17 Moonshot AI released Kimi K3, a 2.8-trillion-parameter open-weight model that autonomously ran a full chip-design flow — architecture through verification — on open-source EDA tools only, closing timing at 100 MHz on a 4 mm² block using the Nangate 45nm library. CDNS fell ~9.5% that day; a six-day slide erased ~$15B of market value and took the stock −21.2% from its June 1 close.

3. The de-rate happened. ~$387 (2026-06-18) → $338.20 (2026-08-10). That is −12.6% from the June dossier's price and −18.8% from the $416.39 all-time high. Trailing 52-week price change is now −3.78%; market capMarket capitalisationThe share price multiplied by the number of shares. What the market says the equity is worth. $93.1B vs the June dossier's ~$105B estimate.

4. Intel Foundry — a multi-year 14A engagement, the first real crack in a twenty-year weak spot. Cadence signed a multi-year agreement covering DTCO on Intel 14A, agentic EDA flows and design IP, plus deepened Samsung Foundry (2nm/3D-IC) and a new Rapidus agentic-AI integration. Devgan: "Intel is a company we tried to work closer for a very long time… this is like a 10- or 20-year-old problem".

5. China nearly doubled — the fragile leg got bigger, not smaller. $236.2M, +96% YoY, 15% of revenue (from 9% a year ago); H1 $425.6M, +64%. Part of this is a base effect — Q2 2025 was suppressed by the May–July 2025 BIS EDA licence requirement, which the filing states explicitly — but the exposure the bear case rests on is now materially larger.

6. A new, dated regulatory cliff. BIS's 2026-09-29 interim final rule extending Entity-List restrictions to 50%-or-more-owned affiliates was suspended on 2025-11-11 for one year — expiring 2026-11-09 absent extension. The FY guide explicitly assumes controls "remain substantially similar for the remainder of the year".

What held

The structural thesis stands, and one leg of it strengthened. The duopoly toll-booth is intact: 84.9% gross margin, no customer ≥10% of revenue, 78% recurring, ~$8.1B contracted-but-unsatisfied performance obligations, all three product categories growing double-digit. The moat widened competitively — Intel, Samsung and Rapidus all deepened this quarter, at exactly the accounts where Cadence was historically weakest.

The June dossier's contrarian call was right, and right early. Lens 12 of the prior dive named the non-obvious risk explicitly: "if AI super-agents genuinely collapse the number of engineering hours needed to design a chip, the long-run per-design seat/license base could shrink." Four weeks later that exact scenario got a public demonstration and cost the stock $15B. The thesis did not break — but the market has now started pricing the risk the prior dive said "almost no one is underwriting."

What did not hold: the balance-sheet cleanliness read. The June dive said inventory (+4.8%) growing far below revenue was evidence of no channel stuffing. Inventory is now +28.6% YTD against +21% H1 revenue. Management's explanation is credible and favourable — hardware is supply constrained, "building systems as quickly as we can to deliver against the backlog" — but the metric flipped sign and it should be watched, not waved through (Lens 10).

Phase A — Understand the business

Company Overview — carried

Unchanged in substance; see the previous dossier. One update: the Hexagon D&E purchase price finalised at $3,101.0M total consideration ($2.2B cash + 3.2M shares worth $902.2M), net $2.9B of cash acquired, allocated to $2,147.2M goodwill + $1,248.0M acquired intangibles. Revenue mix is now 78% recurring / 22% up-front, and management guides ~80/20 for the full year.

Supply Chain — unchanged, see the previous dossier

One material addendum that belongs here rather than a re-run: the hardware line is now supply-constrained, not demand-constrained. CFO Wall: "it still remains supply constrained by customer demand rather than demand constrained. And we're building the systems as quickly as we can to deliver against the backlog". That is the direct cause of the inventory build in Lens 10.

Competitive Advantages — REFRESHED (the new material hits this lens hardest)

Two opposing moves landed in the same six weeks.

The moat widened, commercially. The foundry tie — the thing that makes an EDA vendor structurally hard to displace — got materially stronger in one quarter:

• Intel Foundry, multi-year, 14A DTCO + agentic EDA + IP. Devgan framed Intel and Samsung as the two places Cadence was weak, and said the trajectory "has definitely changed".
• Samsung Foundry, deepened on 2nm and 3D-IC.
• Rapidus, InnoStack super-agent embedded into its Raads design solution.
• Cadence now sits inside TSMC + Intel + Samsung + Rapidus flows. A design-tool vendor certified across every leading-edge foundry is close to un-routable-around at advanced nodes.

