Stock Analysis · Bulls vs Bears

Broadcom Inc. AVGO

🐂 Bulls vs Bears 🐻Report #211 Oct 2026
★ Willem owns this stock · 2.8% of the portfolioOlder version — see the latest →
52
high risk▲ +1 vs last
🐂 8vs9 🐻
Bears lead
62%
▲ Chance of the biggest tailwind the forecasts miss
30%
▼ Chance of the biggest headwind the forecasts miss
The verdict

The engineering franchise looks as strong as three days ago: Q3 beat, TPU 8i shipped first, Apple through 2031, and Marvell's Google deal is mostly peripheral chips. What keeps moving the wrong way is the financing layer, which now covers a guaranteed senior tranche for Anthropic and talks for OpenAI, and its total size is still undisclosed.

🎙️ Bulls vs Bears podcast — Broadcom Inc.
Alex and Sam debate every bull and bear argument · 10:17
0:0010:17
⤓ Download

Written by Claude (Anthropic’s AI) from public sources · Willem owns AVGO · AI can make mistakes — numbers in brackets point to the sources · Not financial advice

🐂 Bull arguments 8

  1. 1Still the default partner for "build your own AI chip"Strong
  2. 2An operating-margin and cash machineStrong
  3. 3Engineering and supply-chain scale as a moatStrong
  4. 4It owns the network inside AI data centresStrong
  5. 5Being the banker locks in the two biggest new customers (new)Medium
  6. 6VMware is a high-margin annuity, but the growth is slowingMedium
  7. 7The non-AI chip business is stabilising and Apple is locked inMedium
  8. 8The CEO is tied in through FY30Weak

🐻 Bear arguments 9

  1. 1Broadcom is becoming a lender and guarantor to its biggest customers, and the exposure keeps growingHigh
  2. 2The guidance arithmetic doesn't reconcile (content per gigawatt)Medium
  3. 3Concentration is rising, and the future 1 customer is in a fight with the US governmentMedium
  4. 4Google is multi-sourcing, and the customer owns the architectureMedium
  5. 5Shrinking gross margin and opaque deal structuresMedium
  6. 6Physical bottlenecks and pre-commitmentsMedium
  7. 7VMware backlash, migrations and EU scrutinyLow
  8. 8Geopolitics: Taiwan, China and tariffsLow
  9. 9Key-person risk and sentiment that swings on headlinesLow

Bull arguments show how well they survive the devil’s advocate (strong / medium / weak); bear arguments show the damage they could do after the counter-argument (high / medium / low).

Bull case — all positive points (most important first)

🐂 Bull #1Still the default partner for "build your own AI chip"

  • The point: every large AI buyer that wants a cheaper alternative to Nvidia GPUs ends up at Broadcom for at least one programme. Broadcom's execution record makes it hard to displace on the next generation.
  • Evidence:
    • Six XPU customers.
    • Google: Long Term Agreement for "future generations of TPUs", plus supply assurance up to 2031 (8-K, 2026-04-06) [11].
    • Apple: multi-generation custom-silicon agreements through 2031 (8-K, 2026-07-06) [12].
    • Broadcom's TPU 8i shipped ahead of MediaTek's version [4].
    • Remaining performance obligations (contracted future revenue) were $179B [2].
    • Counterpoint expects Broadcom to keep ~60% of AI-server ASIC design-partner share in 2027 [48].
  • Why it matters: custom chips are the fastest-growing part of AI compute, and design wins lock in multi-year, multi-generation revenue.
  • 😈 Devil's advocate:
    • The customers own the architecture and are deliberately adding second sources.
      • MediaTek has raised its 2027 AI-ASIC share target to 15–20% of an $80B market and starts production in Q4 2026 [27].
      • Bernstein (reported) sees MediaTek AI-ASIC revenue at $16B in 2027 and $44B in 2028 [27].
    • The Google contract guarantees a relationship, not a share.
  • Verdict: Strong Broadcom keeps winning on time-to-market. MediaTek's numbers are big but still a fraction of Broadcom's ~$115B FY27 path.

🐂 Bull #2An operating-margin and cash machine

  • The point: even as AI growth dilutes gross margin, Broadcom converts almost half its revenue to free cash.
  • Evidence:
    • Q3 FY26 [3,4]:
      • Free cash flow $13.7B (46% of revenue).
      • Operating margin ~68%, with Q4 guided ~66%.
      • Cash $24.0B.
    • It repaid $5.6B of debt in Q3 and $1.5B more after quarter-end [4,5].
  • Why it matters: the cash funds the AI ramp, the dividend ($0.65 a quarter) and deleveraging without new equity.
  • 😈 Devil's advocate:
    • The growth in free cash flow increasingly depends on customers whose payments Broadcom partly guarantees.
    • Pre-buying $126.8B of supply ties up cash if a ramp slips [2].
  • Verdict: Strong for now. Operating margin held even after gross margin fell ~4 points.

🐂 Bull #3Engineering and supply-chain scale as a moat

  • The point:
    • In a market where supply is the bottleneck, the supplier that has locked up wafers, HBM, substrates and lasers wins orders.
    • Broadcom also brings IP that is hard to copy.
  • Evidence:
    • TSMC says packaging is limiting customers' growth and expects strong demand through ~2029–30 [32].
    • Samsung's CFO said HBM4 for 2026 was "completely sold out" [33].
    • Broadcom has $126.8B of purchase commitments [2]. It is tripling indium-phosphide laser capacity and building a substrate fab in Singapore [4].
    • Tan claims "the strongest IP portfolio in semiconductor design", naming SerDes, chip-to-chip interconnect, HBM/SRAM integration and packaging [4].
  • Why it matters: customers choose the partner who can promise delivery dates. Second sources like MediaTek must first secure capacity; MediaTek's board approved a $5B budget to lock supply [27].
  • 😈 Devil's advocate:
    • Nvidia remains TSMC's largest advanced-packaging customer. Nvidia's supply commitments rose from $119B to $279B in one quarter [34].
    • One weak newsletter source even claims Broadcom's packaging allocation was trimmed (unconfirmed).
    • Supply lock-in also means supply risk if demand slips.
  • Verdict: Strong / Medium. The moat is real through 2028 but rests on management claims; the allocation is not independently verified.

