• "Broadcom (AVGO)" = a ~$1.77T semiconductor-and-software conglomerate built by serial acquisition under CEO Hock Tan. Roots in the HP/Agilent semiconductor division; IPO’d as Avago Technologies (Aug 2009), bought Broadcom Corp in 2016 and took its name, then CA Technologies (2018), Symantec enterprise (2019) and VMware (2023, ~$69B). Today it reports its financials in two segments — Semiconductor Solutions and Infrastructure Software (both detailed below).
• "Semiconductor Solutions" (~58% of FY25 revenue) = custom AI accelerators (XPUs), AI/data-center networking (switches, optics), plus wireless (Apple), broadband and storage chips. This is the AI-growth engine.
• "Infrastructure Software" (~42%) = VMware Cloud Foundation (VCF), mainframe and enterprise software (CA, Symantec). Grew +26% in FY25 on the VMware subscription conversion — now maturing (+9% in Q2 FY26).
• "XPU / custom silicon" = application-specific AI chips Broadcom co-designs for a single customer (vs NVIDIA’s merchant GPUs). Reported customers (public confirmation varies by name): Google (TPU) and Apple publicly linked to Broadcom; Meta (MTIA), Anthropic, OpenAI and ByteDance per trade press and earnings-call commentary — with multi-gigawatt 2027 commitments cited by trade press for selected programs (not the full roster).
• "AI semiconductor revenue" = Broadcom’s headline KPI — a management product-category disclosure within Semiconductor Solutions, not a GAAP segment line: $10.8B in Q2 FY26 (+143% YoY, 49% of total). Management guides ~$56B for FY26 (guide) and points to >$100B for FY27 (a roadmap-based target, not contracted backlog) — arguably the single most important number in the bull case, and the one Tan did NOT raise on Jun 3.
• "Non-GAAP vs GAAP EPS" = the gap matters here. GAAP EPS ($1.91 in Q2) is weighed down by years of VMware-acquisition amortization; the Street values AVGO on non-GAAP EPS ($2.44). The ~62× "trailing P/E" is a GAAP artifact — forward non-GAAP is ~35×.
• "Net debt ~$45B" = the VMware deal was debt-funded; cash $19.6B against $64.9B total debt. Deleveraging is why FY25 buybacks ($2.45B) fell well below dividends ($11.1B).
• CEO: Hock E. Tan (President & CEO since Mar 2006 — the architect of the acquisition machine) · CFO: Kirsten Spears (transition to Amie Thuener announced Apr 2, effective Jun 12, 2026) · Chair: Henry Samueli (co-founder).

Balance Sheet — Plain English

The read: $45.28B net debt (cash $19.63B vs $64.91B total debt, $62.66B LT + $2.25B ST, Q2-FY26 close) ≈ 1.5× TTM EBIT — a real leverage position, not the net-cash posture of NVIDIA or AMD, and the single biggest structural gap to its mega-cap AI peers.

It is the bill for the ~$69B VMware deal (2023), down from a ~$74B at-close peak. The position is investment-grade and easily serviced from $33B of TTM free cash flow — coverage is no concern — but the paydown is why FY25 buybacks ($2.45B) ran a fifth of dividends ($11.14B), and it leaves AVGO more rate-sensitive than its net-cash peers.

So the AI-driven FCF surge (Q2 $10.3B) speeds the paydown but does not yet fund both a buyback step-up and another large cash deal without re-levering. The tell: whether Tan reaches for the next acquisition once VMware debt is digested, or pivots free cash back to buybacks.

Operating Segments — FY2025 10-K (XBRL) + Q2 FY26 (Jun 3 release)

Semiconductor end-markets — MD&A narrative (FY25, estimates)

Geography — FY2025 10-K narrative

Segment economics · price-volume-mix, sum-of-the-parts & pricing power

• Inflation: May CPI +4.2% YoY (Jun-10 print) — a three-year high; core +2.9%. Hot inflation repriced the rate-cut path OUT, and long-duration high-multiple tech (AVGO at ~35× forward) is the most rate-sensitive cohort.
• Rates: Fed funds 3.50–3.75% under Chair Warsh; FOMC meets Jun 16–17 with CME FedWatch pricing elevated hold/hike odds after the CPI. 10y near 4.6%, 30y past 5% — a direct headwind to every AI-growth multiple.
• AI-capex cycle: the central swing factor. Hyperscaler AI spending is still climbing (NVIDIA FY26 revenue $215.9B, +65%; TSMC guiding >30% growth), but the Jun-5 semiconductor sell-off — led by AVGO’s own cautious-read outlook — shows the market is newly alert to a spending air-pocket.
• Geopolitics / export control: China is ~17% of AVGO revenue (down from ~20%) under an escalating US export-control regime; the Iran conflict adds an oil/risk-off overlay that is pressuring growth tech specifically.
• Memory crisis: HBM/DRAM tightness and pricing volatility add a second axis of semiconductor risk into H2 — relevant to the AI-system bill-of-materials AVGO’s customers must fund.