The moat was publicly challenged, technically. Kimi K3 is the first credible demonstration that an agentic model plus free tooling can traverse the whole flow unaided. Management's rebuttal is specific and, on the facts available, fair: the block was ~4 mm², on a 45nm library (circa 2008), at a clock 20–30x below current frequency, and it still needed EDA tools to do it — they were just open-source ones. Devgan: "to really do real designs, people use Cadence."

The honest read on the moat: the demonstration does not threaten the 3nm/2nm signoff business where the money is. It threatens something subtler — the narrative that EDA is un-substitutable, which is what the multiple is paying for. And it hands a state-backed Chinese EDA cohort (Empyrean, X-EPIC, Primarius) a proof-of-concept template for the trailing-edge and emerging-market design work that is exactly where domestic substitution starts. The moat at the leading edge is intact and arguably wider. The moat at the perimeter is now demonstrably permeable, and the perimeter is where share loss begins.

Pricing power / value capture: management's answer is the "Three-Layer Cake" — compute and data at the bottom, physically-accurate solvers in the middle, AI agents and orchestration on top — with the claim that agentic AI is a "demand accelerator" because agents invoke the middle-layer engines more often, not less. That is a coherent theory of value capture. It is still a theory: no consumption-pricing model has been disclosed, and agentic revenue is not in the guide.

Segments — REFRESHED (mix moved)

Cadence remains one reportable operating segment; product categories are revenue-mix disclosure only, with no segment operating income.

By product category, Q2:

Derived $ and derived YoY are — the filing discloses whole-percent shares only, so the arithmetic carries ±1pt rounding error. Management's own growth rates are and should be preferred where they differ.

Read: Core EDA's share fell 71%→68% for the second consecutive period. That is not Core EDA weakening (+18–19% is a fine number) — it is IP and SD&A outgrowing it, the first genuinely three-engine quarter this company has printed. SD&A includes the acquired Hexagon D&E business, so its 37% is inorganically flattered; IP's 40%+ is organic and management volunteered a caution against annualising it: "IP revenue can be timing dependent from quarter-to-quarter… I wouldn't annualize any one quarter". Recurring revenue grew ~24%, of which roughly 4 points was Hexagon — ex-acquisition recurring growth is "high teens to 20%" on management's own normalisation.

By geography, Q2 2026 vs Q2 2025:

The single most consequential line in this table: US share fell 46%→42% while China rose 9%→15%. The revenue base is migrating toward the geography with a dated regulatory cliff on it.

Phase B — Measure performance

Earnings Result — REFRESHED (Q2 2026, reported 2026-07-27)

The print [all research-layer: filings/10-q-2026-q2.md unless noted]:

Versus consensus — and the outlets disagree. Quiver Quantitative reports EPS $2.11 vs $2.10 estimate (a $0.01 beat) and revenue $1,584.5M against a $1,608.5M estimate — i.e. a revenue miss. Investing.com reports the same EPS beating a $2.05 consensus by $0.06. Zacks/Yahoo characterise it as surpassing both EPS and revenue estimates. These are not reconcilable from public sources and I am not resolving them. The defensible statement: EPS beat on every version; the revenue line was at or slightly below the highest published estimate, and management's own metric — "all key metrics exceeding our guidance" — refers to guidance, not Street consensus.

Guidance — raised, twice over:

Guidance carries an explicit caveat: it "contains the useful assumption that export control regulations that exist today remain substantially similar for the remainder of the year." Buyback intent stated for the first time as a formula: ~50% of free cash flow in 2026 (H1 tracked at 45% — $400.0M returned against $889.3M FCFFree cash flowCash left after paying to run and maintain the business. Unlike profit, it is hard to flatter with accounting choices. ).

Note the Q3 sequential margin step-down — non-GAAP 45.5% actual → 43.5–44.5% guided. Management attributes it to $20–25M of H2 investment in the Hexagon D&E integration and the Intel engagement, with margin recovery promised for 2027. This is the first guided sequential margin decline of this cycle and it is spend-driven, not demand-driven.