🐂 Bull #4It owns the network inside AI data centres

  • The point: AI clusters are moving to Ethernet, and Broadcom's Tomahawk and Jericho chips sit inside most non-Nvidia Ethernet switches.
  • Evidence:
    • Dell'Oro (2026-06-02): Ethernet was about two-thirds of AI-cluster switch sales in Q1 2026, and its AI back-end sales more than doubled [30].
    • The scale-up standards ESUN and SUE build on Broadcom contributions [26].
    • Broadcom's 51.2T co-packaged-optics switch completed deployment validation with hyperscalers such as Meta. The CPO volume ramp is expected in 2027–28 [31].
  • Why it matters:
    • Networking is sold to every AI builder, not just XPU customers, including GPU clusters.
    • It diversifies Broadcom away from the custom-chip concentration.
  • 😈 Devil's advocate:
    • Nvidia's own Ethernet (Spectrum-X) grew 2.6x year on year, and Nvidia aims to be "the largest and fastest-growing network company" [35].
    • Nvidia shipped a CPO switch first [31].
    • Marvell's Teralynx 102.4T switch supports every scale-up standard [26].
    • InfiniBand revenue more than tripled in Q1 [30].
  • Verdict: Strong / Medium. Ethernet is winning the protocol war, but Broadcom shares that win with Nvidia's own Ethernet.

🐂 Bull #5Being the banker locks in the two biggest new customers (new)

  • The point:
    • By arranging and partly guaranteeing financing, Broadcom becomes almost impossible to replace at Anthropic and OpenAI.
    • Anthropic's $125.2B five-year TPU lease runs through Broadcom's platform [14].
  • Evidence:
    • The first $35B XPV tranche reportedly achieved investment-grade senior ratings because of the Broadcom backstop [15].
    • Anthropic placed cash in a restricted account for Broadcom's benefit in April 2026 [14].
    • The notes convert into Anthropic equity, which is reportedly targeting a ~$2T valuation at IPO [14].
    • Tan: "It's a great investment for us" [4].
  • Why it matters: it converts the labs' weak credit into Broadcom-quality financing. That brings forward gigawatts that would otherwise wait, and the equity link gives upside if the labs succeed.
  • 😈 Devil's advocate:
    • It is the same structure critics call "circular financing": the supplier funds the customer's purchases [47].
    • Anthropic's own prospectus flags "potential conflicts of interest" [14].
    • It works only while capital markets stay open for AI.
  • Verdict: Medium It is a real competitive tool, but its value and its risk are the same bet (see Bear #1).

🐂 Bull #6VMware is a high-margin annuity, but the growth is slowing

  • The point: VMware's installed base is hard to rip out and pays a ~94% gross margin. That funds everything else.
  • Evidence:
    • Software revenue was $8.75B in Q3 (+29%), with ~84% operating margin [4].
    • Software backlog was $73B at FY25 year-end [8].
    • Nine-month software growth of 13% matches the December "low double-digit" guide [8].
  • Why it matters: a stable cash base at a near-software-company margin cushions AI-cycle volatility.
  • 😈 Devil's advocate:
    • ARR growth decelerated from 19% (Q1) to 17% (Q2) to 15% (Q3) [7,6,4].
    • The Q3 jump came from upfront licence recognition ($3.5B) on non-cancellable contracts, not underlying momentum [2].
    • Gartner (via trade press) expects ~70% of enterprise VMware customers to move at least half their workloads by 2028 [43].
    • CISPE's EU complaint (2026-03-19) cites price increases above 1,000% [42].
  • Verdict: Medium (weakened from last time). The cash is real, but the high growth rate is partly an accounting timing effect.

🐂 Bull #7The non-AI chip business is stabilising and Apple is locked in

  • The point: the legacy chip business, once feared to lose Apple, is recovering and now has a deal through 2031.
  • Evidence:
    • Non-AI chips were ~$4.2B in Q3 (+5%), with a ~$4.3B Q4 guide [4].
    • Q2 non-AI bookings were above $6B ("path towards a full cyclical recovery") [6].
    • The Apple multi-generation agreements run through 2031 [12].
  • Why it matters: removes a tail risk to ~15% of revenue.
  • 😈 Devil's advocate:
    • The Apple 8-K gives no value.
    • Wireless is still declining [4].
  • Verdict: Medium

🐂 Bull #8The CEO is tied in through FY30

  • The point: Tan's award vests only if he stays to the end of FY30, so the key person behind the strategy is locked in.
  • Evidence:
    • Payout is 0–300% of 610,521 shares, measured on AI revenue in FY28–30 [9].
    • Tan must hold after-tax shares from his 2023 award through FY30 [9].
  • Why it matters: retention through the critical FY27–28 ramp.
  • 😈 Devil's advocate:
    • The maximum payout is hit at $120B of AI revenue over four quarters.
    • Management already guides ~$115B for FY27 and ~$230B for FY28.
    • Beyond retention, the award gives no extra incentive to choose growth or margin quality.
    • Tan made only a gift (22,000 shares, 2026-04-08) and no open-market purchase [13].
  • Verdict: Weak (weakened). It keeps Tan in place, but as an incentive it is already largely won.