At ~35× forward AVGO behaves like a long-duration asset (~30–35yr effective duration): a sustained +100bps in the 10-year compresses a ~35× multiple ~15–25%, far more than a 12–15× value name — and its ~2.0 market beta (Lens C) compounds it. So the Jun reaction is mostly a discount-rate move with fundamentals intact; the Synthesis carries the full Campbell-Shiller decomposition. The one asymmetry that matters here: an inflation/term-premium-driven rate rise is unambiguously negative, but a growth-driven rise can be partly offset by stronger AI demand.

The AI-accelerator market is in the steep middle of an investment supercycle: roughly half of all semiconductor revenue is now AI-linked, the AI-chip market is projected to grow ~26% in 2026, and TSMC warns advanced-node capacity may fall short for years. Within it, a structural split is hardening: NVIDIA merchant GPUs for general-purpose AI vs custom ASICs (XPUs) that hyperscalers co-design with Broadcom and Marvell to cut cost and power at scale. Broadcom is arguably the leading player in the custom-silicon lane — its reported roster (Google, Meta, Anthropic, OpenAI, Apple, ByteDance; confirmation varies by name) and >$30B of single-quarter AI bookings are among the clearest evidence that XPUs are taking share of incremental AI capex.

Capital-cycle read: capacity capital is flooding in on every front — NVIDIA pushing into AI PCs and networking, AMD taking server-CPU share and committing >$10B to Taiwan, the hyperscalers funding both merchant GPUs AND their own custom chips. That is classic late-stage capital-cycle behavior: extraordinary returns attracting extraordinary investment, which historically compresses future returns. Broadcom’s moat (incumbency on a customer’s XPU roadmap, advanced-packaging know-how, the networking-plus-compute bundle) is real and sticky — but the same hyperscalers funding the ramp have the scale and motive to in-source over time (Google’s TPU is already largely internal IP — though Broadcom can still participate in physical implementation, packaging, networking and supply-chain layers, so “internal IP” is not the same as “Broadcom-out”).

• FV is a range, not a point — upper ~$370 on the optimistic FY26 non-GAAP EPS $10.70, lower ~$339 on the defensible $9.80 → spot ~9% above the conservative fair value, “fair” only on the optimistic EPS.
• Why the multiple is the debate — FCF yield is only ~1.9% at ~$1.77T cap, so the equity does not screen cheap on delivered cash (Financials owns the $33B TTM FCF); at ~35× modeled FY26 EPS the price is an advance on the FY27 $100B plan, paid while software (+9%) cools.
• Conviction & catalyst — low-to-medium: the franchise de-risks long-term earnings power, not the near-term multiple. The catalyst that matters — a FY26 AI-target RAISE — is exactly what management withheld on Jun 3, so the asymmetry improves only on a deeper de-rate toward the 200-day / gap-fill or a confirmed raise at Q3.

AVGO is a mature cash-flow asset and a DCF is defensible, but with the equity debate hinging on one variable — what multiple does the AI growth deserve, and does FY26 AI get raised? — a scenario-weighted forward non-GAAP P/E on modeled FY26 EPS (~$10.70; built bottom-up from the released Q1/Q2 plus the Q3 guide, because public aggregator EPS figures were split-confused and inconsistent) is the more decision-relevant frame and is our valuation anchor in place of a DCF. The legs map to multiple regimes: 26× ≈ a market-multiple-plus AI-digestion de-rate; 44× ≈ the pre-print premium re-asserting on a FY27 $100B that proves credible. Reverse-Bayes check on the weights: for the prob-weighted value to clear spot the bull leg would need ≳30% weight (vs the 25% assigned) — defensible only if a FY26 AI raise at Q3 is more-likely-than-not, which management explicitly withheld on Jun 3; the 30/45/25 split therefore deliberately under-weights the leg the catalyst has not yet confirmed.

All three legs price the same modeled FY26 non-GAAP EPS (~$10.70); the spread is the multiple only. The EPS itself is the model’s key fragility — it embeds a Q4 revenue estimate (~$31.5B, modeled) beyond the Q3 guide, so treat $10.70 as a center-of-range, not a point. A higher FY26 EPS (if AI is raised) shifts every leg up proportionally.