Backlog / RPO. $8.1B contracted-but-unsatisfied at 2026-06-30 (vs $8.0B at Q1, $7.8B at YE2025), including $0.9B of non-cancelable commitments; 58% recognised within 12 months, 40% in months 13–36; current RPORemaining performance obligationsRevenue a company has already signed contracts for but has not yet delivered or booked. ~$4.2B. Backlog grew ~1% QoQ while revenue grew 24% YoY — which looks like deceleration until you take management's framing: 2026 is a low year in the three-year renewal cycle, H1 bookings were up ~55% YoY (an analyst's uncontested figure on the call), and "normally, first half, we draw down on our backlog, but this year has been good growth". Both readings are defensible; I am flagging the metric, not calling it.

Balance-sheet flags:

• Cash $1,440.4M (up from $1,406.7M at Q1). The $425M revolver draw is fully repaid — $600M drawn and $600M repaid across H1. Net debt fell to $1,041.8M from $1,499.5M at Q1.
• Receivables (current) $1,065.0M, +12.7% vs YE2025 — comfortably below revenue growth. DSO 65 days. But unbilled AR is $564.6M vs $456.0M (+23.8%) while billed AR grew only 2.4% ($504.4M vs $492.8M). Unbilled is now ~50% of gross receivables. Normal for ratable software; worth a watch line.
• Inventory $390.4M, +28.6% YTD — now growing faster than revenue, reversing the June read. Raw materials $312.3M (+27.2%). H1 cash flow shows $107.0M consumed by inventory. Management's cause: supply-constrained hardware ramp.
• Deferred revenue (current) $1,025.1M vs $778.4M, +31.7% — a genuinely good forward signal.
• Goodwill $4,914.8M + acquired intangibles net $1,874.5M = $6,789.2M, which is 56.2% of total assets and 99.0% of shareholders' equity. See Lens 10.

Market reaction: +4.37% on the release. But that bounce came off a stock already ~21% below its June 1 close with RSI near 30. A beat-and-raise of this size recovering only 4% is itself information: the market is no longer paying for the earnings, it is repricing the multiple.

Earnings Calls — REFRESHED (first transcript ever on the shelf)

The Q2 2026 transcript is now ingested at transcripts/2026-q2.md [source: AlphaStreet, 2026-07-27]. The June dive had zero transcripts and had to source sentiment entirely from. Sentiment trend vs the prior three-to-four calls:

• Confidence: up, and now numerate. Q1 2026's agentic story was a product reveal with no metrics. Q2 put numbers on it for the first time: ChipStack AI Super Agent — 20+ customer engagements, "40x faster RTL validation"; ViraStack — 25+ engagements, "2x to 10x productivity improvements". That is a real disclosure step-up.
• New defensive posture — the genuine tone shift. Three separate analysts asked about open-source / LLM-driven chip design. Devgan answered each time with the same frame and one memorable line: "the ground truth will prevail." He also reframed the threat as a confirmation: "the recent news just confirms that framework" — arguing agentic AI increases tool invocations. Management has never had to defend the existence of the category before. It does now.
• New vocabulary: "Three-Layer Cake" (now the organising metaphor of every strategic answer), "demand accelerator", "ground truth". "Rule of 60" — the Q1 headline phrase — did not recur in the prepared remarks.
• What they still will not quantify. Agentic-AI TAM ("it opens up a new opportunity" and no further), agentic monetisation timing beyond the current add-on business, and China revenue exposure — which they did not break out despite it nearly doubling. The add-on business is characterised as "encouraging" but "still in early days," ~6 months post-launch.
• New candour markers, both worth crediting. Wall volunteered that 2026 is a low renewal year; Devgan volunteered "I wouldn't annualize any one quarter" on the fastest-growing line (IP); and management pre-announced an H2 margin dip rather than letting Q3 surprise. Companies inflating a narrative do not usually hand you the caveats.
• Highest emotional charge on the call: Intel. "I'm very proud of this new Intel collaboration… this is like a 10- or 20-year-old problem."

Net sentiment: more confident and more specific than any prior call, with one new defensive front. The credibility positive from June holds — they raised hard without putting un-monetised agentic revenue in the number.

Comps — CDNS column refreshed; peer columns carried

CDNS row. Synopsys row carried unchanged from the June dossier — stale by ~7 weeks and not re-verified this run; do not trade the spread off it.