Bear case — all negative points (most important first)

🐻 Bear #1Broadcom is becoming a lender and guarantor to its biggest customers, and the exposure keeps growing

  • The point: to keep the two AI labs ordering, Broadcom has moved from chip supplier to financier. Each month adds a new layer.
  • Evidence:
    • 10-Q (period to 2026-08-02):
      • A backstop of up to ~$29B covers Anthropic's 5-year rack leases. Broadcom pays the gap between 85% of the amount owed and the racks' resale value.
      • Up to $42B of convertible notes may be issued by the customer to Broadcom; none were issued yet.
      • "Residual value guarantees" are possible on future tranches [2].
    • Since the last dive:
      • Bloomberg (2026-10-06, relayed): BofA, Citi and Morgan Stanley started syndicating a $42B senior Class A tranche with a partial Broadcom residual-value guarantee (size not disclosed). An $18B junior Class B tranche is led by Blackstone [16].
      • WSJ (2026-10-08) reports Broadcom is arranging more than $50B for OpenAI's chip. Bloomberg (2026-10-09) says ~$30B, in early talks [19,20].
    • The CFO declined to give a total maximum exposure. Future deals will be "evaluated deal by deal" [5].
    • Credit markets noticed:
      • Broadcom 5-year credit default swaps (CDS, the price of insuring its debt) rose 28bp in August [17].
      • BofA credit research cut Broadcom bonds to Neutral on 2026-08-10, citing "significant uncertainty" around XPV [18].
      • No 2026 Moody's, S&P or Fitch action was found. A relayed Moody's quote about "financial flexibility" is UNCONFIRMED.
  • Why it matters: if one lab stumbles, Broadcom is hit three ways at once: lost orders on a guided $230B FY28, a guarantee call, and losses on the notes.
  • 🛡️ Counter-argument:
    • The guarantee pays only the gap after the racks are resold, and AI hardware resale demand is strong today.
    • Four of six XPU customers fund themselves [5].
    • The junior tranche (Blackstone) takes Anthropic credit risk ahead of Broadcom's guarantee on the senior tranche [16].
    • An Anthropic IPO would sharply improve the borrower's credit.
  • Severity: High It is the one channel where an AI-funding downturn hits Broadcom's revenue and balance sheet at the same time. The total is still undisclosed and still rising.

🐻 Bear #2The guidance arithmetic doesn't reconcile (content per gigawatt)

  • The point:
    • Management's dollar guides and its "per gigawatt" claims imply very different things.
    • Either a large share of the ~30 GW announced lands after FY28, or Broadcom's content per GW is far below what Tan says.
  • Evidence:
    • Q3 call, 2026-09-02 [4,5]:
      • Bernstein's Rasgon: the ~10 GW (2027) and ~20 GW (2028) plans divided into the $115B and $230B guides give "somewhere between 11 and 12 billion per gigawatt".
      • Tan said not all 30 GW reach production by FY28, but gave no number.
      • In another answer he said content is "$20 to $30 billion per gigawatt" and "very sustaining".
    • His per-GW range has drifted: "$15–20B" (cited in June) → "$20–30B" (September) [6,4].
    • Interpretation:
      • At $25B/GW, the $345B FY27–28 total implies only ~14 GW shipped in two years, versus ~30 GW announced.
      • That means either the customers' gigawatt plans are heavily back-loaded beyond FY28, or a large part of the gigawatts uses non-Broadcom content (networking from Nvidia/Arista, memory bought directly).
  • Why it matters:
    • If the gigawatts slip, the FY28 number is at risk.
    • If content is lower, the per-GW pitch overstates Broadcom's share of each AI dollar.
  • 🛡️ Counter-argument:
    • Management has consistently under-promised and over-delivered on dollars.
    • Tan explicitly called the guide "judged" and "conservative" on gigawatts. Slipped gigawatts would be upside for FY29.
  • Severity: Medium It's a transparency problem more than a demand problem, but it makes FY28 harder to underwrite.

🐻 Bear #3Concentration is rising, and the future #1 customer is in a fight with the US government

  • The point: Broadcom's revenue is concentrating fast. Its biggest future customer carries a federal "supply chain risk" label that Broadcom's own filing now names as a reason it might have to stop selling.
  • Evidence:
    • One distributor handled 50% of Q3 revenue (32% a year earlier). The top five end customers were ~55% [2].
    • The new 10-Q risk factor says that if customers "are designated a 'supply chain risk,' we may be required to suspend… selling to such customers". The phrase does not appear in the FY25 10-K [2,1].
    • The Pentagon designated Anthropic a supply chain risk in March 2026 [40]:
      • A California district judge ruled the label unlawful on 2026-08-27 [39].
      • The D.C. Circuit upheld it 2-1 around 2026-09-25, on a different legal basis [38].
      • The Pentagon says the designation stands [38].
    • Anthropic's 3.5 GW is "dependent on Anthropic's continued commercial success" (8-K) [11].
  • Why it matters:
    • The more the FY27–28 plan rests on Anthropic and OpenAI, the more any customer-specific shock — legal, financial or political — becomes a Broadcom shock.
  • 🛡️ Counter-argument:
    • Interpretation: the designation as reported bars use in Defense Department work. It is not a ban on commercial chip sales to Anthropic.
    • Google, Meta, Apple and Broadcom's networking customers spread the base.
    • The distributor share mostly reflects routing, not a single buyer.
  • Severity: Medium A low-probability but high-impact legal tail sits on top of rising concentration.