FY25 deployment: dividends $11.14B + buybacks $2.45B = $13.6B returned (~51% of $26.9B FY25 FCF, computed), with the rest going to VMware debt paydown. The skew is the story: buybacks collapsed from $7.18B (FY24) to $2.45B (FY25) as cash was redirected to deleveraging — gross debt is down from the deal’s ~$74B peak to ~$65B today (net debt $45B). SBC is material and rising ($7.57B FY25, ~12% of revenue), which is why non-GAAP EPS flatters GAAP and warrants scrutiny. The dividend ($0.65/q, 0.70% yield) has a long growth record but is a token yield at this price — AVGO is a capital-appreciation name, not an income name. The FY26 allocation question: as FCF scales past $30B+, does Tan accelerate buybacks, chase another acquisition (his historical pattern), or keep deleveraging?

Last 12 months (SEC Form 4, ≥$100k): open-market purchases of just $1.55M (4 transactions, all by director Harry You — a mild positive signal) against $512M of discretionary (non-10b5-1) sales across 184 transactions (President-ISG Ram Velaga, CFO Kirsten Spears, others) plus $630M of planned 10b5-1 sales (largely co-founder Henry Samueli — compensation/diversification noise). Net discretionary −$511M. Read: routine heavy executive selling into the run to $495 — normal for a high-SBC mega-cap, but the near-total absence of insider buying through the post-earnings drawdown means nobody inside is signaling the $372 level is a gift.

See Financials for the detail. The value-lens read: net debt $45.28B ≈ 1.5× TTM EBIT is investment-grade and serviceable, but it gates the buyback and makes the equity more rate-sensitive than net-cash peers (NVDA, AMD) — supporting dividend + deleveraging, not a buyback surge or a large cash acquisition without re-levering.

A.3 is an inside view — this name’s AI story, weighted 30/45/25. Below it is tested against the outside view (reference-class base rate first, story second; Kahneman & Lovallo 1993) and against a valuation method A.3 does not use. Five checks; each tightens the HOLD/WAIT but moves no fair value.

• Single-name prior — the reference-class base rate is humbling: the majority of individual stocks underperform T-bills over long horizons, with aggregate gains concentrated in a thin minority of names (Bessembinder 2018; a portfolio base rate, not a single-name forecast). AVGO is a plausible member of that minority — so the discipline falls on the bull leg: the case for paying up must clear the skew bar, not assume past it.
• Optimism haircut — forecast optimism is the norm, so management/Street FY27 targets warrant a consensus-optimism haircut (a base-rate caveat, not a claim about this guide); the >$100B FY27 roadmap is the line most exposed to it (the Evidence Ladder ranks it the softest rung).
• Multiple-expansion prior — a mega-cap already at ~35× forward non-GAAP has a low base rate of further multiple expansion; reversion toward the median is the default frame absent documented moat-persistence (Mauboussin et al. 2016, industry-median pattern, illustrative).
• Net effect — the prior argues for shrinking the 25% bull leg and treating the 35× base leg as the anchor (spot ≈ base, −1%).
• EPV / RIV floor — a Greenwald earnings-power-value (EPV = normalized no-growth earnings ÷ cost of capital) and a residual-income cross-check (RIV = book equity + present value of earnings above a normal charge on that equity; Penman 2011) each point to a no-growth value far below spot. Illustratively, EPV ≈ FCF $33B ÷ WACC ~9% − net debt $45B ≈ ~$320B equity ≈ ~$67/sh (illustrative) — far below spot, confirming that essentially all of AVGO’s value sits in the growth / competitive-advantage period, not in steady-state earnings. That is precisely why the margin of safety is thin at the top of the band.
• Single-method terminal — A.3 uses a forward-P/E terminal frame only; method (Damodaran) asks for a second terminal — a Gordon-growth perpetuity (terminal value = normalized cash flow ÷ [discount rate − long-run growth]) — reconciled to it. The $9.80–$10.70 EPS-band triangulation already in A.3 is the practical analogue; a single-method terminal is a known fragility to flag, not to hide.

• Path — 52-wk low $243.80 → ATH $495 (Jun-3) → gapped −15% at the Jun-4 open ($409 vs $479) → $372.10 (Jun-10 close), now ~22% off the Jun-3 close.
• Moving averages — below the 50-day $403.55 (overhead resistance); just above the rising 200-day $357.22 (key support).
• Volume — Jun-4 gap on 81M shares vs ~25M typical → a high-conviction distribution day, not quiet profit-taking.
• Support — ~$357 (200-day, THE line that defines dip vs. trend change), ~$345 (pre-run-up gap/consolidation + round number), then the $300 zone.
• Resistance — ~$404 (50-day, first overhead reclaim), ~$419 (Jun-4 gap-down close), then $479 (Jun-3 close) and the $495 all-time high.
• Stage trigger (Weinstein) — a discretionary heuristic, not a prediction: a weekly close below the 200-day (~$357) would tip it from Stage-3 distribution toward Stage 4; a reclaim of the 50-day (~$404) on volume would lean back toward the Stage-2 advance — a tactical tell, not a forecast (chart rules carry weak, decay-prone edge).