Cross-check on the CDNS row (this is the useful part). Three independent reconciliations between my primary-filing extraction and the quoted market data all tie:

• LTM revenue from our figures = $5.838B; $94.26B EVEnterprise valueWhat it would cost to buy the whole company: its market value plus its debt, minus the cash you would get with it. Often very different from market cap. / $5.838B = 16.15x vs quoted 16.20x ✓
• LTM GAAP net income = $1.378B; $93.14B / $1.378B = 67.6x vs quoted trailing P/E 67.44 ✓
• LTM FCF from our figures = $1.679B vs the quoted $1.68B ✓

The extraction is sound. Note the 41.8x figure used elsewhere in this dossier is $338.20 ÷ FY2026 guided non-GAAP EPS $8.10; the quoted "forward P/E 39.52" implies ~$8.56 of EPS and is therefore an NTM or FY2027-blended figure, not the company's FY2026 guide. Both are stated so the reader can see which is which.

Stock-Price Catalysts — REFRESHED (the pattern itself changed)

Moves >5% since the 2026-06-23 boundary:

Price path: $416.39 ATH (2026-06-02) → ~$387 (2026-06-18, June dossier price) → the July slide → $338.20 (2026-08-10). Trailing 52-week price change −3.78% — a name that was +45% YoY at the June dive is now flat-to-down over a year.

The structural change to the catalyst list. The June dossier named four drivers: bookings/backlog + guidance raises; AI-narrative beats; China/export headlines; AI-capex sector beta. A fifth now sits at the top: credible public demonstrations that AI reduces the need for licensed EDA. That is a new category of downside catalyst — it is not earnings-sensitive, not guidable, not defensible by a good quarter, and it can recur at the cadence of frontier-model releases (which is roughly monthly). Note that a 24%-growth beat-and-raise recovered only 4.37% of a 21% drawdown: the earnings channel is now a weaker lever on this stock than the narrative channel. That asymmetry is the single most important thing this refresh found in the tape.

Dated forward catalyst added: BIS 50%-affiliate rule suspension expires 2026-11-09.

Phase C — Judge people & books

Management — REFRESHED (light; no personnel change)

CEO Anirudh Devgan and CFO John M. Wall both in seat; Wall signed the 2026-07-27 8-KForm 8-KA filing for something that happened between reports and matters enough to tell shareholders now.. No management change, no departure, no governance event surfaced in the window.

Capital allocation — now stated as a rule. For the first time management committed to a formula: ~50% of free cash flow to buybacks in 2026. Execution: Q2 580k shares for $200.0M; H1 1,251k shares for $400.0M (vs 1,968k / $525.0M in H1 2025 — fewer dollars deployed at a lower price, worth noting). Authorization remaining fell to $1.0B from $1.4B at YE2025. No dividend.

M&A engine: paused. No new acquisition announced since Hexagon closed 2026-02-23. Transaction costs fell to $2.8M in Q2 from $4.0M a year prior — consistent with a digestion year, and consistent with the DOJ-probation constraint the June dive flagged.

One judgment call to credit: management raised the guide by the largest increment in company history in the same quarter the stock fell 21% on an existential-sounding narrative — and still did not put un-monetised agentic revenue into the number to defend the story. They also volunteered three separate caveats (low renewal year, don't annualise IP, H2 margin dip). That is disciplined behaviour under narrative pressure and it is the strongest evidence in this refresh that the guide is real.

Forensic Red Flags — REFRESHED

Every figure labeled. The books remain clean; three items changed grade since June.

1. Net income quality — the headline growth is doubly flattered. (NEW, and the most useful finding here.) Q2 GAAP net income "+129%" is not an operating result. Two distinct distortions:

• Q2 2025 carried the $128.5M BIS/DOJ contingent-liability charge and a 43.0% effective tax rate.
• Q2 2026 carried a $71.9M gain on investments inside other income — 14.1% of pre-tax income, non-operating and non-recurring.

Strip both: pre-tax income ex-investment-gains was $438.3M vs $242.2M. Still a genuinely strong +81%, but score the operating line (+24% revenue, 28.4% GAAP op margin), not the net-income line. Also note interest income collapsed to $9.9M from $26.0M (−62%) as the Hexagon cash left the balance sheet — a real, recurring ~$16M/quarter headwind that the investment gain happens to mask this quarter.