🐻 Bear #4Google is multi-sourcing, and the customer owns the architecture

  • The point: Broadcom's original anchor franchise, Google's TPU, is being shared.
  • Evidence:
    • MediaTek designs one TPU 8 variant, and TrendForce reports market watchers think it may also win v9 [23].
    • Marvell's Google deal (2026-07-29 agreement, warrant issued 2026-08-18) gives Google up to ~59M shares at $206.58, vesting per $500M of purchases through FY33 [24].
    • Marvell calls it a "$120 billion opportunity over six and a half years" [25].
    • Macquarie (June) saw Broadcom's TPU share falling from ~95% in 2026 to ~65% by 2028 (reported) [28].
    • Broadcom fell 4.6% to $362.48 on the Marvell news (secondary aggregator relay, not re-checked).
  • Why it matters: each second source caps price and share on future generations of Broadcom's largest single programme.
  • 🛡️ Counter-argument:
    • Marvell's 8-K lists inference accelerators, storage and memory controllers, NICs and near-memory compute that "attach to the TPU ecosystem". That is mostly peripheral silicon, not the flagship training/inference XPU [24].
    • Marvell's own FY31 custom target is ~$30B across all customers [25].
    • Macquarie re-upgraded Broadcom to Outperform after Q3 (secondary aggregator relay, not re-checked).
  • Severity: Medium (unchanged). Share erodes but dollars still grow, unless MediaTek wins the next flagship.

🐻 Bear #5Shrinking gross margin and opaque deal structures

  • The point:
    • More pass-through memory and rack hardware means less gross profit per dollar of revenue.
    • Management's account of what it sells keeps changing.
  • Evidence:
    • Gross margin: 77.1% (Q2) → 75% (Q3) → ~73% guided (Q4), versus ~78% a year earlier [4,6].
    • No FY27 gross-margin guide: "we guide one quarter at a time" [4].
    • The Q3 call transcript even shows a garbled mid-call correction of the chip-segment gross margin.
      • Interpretation: our derivation from the press-release figures is ~67%.
    • The racks story shifted three times (see section 1) [8,6,2].
  • Why it matters: less margin buffer if XPU pricing pressure arrives. It is also harder to judge how much of "AI revenue" is Broadcom value-add.
  • 🛡️ Counter-argument: operating margin has held at ~66–68%, and absolute gross-profit dollars are surging [4].
  • Severity: Medium

🐻 Bear #6Physical bottlenecks and pre-commitments

  • The point: the FY27–28 plan needs packaging, HBM, power and buildings to show up on time, and Broadcom has pre-committed to buy much of the supply.
  • Evidence:
    • TSMC: packaging is "limiting my customers' growth"; the gap is "very big" [32].
    • SK hynix calls 2027 the "worst year" for HBM supply [33].
    • New York's 1-year data-centre permit pause [46].
    • Purchase commitments of $126.8B [2].
  • Why it matters: a two-quarter slip at Anthropic moves tens of billions of revenue and leaves Broadcom holding inventory.
  • 🛡️ Counter-argument:
    • Hyperscaler 2026 capex keeps rising: Google $195–205B, Amazon ~$220B, Meta $130–145B [36].
    • Google says 2027 capex will "increase significantly" [36].
  • Severity: Medium

🐻 Bear #7VMware backlash, migrations and EU scrutiny

  • The point: the price increases that lifted VMware margins are pushing customers to plan exits, and regulators are looking.
  • Evidence:
    • The CISPE complaint went to the European Commission's competition department (DG COMP) on 2026-03-19 [42].
    • The EU General Court refused Broadcom's request to suspend the Commission's information requests (April 2026, via [42]).
    • ARR growth is slowing (19%→15%) [7,4].
    • Nutanix says 40% of its FY25 bookings were direct VMware replacements (trade press) [43].
  • Why it matters: VMware is the cash cushion. A slow bleed would show up at the 2027 renewal of the 2024 three-year contracts. That renewal timing is inferred; the company doesn't disclose cohorts.
  • 🛡️ Counter-argument:
    • Non-cancellable contracts and a $73B software backlog lock in revenue for years [2,8].
    • No company-disclosed churn.
  • Severity: Low Medium.

🐻 Bear #8Geopolitics: Taiwan, China and tariffs

  • The point: almost all advanced chips come from Taiwan, China is reviewing Broadcom switches, and US chip tariffs are expanding.
  • Evidence:
    • FT (2026-09-23, via Reuters): China's state-asset regulator (SASAC) surveyed Broadcom switch use in state-backed data centres, where Broadcom reportedly has up to ~90% share. Informal guidance to reduce use is possible [21,22].
    • Section 232 phase two (tariffs on imported chips and possibly servers) is still unsigned. Lutnick: "If you build in America, we will give you tariff relief" [44].
    • 17% of FY25 revenue shipped to China, though "a substantially smaller percentage" depends on Chinese end customers [1].
  • Why it matters: Taiwan is the existential tail. China and tariffs are margin and share nicks.
  • 🛡️ Counter-argument:
    • The US–China truce was extended to 2027-01-10.
    • US data centres are exempt from the phase-one 232 tariff [45].
  • Severity: Low Medium (Taiwan aside).

🐻 Bear #9Key-person risk and sentiment that swings on headlines

  • The point:
    • One 74-year-old CEO with no named successor runs the strategy, alongside a CFO in the seat for four months.
    • The stock moves 4–6% on single AI-financing headlines.
  • Evidence:
    • No successor is named [9].
    • Stock moves: -4.6% on the Marvell news (2026-08-19) and +3.7% on the syndication news (2026-10-06) (secondary aggregator relay, not re-checked).
    • The shares trade ~$360–365 (2026-10-09, secondary), ~26% below the ~$495 high [29].
    • Insiders [13]:
      • Co-founder Samueli sold ~$500M under a pre-arranged trading plan (10b5-1, adopted 2025-12-16).
      • Executives Velaga ($13.6M) and Brazeal ($29M) sold.
      • The only open-market buy was director Harry You's $374K.
  • Why it matters: succession uncertainty could hit the multiple, and narrative risk ("circular financing") can swamp fundamentals for quarters.
  • 🛡️ Counter-argument:
    • Tan's award keeps him through FY30 [9].
    • The Samueli sales were pre-planned and small against his ~29.9M-share holding [13].
  • Severity: Low Medium.

