The Hock Tan franchise-consolidation machine — custom AI accelerators co-designed with Google, Meta and OpenAI, the dominant AI networking franchise, and a VMware software annuity. The arms dealer to the companies trying to escape NVIDIA.
Our Position
Research
The verdict
The most dangerous competitor bulls underestimate — the named beneficiary of NVIDIA's own inference-share vulnerability, growing AI +143% — but priced at a PREMIUM to NVIDIA (45× vs 30× EV/EBITDA) on one-third the ROE, with lower earnings quality (acquisition-amortization add-backs, $64B debt) and the same hyperscaler concentration. The most expensive, lowest-quality way to play the custom-silicon thesis. NEUTRAL / WATCHING / MEDIUM.
Two businesses fused by Hock Tan's M&A machine: Semiconductor Solutions ($15.0B Q2 FY26, +79% Y/Y, ~68%) and Infrastructure Software ($7.2B, +9%, ~32%). The semi half is two engines — custom AI accelerators (XPUs) co-designed with hyperscalers, and the dominant merchant AI networking franchise (Tomahawk/Jericho/optical). Software = the VMware roll-up ($69B, 2023) + CA + Symantec. The model: not an edge innovator but a franchise consolidator — buy mission-critical high-switching-cost tech, cut non-core R&D, re-price to the top ~1,000 customers, harvest cash. Named customers: Google (TPU), Meta (MTIA), OpenAI (first custom chip 2027), Anthropic (1GW 2026→3GW 2027) + two unnamed — "six mega-deals." Investable identity: the arms dealer to the companies trying to escape NVIDIA, with a software annuity bolted on for cash stability.
Fabless, one layer below NVIDIA. Upstream: TSMC (3nm/5nm) + CoWoS (the same chokepoint NVIDIA/AMD queue for), HBM from SK Hynix/Samsung/Micron, EDA from Synopsys/Cadence. Downstream: unlike NVIDIA (sells finished GPUs/racks), Broadcom sells the design + silicon to the hyperscaler, who owns the system and the margin on integration. NVIDIA monetizes the rack; Broadcom monetizes the wafer and hands integration (and margin) to the customer. Capex trivially small — $231M Q2 vs $10.5B operating cash — TSMC carries the capital intensity. The asset-light mirror of the fab-heavy names in this sweep.
| Segment | Q2 FY26 | Y/Y | % | Op margin |
|---|---|---|---|---|
| Semiconductor Solutions | $15.0B | +79% | ~68% | 62% |
| — of which AI semi | $10.8B | +143% | ~49% of total | — |
| Infrastructure Software | $7.2B | +9% | ~32% | 79% |
| Total | $22.19B | +48% | 100% | 67% |
| AI is ~49% of revenue and the whole growth story: FY25 AI ~$20B → FY26 guided $56B → FY27 reiterated >$100B; split ~60% XPU / 40% networking. A slow-growth ultra-high-margin software annuity (32%) financing a hyper-growth, customer-concentrated AI silicon business (49%) — the inverse of NVIDIA's 92%-Data-Center monoline. |
Revenue $22.19B, +48%; AI semi $10.8B, +143% above guide. Non-GAAP EPS $2.44 vs ~$2.32 (beat ~5%). 67% op margin, 69% adj-EBITDA margin, FCF $10.26B (46% of revenue). Guidance: Q3 ~$29.4B (vs ~$28.5B Street), Q3 AI $16.0B (>200% Y/Y), FY26 AI $56B, FY27 AI >$100B reiterated. Stock reaction: −12.6% next session (June 4), ~$280B erased — beat-and-fall: the FY-AI number was left unchanged and software was soft, so a stock priced for an AI raise sold off on its absence. A structural echo of the NVIDIA pattern (DeepSeek-day −17%/−$600B) — elevated AI expectations make even good prints a downside source.
Hock Tan escalated from measured to near-evangelical: XPU/networking demand "simply insatiable," $30B+ Q2 AI bookings, six mega-deals, an "Apollo" first tranche ~$35B. The tell: reiteration, not expansion, of FY27 >$100B — disciplined enough to be credible, but the market front-ran a raise. Everything anchored to a small set of named whales — confidence on the call, concentration risk in the 10-K.
| Ticker | Mkt cap | EV/EBITDA | P/E (TTM) | Fwd P/E | Div yld | ROE |
|---|---|---|---|---|---|---|
| AVGO | $1.87T | 45.5x | 66.0x | 24.5x | 0.66% | 37.3% |
| NVDA | $5.05T | 30.3x | 32.0x | 21.0x | 0.49% | 114.3% |
| AMD | $797B | 101.2x | n/a | 53.5x | — | 8.1% |
| MRVL | $230B | 85.5x | 91.0x | 58.1x | n/a | 16.0% |
| TSM | $1.89T | 20.0x | 31.2x | 21.8x | 0.76% | 36.2% |
| ; NVDA row carried from the anchor; EV/EBIT + 5yr-avg ROE n/a. Where AVGO screens: a clear premium to NVIDIA on every earnings multiple (45.5× vs 30.3× EV/EBITDA, 66× vs 32× P/E, 24.5× vs 21× forward) despite one-third the ROE (37% vs 114%). The market pays up for Broadcom's acceleration and down for NVIDIA's deceleration from dominance. The premium is only defensible if you underwrite FY27 >$100B; on current earnings quality NVIDIA is the higher-quality compounder. TSM at 21.8× forward is the value anchor — and the name every other ticker here depends on. |
Dec 2024 ~+24% — SAM disclosure ($60–90B AI by 2027, 3 hyperscalers, 1M-XPU clusters) launched the re-rate; AVGO doubled, crossed $1T. June 4, 2026 −12.6% / ~−$280B — beat-but-no-AI-raise + soft software. Macro/correlated drawdowns (DeepSeek Jan'25, tariffs Apr'25) hit it with the whole complex. The pattern is identical to NVIDIA's: up-moves from AI disclosure, down-moves from AI expectations.