2. Goodwill and intangibles — grade raised from "rising risk" to the primary balance-sheet exposure. Goodwill $4,914.8M (from $2,749.1M at YE2025) plus acquired intangibles net $1,874.5M (from $718.2M) = $6,789.2M, equal to 56.2% of total assets and 99.0% of shareholders' equity. Total amortization of acquired intangibles ran $79.9M in Q2 vs $23.7M a year ago (3.4x), and the filing schedules $158.1M for the remainder of 2026, $294.8M in 2027, $287.5M in 2028. This is a permanent, disclosed, multi-year GAAP earnings drag and the mechanical reason the GAAP/non-GAAP gap widened to $3.29/share on the FY guide. If SD&A growth disappoints, the impairment exposure is now roughly the size of the equity base.

3. Inventory — flag flipped, then explained. $390.4M, +28.6% YTD vs +21% H1 revenue — inverting the June reading. Raw materials $312.3M (+27.2%); finished goods $74.9M (+72.5%); $107.0M of H1 operating cash consumed. Management's cause is supply-constrained hardware demand, not soft sell-through, which is the benign explanation and is corroborated by hardware being called out as a driver of the guide raise. But the FY2025 10-KForm 10-KA company’s audited annual report to the US regulator. The most complete thing it publishes. disclosed prior write-downs for "excess and obsolete inventory related to previous generations of our hardware products" — this company has taken that charge before. Watch, do not dismiss.

4. Revenue recognition — no deterioration. 78% recurring / 22% up-front, essentially flat YoY (78/22 vs 78/22 in Q2 2025; H1 78/22 vs 80/20). Deferred revenue +31.7%. Current receivables growing well below revenue. Unbilled AR +23.8% vs billed AR +2.4% is the one line that moved and it is a watch item, not a finding — unbilled receivables are the normal consequence of ratable recognition running ahead of invoicing.

5. SBC — unchanged, still material. $146.9M = 9.3% of revenue (identical ratio to Q2 2025); unrecognized $970.8M over 2.2 years. The non-GAAP EPS the market quotes ($8.10 guide) overstates economic earnings versus the GAAP guide ($4.81) by 68%.

Regulatory findings (required, re-run):

• SEC EDGAR EFTS (LR + AAER), window 2021-08-10 → 2026-08-10: zero findings. Accounting-enforcement history remains clean.
• DOJ/BIS export-control settlement — no change, no new development. Note 14 restates the July 2025 guilty plea to one count of conspiracy to commit export-controls violations, the $140.6M aggregate net penalties paid in Q3 2025, and the ongoing audit/compliance obligations. No new charge, no new accrual, no probation violation disclosed.
• NEW — a dated regulatory cliff. BIS's interim final rule effective 2025-09-29 extended Entity-List/MEU restrictions to entities 50%-or-more owned by listed parties. BIS published a one-year suspension on 2025-11-11, set to expire 2026-11-09 absent extension. Cadence's own language: "We expect the impact… to be limited, but we will continue to monitor." Management's FY guide assumes no change. This is the first hard date the bear case has ever had.
• Tariffs: the U.S. Supreme Court struck down IEEPA-authorised tariffs in February 2026; tariffs under other statutes remain and new measures are being pursued. Cadence flags direct exposure via hardware component imports.
• No other agency action surfaced.

Bottom line on Lens 10: revenue recognition and cash conversion remain clean. What changed is that the balance sheet now carries acquisition risk roughly equal to the equity base, the income statement's headline growth is flattered by a one-off investment gain against a prior-year charge, and inventory flipped from a comfort to a watch item. None of these is a thesis-breaker. Together they mean the GAAP line deserves more weight than it did in June.

Phase D — Project & stress-test

Forward Projection (FY2026 → FY2028) — REFRESHED

Model status — read this before any number below. model.xlsx was rebuilt this run from the newly-populated our figures (8 quarters, 2024-Q3 → 2026-Q2; opening balance sheet sourced: yes). It has no computed values — our model reports 51 formulas with no cached results, so every output cell (value per share, EV, FY+3 EPS, balance check) reads —. I am therefore citing no model output. Three seeded assumptions would also need overriding before its outputs could be trusted: share price seeded at 0, tax rate seeded at 15% (Cadence's H1 2026 effective rate was 25.1% ), unlevered beta seeded at 1.0, and depreciation seeded at 50% of opening net PP&E — a distortion here, because most of Cadence's D&A is acquired-intangible amortization, not PP&E depreciation. Everything below is with arithmetic shown.

Anchor: FY2026 guide, non-GAAP EPS $8.05–8.15, midpoint $8.10; GAAP EPS $4.76–4.86, midpoint $4.81; revenue $6.260–6.340B.