Tailwinds NOT in the forecasts

  • Anthropic IPO prices and raises enough to fund its TPU leases — 62% chance within 12 months · Impact if it happens: High
    • What it is: a public listing that lifts Anthropic's credit standing and gives the up-to-$42B notes an equity value.
    • Why it's not priced in: consensus earnings models do not value the backstop's risk reduction or any note conversion. The stock still trades at a discount on "circular financing" fears.
    • What would trigger it / the sign to watch: a public S-1 (it must be out ≥15 days before the roadshow), a price range, then pricing [37].
    • Why this probability:
      • In favour: confidential filing since 2026-06-01, Nasdaq chosen, banks lined up [37].
      • Against: no public S-1 by 2026-10-11, an unresolved Pentagon dispute, and October slippage reports.
  • A seventh large XPU customer is named (e.g., a Microsoft Maia programme) — 38% chance within 36 months · Impact: Medium
    • What it is: a new hyperscaler programme beyond the six.
    • Why it's not priced in: Microsoft's reported talks to move Maia to Broadcom are unconfirmed, and consensus covers only named programmes.
    • Sign to watch: a new XPU customer named on a call, or an 8-K long-term agreement.
    • Why this probability: Marvell holds the Maia and Trainium programmes and Marvell says it serves all four US hyperscalers [25], so a switch is plausible but not likely.
  • Co-packaged optics and lasers become a second networking leg — 35% chance of material revenue within 36 months · Impact: Medium
    • What it is: optics built into switch packages, plus Broadcom's own lasers.
    • Why it's not priced in: today it is limited shipments only [31].
    • Sign to watch: Tomahawk 6 Davisson CPO in a hyperscaler production build. OCP Summit (2026-10-12/15) demos.
    • Why this probability: the industry expects volume in 2027–28, but Nvidia shipped first and pluggable optics keep improving.
  • Capital-return step-up in December — 45% chance within 3 months · Impact: Low
    • What it is: a larger dividend or buyback with Q4 results.
    • Why it's not priced in: management prioritised debt repayment, and no buyback was discussed on the Q3 call [5].
    • Why this probability: Broadcom raises its dividend every December, but a big buyback competes with financing needs (lowered from 70%).

Headwinds NOT in the forecasts

  • An AI-lab funding squeeze delays deployments, so FY28 AI revenue lands >20% below ~$230B — 30% chance within 36 months · Impact: High
    • What it is: Anthropic or OpenAI can't raise enough to pay for the 2027–28 gigawatts.
    • Why it's not priced in: consensus FY27 revenue (~$174B) already embeds the full ~$115B AI path [29], and management guides FY28 at ~$230B.
    • Sign to watch: a delayed or downsized Anthropic IPO; OpenAI's $30–50B financing failing to close; "secured"/"line of sight" wording softening.
    • Why this probability: both labs' revenue is growing fast, but a ~$350B two-year plan needs open capital markets every quarter.
  • Broadcom adds OpenAI guarantees on top of Anthropic's — 55% chance within 12 months · Impact: Medium
    • What it is: a second financing structure of $30–50B with a Broadcom backstop or residual guarantee.
    • Why it's not priced in: earnings models ignore contingent liabilities.
    • Sign to watch: an 8-K or 10-K note; Bloomberg/WSJ follow-ups.
    • Why this probability: talks are confirmed by two outlets, but they are "early", and Tan called OpenAI's finances "less clear" [5,19,20].
  • Rating-agency negative outlook or debt-like treatment of the guarantees — 30% chance within 24 months · Impact: Medium
    • What it is: S&P or Moody's adding guarantees to adjusted debt, or moving outlooks to negative.
    • Why it's not priced in: no 2026 agency action yet, while CDS and BofA credit research already moved [17,18].
    • Sign to watch: agency releases after the FY26 10-K (~mid-December).
    • Why this probability: the exposures are large relative to ~$60B of debt, but the cash flow is huge.
  • China formally or informally steers state data centres off Broadcom switches — 45% chance within 24 months · Impact: Low
    • What it is: SASAC guidance following its survey [21].
    • Why it's not priced in: no order has been issued yet.
    • Why this probability: it follows the pattern of Chinese guidance against Micron and Nvidia, but private hyperscalers are reportedly unaffected.
  • Google moves the next TPU flagship mainly to MediaTek — 25% chance within 36 months · Impact: Medium-High
    • What it is: MediaTek wins lead design on TPU v9.
    • Why it's not priced in: consensus assumes Broadcom stays Google's main partner.
    • Sign to watch: TPU v9 attribution; MediaTek's 2027 data-centre revenue tracking above $12–16B [27].
    • Why this probability: Broadcom just shipped first on 8i, but Google has every reason to keep a credible second source.
  • The Anthropic "supply chain risk" dispute widens beyond the Pentagon — 12% chance within 24 months · Impact: Medium
    • What it is: the label spreads to broader federal procurement, or to rules that pressure suppliers.
    • Why it's not priced in: Broadcom's new risk factor is new and barely discussed.
    • Why this probability: the courts are split and the label currently applies only to Defense work [38,39].
  • EU formal antitrust proceedings on VMware — 25% chance within 36 months · Impact: Low-Medium
    • Why this probability: the CISPE complaint and the information requests are live, but no formal case has opened in seven months [42].
  • Hock Tan leaves before FY2030 — 10% chance within 36 months · Impact: Medium-High
    • Why this probability: the award structure is a strong retention tie, so the remaining risk is mainly age and health.