Three quantitative priors — one peer-reviewed, two practitioner, all portfolio-level, not single-name proof — currently point the same near-term way as the chart.

• PEAD on SUE — Q2 was a positive earnings surprise (a beat: GAAP $1.91 / non-GAAP $2.44); SUE (standardized unexpected earnings = surprise ÷ the firm’s own surprise volatility) is positive in sign — we do not standardize the magnitude here. The textbook base rate is post-beat upward drift over ~60 trading days — but that anomaly is sorted on SUE cross-sectionally, not on the price reaction (Bernard–Thomas 1989, academic; portfolio base rate, not a single-name forecast). Here the stock gapped −15% on the beat because the ~$56B FY26 AI target was reaffirmed-not-raised; a negative price reaction to a positive surprise is a guidance reset that overrides the post-beat lean. Net: PEAD tailwind neutralized — confidence Medium, and PEAD is known to decay in liquid mega-caps.
• Relative-strength line — the RS line (price divided by the index/sector, judged by new highs vs new lows — Mansfield/O’Neil construct, practitioner, not peer-reviewed). The 12-month line is strong (+52% 1y), but with spot below the 50-day and the Jun-4 distribution day the near-term line is rolling over, not making fresh highs. This is the same momentum vote Lens C casts (deduped in Synthesis). Confidence Medium.
• 52-week-high prior — nearness to the 52-week high predicts continuation cross-sectionally (George–Hwang 2004, academic; portfolio prior, not single-name proof). At ~$372 the stock sits ~25% below the $495 ATH (Jun-3), which is the unfavorable/bottom bucket — the 52wk-high anchor is currently broken, a near-term headwind to the momentum prior, not a tailwind. The Weinstein stage label (Stan Weinstein, practitioner) already carried in the prose is consistent. Confidence Medium.

• Low-vol / Beta: high-beta (~2.0) — the opposite of a defensive/low-vol tilt; structurally a leveraged market/sector bet (Betting-Against-Beta makes low-β the risk-adjusted tailwind, not high-β).
• Quality: long — top-decile profitability (39.9% EBIT margin, $33B TTM FCF) and Piotroski 9/9 (see Skeptic’s Audit); the most durable factor.
• Value: short — ~35× fwd / ~62× trailing GAAP; rich on every multiple shown.
• Momentum: mixed — +52% on 12 months (positive 12–1) but the near-term trend flipped negative (below the 50-day) — the same momentum signal a GARP/CANSLIM read surfaces (deduped in Synthesis).
• Size: mega-cap (large) — liquid, negligible size premium.
• Duration: long — a ~35× multiple ≈ ~35-year equity duration → acutely rate-sensitive (see Macro Overlay).

A conviction-weighted join (each lens enters with weight ∝ its conviction tier, not a flat one-vote) keeps the lenses from out-voting each other on labels alone. The point is not a fourth verdict — it is that the three converge, with their shared evidence counted once.

• Dedup — momentum once — the broken trend (Lens B), the momentum factor (Lens C) and the GARP/CANSLIM read are the same underlying momentum signal, not three votes — a 52-week-high read and a GARP momentum read are one signal (George & Hwang 2004). Counting correlated factor premia once (they decay and co-move post-publication — McLean & Pontiff 2016) avoids over-weighting one input. Likewise rate-sensitivity shows up in both Lens C (duration) and the Macro Overlay — one risk, not two.
• Reaction attribution — the Jun move splits (Campbell–Shiller: a return is cash-flow news − discount-rate news) into a large discount-rate leg (hot CPI ≈ +4.2% / a hawkish FOMC compressing every long-duration multiple) and a small cash-flow leg (guide reaffirmed-not-raised, software +9% decel). A move that is mostly multiple-repricing with forward estimates roughly intact is the textbook variant-perception setup — price moved more than value (consistent with the rate-sensitivity read in the Macro Overlay).
• Convergence, not conflict — all three point to one action: own the franchise with a margin of safety, not at the top of the band. Spot at the top of the band (Lens A), below the 50-day (Lens B), on a ~2.0-beta long-duration chassis into a hawkish print (Lens C) is the same caution seen three ways.