A detail the GAAP line exposes: H1 2026 GAAP diluted EPS was $2.56; the FY guide midpoint of $4.81 implies ~$2.25 of H2 GAAP EPS — a sequential step-down, driven by the $20–25M H2 investment and the rising intangible amortization. The non-GAAP line hides this entirely.

Buyback drag assumption for all three cases: FY2026 op cash flow ~$2.0B less ~$0.20B capexCapital expenditureMoney spent on long-lived things — buildings, machines, servers — rather than on running costs. ≈ $1.8B FCF; 50% deployed = $0.9B; at $338 that retires ~2.66M shares ≈ 0.97% of the 275.4M outstanding, largely offset by SBC issuance → ~0.5% net annual share reduction.

Cross-check vs Street: consensus PEG of 2.57 against a 39.52x forward P/E implies ~15.4% expected EPS growth — the base case sits just above it, which is the right place for a base case to sit.

Valuation sanity at $338.20 [all estimate: arithmetic shown]:

Asymmetry — materially better than June, still not good:

• De-rate to 30x on base FY2027 $9.48 → $284 (−16%)
• De-rate to 30x on bear FY2027 $8.77 → $263 (−22%)
• Hold 36x on bull FY2027 $9.94 → $358 (+6%)
• Hold 40x on bull FY2027 $9.94 → $398 (+18%)
• Entry at 32x base FY2027 → $303, i.e. 11% below spot (was ~25% below in June)

Street average PT is quoted at $403.67 across 25 analysts and separately at $375 across 14 analysts — a conflict I am surfacing, not resolving; the coverage universes differ. BofA raised to $400 and KeyBanc to $425 post-print.

(No Brier forecast logged — unattended-run rules. If promoted to a position, the natural falsifier is: "CDNS FY2026 non-GAAP EPS ≥ $8.05," resolving 2027-02.)

Bull vs Bear — REFRESHED

Bull case (strengthened this quarter). The toll-booth is collecting more, from more places, than at any point in its history: +24% revenue with all three product categories growing double-digit, IP +40%, SD&A +37%, hardware supply-constrained rather than demand-constrained, record $8.1B backlog in what management says is a low renewal year, and 45.5% non-GAAP operating margin. On top of that, two things happened in six weeks that the June dive could not have priced: Intel Foundry signed a multi-year 14A engagement that closes a two-decade competitive gap, and Samsung and Rapidus deepened alongside it — Cadence is now embedded in every leading-edge foundry flow on earth. And agentic AI finally has numbers attached: 45+ customer engagements across two super-agents, with 2x–40x cited productivity gains. If Devgan's "demand accelerator" thesis is right — agents invoke the physically-accurate engines more, not less — then the July selloff bought a wider moat at a lower price.

Bear case (a real catalyst now exists, with a date).

1. The Kimi K3 problem is not that it was right — it is that it is repeatable. A frontier open-weight model, released roughly monthly by somebody, can now generate a headline that costs this stock 9% in a session. The demo was 45nm and trivial, but the narrative channel has proven ~5x more powerful on this tape than a record beat-and-raise (−9.5% on a demo vs +4.4% on the print). You cannot guide your way out of that.
2. China is now 15% of revenue and the BIS 50%-affiliate rule un-suspends 2026-11-09. The guide explicitly assumes no change. The mechanism has already fired once (May–July 2025) and measurably suppressed China revenue — that is precisely why the +96% YoY comp is so flattering.
3. Valuation still has no cushion. 41.8x FY2026 non-GAAP, 70x FY2026 GAAP. The de-rate has run 19% from the high and has plenty of room left before it reaches anything resembling a floor.
4. $6.79B of goodwill + intangibles ≈ 99% of equity, with $294.8M/yr of scheduled amortization from 2027, in a business that just paused its M&A engine and must now organically prove SD&A.

Pre-mortem (18 months out, thesis broke). The most likely failure mode has changed shape since June. In June it was a passive de-rate on a cooling AI-capex narrative. Now it is active and two-handed: (a) a second, better open-model chip-design demonstration — this time at a modern node, or from a Chinese domestic EDA stack — resets the terminal-value question the market has never had to ask about EDA; (b) the BIS rule un-suspends in November, China steps from 15% back toward 9%, and the 19% growth guide becomes a low-teens reality. Either alone is a 20% de-rate on intact earnings. Together, at 42x, it is 35%+. The franchise still survives in every branch — this remains a multiple risk, not a business risk.