Investment Risk Score: 52/100

Risk area Weight Score (0–100) Why
Business & competition 25% 44 The networking and execution moats held: TPU 8i shipped first. But MediaTek's 15–20% share target and Marvell's TPU-attach deal show customers building real second sources. Up 2 from 42.
Financial health & dilution 20% 46 Huge FCF, falling debt, no dilution. But the contingent stack keeps growing: $29B backstop, up-to-$42B notes, a partial guarantee on a $42B senior tranche, and OpenAI talks. The total is undisclosed and CDS widened. Up 6 from 40.
Execution (delivering the plan) 20% 53 Three AI beats in a row. But the per-GW gap, gross-margin misses versus promises and the packaging/HBM/power bottlenecks make FY28 harder to underwrite. Up 1 from 52.
Dependency & concentration 15% 74 One distributor is 50% of revenue, top five are ~55%, two AI labs become #1 and #2, the future #1 customer carries a federal risk label, ~95% of wafers come from TSMC, and one 74-year-old CEO runs it. Up 2 from 72.
Regulation, policy & geopolitics 10% 52 SASAC switch survey, unsigned Section 232 phase two, the new "supply chain risk" factor and EU VMware scrutiny. Up 2 from 50.
Sentiment & expectations 10% 52 The stock is ~26% below its high at ~18.6x FY27 EPS, with consensus already at the full FY27 AI path. It still swings 4–6% on single financing headlines. Down 3 from 55.

Overall = 0.25×44 + 0.20×46 + 0.20×53 + 0.15×74 + 0.10×52 + 0.10×52 = 52.3 → 52/100 (high).

The score barely moves: the engineering business looks at least as strong as three days ago and expectations have cooled. The financing layer is the area that worsened. Since the last dive, a Broadcom-guaranteed senior tranche entered syndication, OpenAI financing talks surfaced, and the 10-Q's new risk factors (AI "bubble", "supply chain risk", "customer-owned tooling") were read in full.

Conclusion

  • What kind of company this is:
    • A cash-rich chip-and-software conglomerate that has become the main custom-AI-chip and AI-network supplier, and now a partial financier, for the biggest AI spenders.
    • Owning it is a bet that frontier labs and hyperscalers deploy the gigawatts signed for 2027–28, that Broadcom keeps most of the custom work, and that the financing it arranges never has to be called.
  • The 2–3 things that will decide the outcome:
    1. Whether Anthropic (IPO) and OpenAI (new financing) can fund the 2027–28 build.
    2. How many of the ~30 GW announced actually ship by FY28. The content-per-GW gap says it's fewer than headlines suggest.
    3. Whether Broadcom stays first on Google's next TPU while MediaTek ramps.
  • Signposts confirming the bull case:
    • Anthropic's S-1 goes public and the IPO prices.
    • Q4 FY26 (expected ~2026-12-10, unconfirmed) shows AI revenue ≥ $21.7B and an FY27 outlook held or raised.
    • The FY26 10-K gives a full, netted contingent-exposure figure that is smaller than the headline sum.
    • Tomahawk 6/7 and CPO enter hyperscaler production.
  • Signposts that would break it:
    • The OpenAI financing closes with a large Broadcom guarantee.
    • A rating agency moves to a negative outlook.
    • "Secured" or "line of sight" softens, or purchase commitments are cut.
    • MediaTek is confirmed as lead on TPU v9.
    • Operating margin falls below ~62%.
    • VMware ARR growth drops below ~10%.
  • Balanced verdict:
    • Strongest bull: no company other than Nvidia has this much contracted, supply-backed AI growth, at a ~66% operating margin, with a 94%-gross-margin software annuity funding it. The engineering lead was re-proven when TPU 8i shipped ahead of a rival that started earlier.
    • Strongest bear: the growth rests increasingly on two loss-making labs. Broadcom is underwriting them through structures whose total size it won't state. Its own FY27–28 guides imply far fewer gigawatts than the headlines it uses to sell the story.
    • Interpretation: today the evidence leans toward FY27 being delivered: contracts, RPO, supply and three straight beats are hard data. FY28 is increasingly a capital-markets outcome, and the financing exposure is the variable that has worsened most in the last month.