Consensus says: phenomenal AI franchise, expensive stock, fairly valued after the reset — own it, don’t chase it. The variant worth monitoring is that the market is under-appreciating how much of the FY27 $100B AI target is already de-risked by >$30B of in-quarter bookings and six reported XPU customers — some with multi-GW 2027 commitments cited by trade press (confirmation varies) — i.e. the "reaffirmed not raised" guide is conservatism, and a raise is, on the bull variant, more likely than not (still a minority ~25% in our weighting). The equally live opposite variant: the >$100B FY27 number is a customer-roadmap aspiration, not contracted backlog, and a single hyperscaler in-sourcing decision (Google’s TPU is already internal) or capex pause re-rates the whole complex. The Q3 print — does FY26 AI get raised? — discriminates between these worlds better than any narrative.

What tilts it — management’s track record: the "conservatism vs soft ceiling" question is not symmetric. Tan has a multi-quarter pattern of setting AI targets deliberately below the eventual print and then raising them (the ~$56B FY26 figure was itself stepped up over prior quarters as bookings landed). That history argues the reaffirmation is sandbagging, not a wall — which is why the bull variant deserves real weight. The check on it is the ~$73B disclosed AI backlog (per Q2 FY26): it covers ~73% of the FY27 $100B (see the Evidence Ladder), so ~27% still has to be won, not just converted — the part where a raise could stall.

Competitive frame: AVGO is not "cheap NVIDIA". It is the larger half of the custom-silicon (XPU) duopoly with Marvell — winning the share of AI capex that hyperscalers deliberately route away from merchant GPUs. The two live competitive threats are (a) Marvell taking design wins at the margin, and (b) NVIDIA pushing NVLink/Spectrum networking into Broadcom’s Tomahawk/Jericho switching turf — the AI-networking franchise that rides alongside the XPUs. Neither is acute today, but both are why "custom silicon is structurally Broadcom’s" is a thesis, not a fact.

Bull loop: AI revenue beats and gets raised → multiple re-rates → cheap equity/credibility funds more design wins and packaging capacity → bookings compound → the FY27 $100B looks conservative → confidence builds.

Bear loop: a reaffirmed-not-raised guide or a capex wobble → "AI-spending peak" narrative hardens → correlated AI-cohort de-rate (the Jun-5 $1T session) → multiple compresses → a hyperscaler accelerates in-sourcing to cut cost → the XPU concentration that powered the ramp becomes the overhang.

• Jun 16–17, 2026 — FOMC under Chair Warsh after the hot CPI: sets the multiple environment for every long-duration AI name.
• ~late Jun 2026 — quarterly dividend ($0.65/share) payment.
• ~early Sep 2026 (est) — Q3 FY26 earnings: THE catalyst — does FY26 AI revenue get RAISED above $56B? Q3 AI run-rate vs the $16B guide; software re-accelerate or decelerate; gross-margin trajectory as AI scales.
• H2 2026 — hyperscaler capex updates (Google, Meta, Microsoft, Amazon prints) — the demand signal underneath the XPU bookings.
• Ongoing — any newly confirmed or newly reported XPU customer or multi-GW commitment; export-control changes affecting the ~17% China exposure.

1. The multiple, not the business: ~35× forward non-GAAP / ~62× trailing GAAP leaves a thin margin of safety — a record print with a strong near-term guide still produced a ~22% drop, which is what "priced for perfection" looks like.
2. Software is decelerating: +9% YoY (vs +26% in FY25), below unofficial whisper expectations — the VMware engine that justified part of the multiple is lapping out, and there is no obvious third growth leg.
3. Customer concentration / in-sourcing: see the Concentration section — a hyperscaler can slow, slip, or in-source, and the FY27 >$100B is roadmap, not backlog.
4. Margin mix headwind: custom AI silicon is lower-margin than software (and than mature wireless ASICs); as AI grows to ~half of revenue, blended gross margin faces structural pressure even as dollars surge.
5. Balance-sheet weight: net debt ~$45B (vs net-cash NVDA/AMD) plus $7.6B of annual SBC mean the GAAP-to-non-GAAP gap is large and the buyback is gated by deleveraging.
6. Correlated AI de-rate + the semi cycle: when half the market’s growth is one capex theme, a repricing is violent and correlated — AVGO’s own outlook triggered the Jun-5 $1T sector drawdown. Layer on the classic semiconductor risks the AI narrative has masked: a cyclical downturn, AI-accelerator oversupply / supply glut once capacity catches demand, the memory cycle (HBM/DRAM volatility in the system BOM), and tightening China export controls on the ~17% China book.

What flips the thesis, and how hard the evidence under the bull case actually is.