The most accomplished serial acquirer in semis — the "Broadcom Playbook" across LSI/Brocade/CA/Symantec/VMware. FY25 returned $13.6B ($11.1B dividends + $2.5B buyback), 15th consecutive dividend raise. Large equity stake, performance-weighted comp — strong alignment. The distinct tension vs NVIDIA: the model runs on debt — VMware pushed long-term debt to ~$63.8B, and ~$20B/yr FCF splits between dividends and paydown rather than NVIDIA's near-pure buyback. Red flags: (1) the growth algorithm requires an ever-larger next acquisition (accretion treadmill); (2) VMware per-core re-pricing alienated customers, software grew only +9% — if churn accelerates the "annuity" weakens (it contributed to the June-4 drop); (3) post-deal R&D cuts maximize near-term margin but raise a long-term innovation question outside AI. Best-in-class capital allocator, but levered, acquisition-dependent, customer-unfriendly software playbook.
Forensic bottom line: Broadcom is the most dangerous competitor bulls underestimate — the named beneficiary of NVIDIA's named vulnerability — but its earnings quality is lower, not higher: where NVIDIA's risk sits off the income statement on clean cash earnings, Broadcom's sits inside the earnings construction (amortization add-backs, $64B debt, wider GAAP gap) plus it shares NVIDIA's hyperscaler concentration. The pair is doubly-levered to one event — a hyperscaler capex slowdown takes both.
Bottom-up; Broadcom FY ends early Nov (FY26 = Nov'25–Nov'26).
| Scenario | FY26 rev / non-GAAP EPS | FY27 | FY28 | Assumptions |
|---|---|---|---|---|
| Base | ~$98B / ~$10.20 | ~$135B / ~$15.00 | ~$160B / ~$18.50 | AI $56B FY26 → ~$100B FY27 (guided); software +8%; margin holds; modest debt paydown |
| Bull | ~$100B / ~$10.60 | ~$150B / ~$17.50 | ~$200B / ~$24 | custom-silicon share grab accelerates; new mega-deal; networking attach scales |
| Bear | ~$94B / ~$9.40 | ~$110B / ~$11 | ~$110B / ~$10.50 | one hyperscaler re-insources/re-bids an XPU program; VMware churn; software stalls; AI growth halves |
Log: npx tsx scripts/research/forecast.ts create --topic hardware --question "AVGO FY27 AI semi revenue >= $100B" --p 0.55 --resolves 2027-12-15 --tags broadcom,deep-dive |
Bull. Broadcom is the structural winner of the one thing NVIDIA's own forensic file admits it loses — commodity inference to custom silicon (90%→20-30% by 2028). It's growing AI +143% off a backlog of $30B+ bookings and six multi-year hyperscaler mega-deals, it owns the networking layer NVIDIA must defend, and a 79%-margin VMware annuity funds it all. FY27 >$100B AI is the number; if it lands, the premium is earned. Bear (impairment vectors). (1) The premium multiple (45× EV/EBITDA, 24.5× forward) on one-third NVIDIA's ROE leaves no room for a merely-excellent quarter — the −12.6% June-4 drop proved it. (2) Earnings quality is structurally lower than NVIDIA's: amortization add-backs flatter non-GAAP, $64B debt, a self-defined AI metric. (3) It shares NVIDIA's hyperscaler concentration in the same customers — doubly-levered to one capex cycle, with the added single-customer-program reset risk. Pre-mortem (2027): Google re-bids its TPU program or brings more in-house, an XPU mega-deal resets, VMware churn accelerates as enterprises finish escaping the per-core repricing, and the FY27 $100B AI number slips to $80B — on a 24.5× forward multiple priced for the raise, the stock halves, and the $64B debt turns a growth story into a deleveraging one. Contrarian view: the market treats Broadcom as the "safer" AI bet than NVIDIA (it has software, it has the custom-silicon future) — but the data say the opposite: it's the higher-multiple, lower-ROE, more-levered, lower-earnings-quality name, exposed to the same customers. The variant perception is that Broadcom is the riskier of the two, not the hedge.
A custom-silicon "win" is a single-customer dependency: if Google re-bids or re-insources one program, an entire revenue line resets overnight — there's no CUDA-style switching cost protecting Broadcom, because the customer owns the stack. The growth algorithm is an accretion treadmill that needs an ever-bigger next acquisition; at $1.9T that next deal barely moves the needle. The non-GAAP EPS is flattered by billions of amortization that is real money spent. $64B of debt in a higher-rate world is a deleveraging anchor. And the whole AI number is management-defined guidance, not a booked, auditable segment. Strip the amortization add-back and the leverage, share NVIDIA's exact customer concentration, and pay a premium to NVIDIA for it — that's the bear case in one breath.
Number-two in high-bandwidth memory and the only US maker of both DRAM and NAND.