Open questions & data gaps

  • Total contingent exposure. High importance, still unresolved. The overlap between the $29B backstop, the $42B notes, the partial guarantee on the $42B Class A tranche, and any OpenAI structure is undisclosed. The FY26 10-K (~mid-December) is the next chance.
  • Content per GW. High importance. There is no company bridge between $20–30B/GW and the ~$11.5B/GW implied by guidance, and no split between XPU, networking and rack content.
  • Credit ratings. Medium importance. No 2026 Moody's, S&P or Fitch action was found. The latest found (Moody's A3 positive, S&P A- positive) are stale (~2025).
  • TPU 8t/8i naming. Low-Medium importance. Earnings-call transcripts say Broadcom ships 8i and MediaTek 8t. TrendForce and others say the reverse. We follow the call.
  • Racks. Medium importance. Whether the financial partner buys racks from assemblers while Broadcom books only chips is an inference, not a disclosure. Rack margin: not found.
  • VMware. Medium importance. Not found: VMware-only revenue, the renewal cohort timing of 2024 deals, and churn.
  • Consensus. Medium importance. Not found: FY28 consensus and consensus gross margin (paywalled).
  • Transcripts. All call quotes are from third-party transcripts (Benzinga/Webull, Motley Fool), which contain speaker-name errors. Broadcom's IR site blocks automated access.
  • Price. The Oct 9 close was not confirmed from a primary market source (~$360–365 per secondary sources).
Sources
  1. Broadcom Form 10-K FY2025 — SEC EDGAR — 2025-12-18 — https://www.sec.gov/Archives/edgar/data/1730168/000173016825000121/avgo-20251102.htm
  2. Broadcom Form 10-Q, quarter ended 2026-08-02 — SEC EDGAR — 2026-09-10 — https://www.sec.gov/Archives/edgar/data/1730168/000173016826000080/avgo-20260802.htm
  3. Broadcom Q3 FY26 earnings press release (Ex. 99.1) — SEC EDGAR — 2026-09-02 — https://www.sec.gov/Archives/edgar/data/1730168/000173016826000076/avgo-08022026x8kxex99.htm
  4. Broadcom Q3 FY26 earnings call transcript — Benzinga via Webull — 2026-09-02 — https://www.webull.com/news/15511095937917952
  5. Broadcom Q3 FY26 earnings call transcript — Motley Fool — 2026-09-09 — https://www.fool.com/earnings/call-transcripts/2026/09/09/broadcom-avgo-q3-2026-earnings-call-transcript/
  6. Broadcom Q2 FY26 earnings call transcript — Motley Fool — 2026-06-03 — https://fool.com/earnings/call-transcripts/2026/06/03/broadcom-avgo-q2-2026-earnings-transcript/
  7. Broadcom Q1 FY26 earnings call transcript — Motley Fool via AOL — 2026-03-04 — https://www.aol.com/articles/broadcom-avgo-q1-2026-earnings-000708518.html
  8. Broadcom Q4 FY25 earnings call transcript — Motley Fool — 2025-12-12 — https://www.fool.com/earnings/call-transcripts/2025/12/12/broadcom-avgo-q4-2025-earnings-call-transcript/
  9. Broadcom DEF 14A proxy statement — SEC EDGAR — 2026-03-02 — https://www.sec.gov/Archives/edgar/data/1730168/000119312526085691/d49254ddef14a.htm
  10. Broadcom 8-K, CFO transition — SEC EDGAR — 2026-04-02 — https://www.sec.gov/Archives/edgar/data/1730168/000119312526140574/d109450d8k.htm
  11. Broadcom 8-K, Google long-term agreement and Anthropic 3.5 GW — SEC EDGAR — 2026-04-06 — https://www.sec.gov/Archives/edgar/data/1730168/000119312526144028/d87999d8k.htm
  12. Broadcom 8-K, Apple agreements through 2031 — SEC EDGAR — 2026-07-06 — https://www.sec.gov/Archives/edgar/data/1730168/000119312526295589/d84378d8k.htm
  13. Broadcom Form 4 filings, 2026-04-01 to 2026-09-25 — SEC EDGAR submissions feed — 2026-04 to 2026-09 — https://data.sec.gov/submissions/CIK0001730168.json
  14. Exclusive: Broadcom to lend Anthropic up to $42 billion to lease its chips, filing says — Reuters via The Star — 2026-10-01 — https://www.thestar.com.my/tech/tech-news/2026/10/01/exclusive-broadcom-to-lend-anthropic-up-to-42-billion-to-lease-its-chips-filing-says
  15. Broadcom seeks more than $60bn in debt to fund AI chips for Anthropic (relaying Bloomberg) — TheNextWeb — 2026-08-21 — https://thenextweb.com/news/broadcom-60bn-ai-chip-debt-anthropic
  16. Broadcom is Anthropic's chip supplier, lender, and debt guarantor: $60 billion syndication begins (relaying Bloomberg) — TechTimes — 2026-10-06 — https://www.techtimes.com/articles/328625/20261006/broadcom-anthropics-chip-supplier-lender-debt-guarantor-60-billion-syndication-begins.htm
  17. Broadcom credit risk soars on mega AI debt financing backstops — Bloomberg Government — 2026-08-24 — https://news.bgov.com/legal-ops-and-tech/broadcom-credit-risk-soars-on-mega-ai-debt-financing-backstops
  18. Bank of America downgrades Broadcom credit to Neutral (relay of BofA credit research) — Longport — 2026-08-10 — https://longportapp.cn/news/295504198
  19. Broadcom eyes US$30bil AI chip financing deal (Bloomberg) — The Star — 2026-10-09 — https://www.thestar.com.my/business/business-news/2026/10/09/broadcom-eyes-us30bil-ai-chip-financing-deal
  20. Broadcom seeks more than USD 50 billion (relaying WSJ/Business Times) — ad-hoc-news — 2026-10-08 — https://www.ad-hoc-news.de/boerse/news/corporate-news/broadcom-seeks-more-than-usd-50-billion-broadcom-stock-sits-26-17-percent-below-high/70279549
  21. China surveys Broadcom switch use in state data centers, FT reports — Reuters via The Standard — 2026-09-23 — https://www.thestandard.com.hk/innovation/article/343583/China-surveys-Broadcom-switch-use-in-state-data-centers-FT-reports
  22. Broadcom data center network switches under state microscope in China — SDxCentral — 2026-09-24 — https://www.sdxcentral.com/news/broadcom-data-center-network-switches-under-state-microscope-in-china-report/