Contrarian view — updated, and it inverts the June one. In June the contrarian point was "the market treats agentic AI as pure upside; the reverse risk is un-underwritten." That trade is now largely done — the market spent July underwriting it, violently. The new contrarian position is the mirror: the market is now over-extrapolating a 45nm toy block into a terminal-value threat, while systematically under-weighting the thing that actually happened in the same six weeks — Cadence closing the Intel gap it had failed to close for twenty years. One of those is a demo; the other is a multi-year contract at the second-largest leading-edge foundry. The tape paid ~$15B of attention to the first and roughly none to the second. If the Three-Layer Cake thesis holds, July was a narrative dislocation, not a re-rating. If it doesn't, July was the first honest quote.

Devil's Advocate — REFRESHED

Dismantling the refreshed bull case.

• The +24% is not as clean as it reads. Roughly 4 points of recurring growth was Hexagon — bought, not earned. SD&A's +37% is substantially inorganic for the same reason. China's +96% laps a quarter that BIS deliberately suppressed — the filing says so outright. IP's +40%+ is the one line management itself told you not to annualise. Strip the acquisition, the regulatory base effect, and the timing-dependent line, and the organic core is high-teens — good, not the 24% headline. And the headline net income +129% is flattered by a prior-year charge on one side and a $71.9M investment gain on the other (Lens 10).
• R&D intensity fell 115bps (34.7% → 33.5% of revenue) in the exact quarter the market decided AI might commoditise your product. Some of that is operating leverage. Some of it, in the bull's own framing, is under-investing in the top layer of the cake while an open-weight model shipped the demo you should have shipped.
• Where the moat is genuinely weaker than bulls think. The Kimi rebuttal — 45nm, 4mm², 20-30x slow — is correct and beside the point. The threat vector was never leading-edge signoff; it is that the trailing edge and the emerging-market design tier now have a demonstrated free path, and that is exactly the tier where Huada Empyrean, X-EPIC, Primarius, Xpeedic, Univista and Giga Design Automation compete. US export policy is handing that cohort a captive market and Moonshot just handed them a reference architecture. The two combine.
• Most dangerous competitor bulls underestimate: unchanged from June — the state-backed Chinese EDA cohort, not Synopsys. But the threat upgraded this quarter: it now has an open-weight agentic layer to build on.
• The Intel win cuts both ways. Management said outright: "we will also invest more, right, in Intel and Intel customers, which is to be expected" — and $20–25M of H2 margin is already committed to it. A landmark customer that requires you to spend to serve, at a foundry whose own volume ramp is unproven, is optionality priced as certainty.
• Assumptions that must hold for $338.20: ~15–17% EPS growth sustained for three years, the ~42x FY2026 multiple not compressing, China staying at 15% through a November regulatory expiry, agentic monetisation eventually appearing as revenue, Hexagon's $2.1B of goodwill proving out, and no better open-model chip-design demonstration. That is still a lot of "ands" — though it is a shorter list than June's, because the market has already taken 19% out.
• If growth disappoints 20–30%: bear FY2027 ~$8.77 at a 30x de-rate → ~$263, a 22% drawdown.
• The single scenario that permanently impairs: unchanged in kind, upgraded in probability — a structural EDA-to-China export ban that both zeroes a now-15% revenue leg and accelerates a domestic Chinese EDA stack that, post-Kimi, has a credible open-source agentic foundation to build on and can then compete on price in every emerging market. June called this "low-to-moderate but rising." The 2026-11-09 expiry date and the Kimi demonstration together move it to moderate.

Management Questions — unchanged, see the previous dossier

The fifteen questions there survive intact; Q2 partially answered #1 (agentic engagement counts, still no monetisation model), #4 (Hexagon "progressing well," no milestones given) and #8 (winning at Intel and Samsung). Three to add: 16. Kimi K3 was 45nm and trivial. At what node, and at what design complexity, would an open-source-plus-agent flow start taking real revenue from you — and what is your internal tripwire for that? 17. China went from 9% to 15% of revenue while the BIS 50%-affiliate rule sits suspended until 2026-11-09. What is the revenue at risk if it un-suspends, and why have you never broken out China exposure quantitatively? 18. R&D intensity fell 115bps this quarter. Is the top layer of the Three-Layer Cake — agents and orchestration — getting more absolute R&D dollars than a year ago, or is operating leverage coming out of the exact layer under competitive attack?