  23. Marvell, AMD reportedly shake up Google TPU race — TrendForce — 2026-08-20 — https://www.trendforce.com/news/2026/08/20/news-marvell-amd-reportedly-shake-up-google-tpu-race-putting-broadcom-mediatek-under-pressure/
  24. Marvell 8-K, Google agreement and warrant — SEC EDGAR — 2026-08 — https://www.sec.gov/Archives/edgar/data/0001835632/000119312526356217/d412696d8k.htm
  25. Marvell Investor Day 2026 transcript — StockAnalysis — 2026-10-06 — https://stockanalysis.com/stocks/mrvl/transcripts/724088-investor-day-2026/
  26. Marvell Investor Day 2026 targets $80B FY2031 revenue — Futurum — 2026-10-09 — https://futurumgroup.com/insights/marvell-investor-day-2026-targets-80b-fy-2031-revenue-with-custom-silicon-and-optics/
  27. MediaTek Q2 2026 prepared remarks — MediaTek IR — 2026-07-31 — https://www.mediatek.com/hubfs/MediaTek%20Assets/Pdfs/Quarterly%20Earnings%20Release/2026/Quarterly%20Earnings%20Release-2026Q2/Prepared%20remarks.pdf (Bernstein MediaTek forecast via Webull — 2026 — https://www.webull.com/news/15580774049924096)
  28. Broadcom just got downgraded by a major bank to a $437 target (Macquarie) — TIKR — 2026-06 — https://www.tikr.com/blog/broadcom-just-got-downgraded-by-a-major-bank-to-a-437-target-is-the-bear-case-right
  29. AVGO analyst forecast and estimates (S&P Global data) — StockAnalysis — 2026-10-09 — https://stockanalysis.com/stocks/avgo/forecast/
  30. Ethernet extends lead in AI scale-out networks despite strong InfiniBand rebound — Dell'Oro Group — 2026-06-02 — https://www.delloro.com/news/ethernet-extends-lead-in-ai-scale-out-networks-despite-strong-infiniband-rebound/
  31. Co-packaged optics press release — TrendForce — 2026-07-27 — https://trendforce.com/presscenter/news/20260727-13151.html
  32. TSMC Q2 2026 earnings call transcript — Investing.com — 2026-07-16 — https://www.investing.com/news/transcripts/earnings-call-transcript-tsmc-lifts-2026-outlook-as-ai-demand-stays-hot-in-q2-2026-93CH-4794777
  33. TrendForce: 2027 HBM contract prices may rise 70–140% — Herald Corp — 2026-08/09 — https://biz.heraldcorp.com/article/10852833
  34. Nvidia Form 10-Q, quarter ended 2026-07-26 — SEC EDGAR — 2026-08-26 — https://www.sec.gov/Archives/edgar/data/0001045810/000104581026000075/nvda-20260726.htm
  35. Nvidia Q2 FY27 earnings call transcript — Webull — 2026-08-26 — https://www.webull.com/news/15497781742806016
  36. Follow the CAPEX: Q2 2026 scoreboard — Platformonomics — 2026-07 — https://platformonomics.com/2026/07/follow-the-capex-q2-2026-scoreboard/
  37. Anthropic chose Nasdaq for October IPO — TheNextWeb — 2026-09-13 — https://thenextweb.com/news/anthropic-ipo-nasdaq-october-listing
  38. Federal court says Pentagon can label Anthropic a supply chain risk (AP) — CP24 — 2026-09-25 — https://www.cp24.com/news/world/2026/09/25/federal-court-says-pentagon-can-label-anthropic-a-supply-chain-risk/ (also TheNextWeb, "US appeals court upholds Pentagon's supply chain risk label on Anthropic" — 2026-09 — https://thenextweb.com/news/anthropic-pentagon-supply-chain-risk-appeals-court-ruling)
  39. Judge rules the Pentagon's supply chain risk label for Anthropic unlawful (CNN) — KEYT — 2026-08-27 — https://keyt.com/news/money-and-business/cnn-business-consumer/2026/08/27/judge-rules-the-pentagons-supply-chain-risk-label-for-anthropic-unlawful/
  40. It's official: the Pentagon has labeled Anthropic a supply chain risk — TechCrunch — 2026-03-05 — https://techcrunch.com/2026/03/05/its-official-the-pentagon-has-labeled-anthropic-a-supply-chain-risk/
  41. Why Wall Street thinks Broadcom is a buy (Susquehanna note on Anthropic rack revenue, relayed) — Yahoo Finance — date not found — https://finance.yahoo.com/markets/stocks/articles/why-wall-street-thinks-buy-143722193.html
  42. CISPE files competition complaint against Broadcom — CISPE — 2026-03-19 — https://www.cispe.cloud/?p=25724
  43. VMware cost trap 2026: alternatives Proxmox, Nutanix (Gartner, Nutanix figures relayed) — cloudmagazin — 2026-03-18 — https://www.cloudmagazin.com/en/2026/03/18/vmware-broadcom-kostenfalle-2026-alternativen-proxmox-nutanix/
  44. Lutnick says upcoming Section 232 chip tariffs will include onshoring incentives — International Trade Today — 2026-09-03 — https://internationaltradetoday.com/article/2026/09/03/lutnick-says-upcoming-section-232-chip-tariffs-will-include-onshoring-incentives-2609030051
  45. Administration policies on advanced AI chips codified — Mayer Brown — 2026-01-16 — https://www.mayerbrown.com/en/insights/publications/2026/01/administration-policies-on-advanced-ai-chips-codified
  46. New York enacts first-in-the-nation statewide data center moratorium — Sheppard Mullin — 2026-07 — https://www.sheppard.com/insights/blogs/new-york-enacts-first-in-the-nation-statewide-data-center-moratorium
  47. Broadcom lending Anthropic up to $42 billion in yet another AI circular financing deal — IEEE ComSoc Techblog — 2026-10-01 — https://techblog.comsoc.org/2026/10/01/broadcom-lending-anthropic-up-to-42-billion-in-yet-another-ai-circular-financing-deal/
  48. Custom AI ASICs examined, from Broadcom to MTIA (Counterpoint share data) — Tom's Hardware — 2026-05 — https://www.tomshardware.com/tech-industry/semiconductors/custom-ai-asics-examined-from-broadcom-to-mtia

Notes: "secondary aggregator relay, not re-checked" marks stock-price moves and analyst actions relayed by TradingKey or The Fly via Yahoo. Scores are model judgements, not facts. This is research, not advice.

All AVGO Bulls vs Bears reports
Disclaimer. AI-generated research and podcast (Claude by Anthropic, voices by Google Gemini) for educational purposes — not financial advice and not a recommendation to buy or sell. Willem (MR.INVEST) owns this stock. Check the company’s own filings before making any decision.
← All Bulls vs Bears reports