We value AVGO at $318 a share. The report's $329.29 price (Oct 26, 2026) was 4% above fair value: Hold.
- Five-year return
- 9.7%
- a year, at the price
What would change the view
- FY27 AI revenue guide at or above $115B and the FY28 path reaffirmed without new Broadcom-funded lending: move toward the bull case
- Customer financing or guarantees on Broadcom's balance sheet beyond the first tranche of XPV (the financing vehicle set up with Apollo and Blackstone): move toward the bear case
- A frontier-lab customer fails to fund its gigawatt commitment: bear case
- V3: Broadcom's notes to Anthropic step from 9% to 13% paid in kind (issued while Anthropic misses its revenue conditions, S-1 p. 167), or draws stop because the conditions are missed for a sustained period: move toward the bear case, whose AI revenue falls to $110B by FY31
- Broadcom pays any amount under the XPV Backstop, or names the XPV customer as Anthropic with the Backstop above $29B: move toward the bear case
Catalysts
- Q4 FY26 results (Broadcom's fiscal year ends in early November): the FY27 AI revenue guide against about $115B, and the FY26 10-K's note on the XPV Backstop and any Anthropic notes issued
- Anthropic's IPO and the first quarters after it: notes can be issued only after the draw period starts and are not expected before the offering (S-1 p. 167); any issue, and its rate (9% or 13%), shows in Anthropic's and Broadcom's next filings
We keep Broadcom a Hold at $318 a share, up $1 from $317, against $329.29. Anthropic's S-1 and Broadcom's own 10-Q now give the terms of the financing Broadcom provides around its TPU leases to Anthropic. Our 2026-10-15 report relied on press accounts of those terms. The loan is small against Broadcom's value. The bigger risk is in the revenue: by our inference, the customer Broadcom finances rents about a fifth of the AI chips in our central case.
What the filings show
The notes (fact, S-1 p. 167-168, confirmed in part by Broadcom's 10-Q note 10):
- Size and purpose. Anthropic may issue Broadcom up to $42.0B of convertible notes, "if needed". The money may be used only to pay the quarterly rent on Anthropic's 2026 lease of TPU equipment from Broadcom. None has been issued, and Anthropic expects none before its IPO.
- Interest. 9.00% a year, paid in kind, that is, added to the principal instead of paid in cash. Notes issued while Anthropic misses certain revenue performance conditions carry 13.00%.
- Conversion. At Broadcom's choice, into Anthropic Class N stock at 90% of the 30-day average Class A price. Notes issued during a miss convert at 85%.
- Conditions. The revenue-growth conditions are "significantly below" Anthropic's actual growth in every quarter of 2025 and 2026. A sustained miss ends Anthropic's right to issue notes.
- Security. Secured on a junior basis by the same equipment that secures the lease.
- Maturity and default. Five years from the start of the draw period for the first $24.0B. The notes accelerate on a lease payment default or a cross-default.
The leases (fact, S-1 p. 58, 75, 172-173): five-year master equipment leases with Broadcom, about $161.2B of rent excluding interest, of which about $125.2B is the March 2026 lease. They are non-cancelable except on default, guaranteed by Anthropic, PBC, and secured by the equipment. Anthropic has posted restricted cash for Broadcom. The S-1 also says Broadcom intends to assign the 2026 lease to a "designated financial partner" (p. 58), and that in calendar Q2 2026 Broadcom assigned part of one lease to a third party.
Broadcom's own disclosure (fact, 10-Q for the quarter to 2026-08-02, note 10 and MD&A). Broadcom arranged for a financial partner to buy AI racks built on its chips and take over the related leases "with a customer", which it does not name. That is the first $35B tranche of the "AI XPV" platform, for more than 1 GW. Broadcom backstops the customer's lease payments over the five-year terms. If the customer defaults, Broadcom pays 85% of the amount owed less what the racks sell for. The maximum is about $29B, undiscounted, once all racks are deployed. Broadcom puts the backstop's fair value as "not material" and has paid nothing. The 10-Q also mentions the $42B of notes, without their rate.
Inference: the XPV customer is probably Anthropic. The S-1's assignment of part of a lease in calendar Q2 matches the 10-Q's first tranche in June. Neither filing says so.
Meta (fact, FWP 2026-10-26): NVIDIA, AMD and Broadcom, each on its own account, indicated non-binding interest in buying up to $5.0B of Meta stock in total in its offering. Broadcom's share is not stated. We count none of it.
No Broadcom 8-K has been filed since the 10-Q (EDGAR, read 2026-10-26). Broadcom has not disclosed the note terms itself; they come from Anthropic's filing.
What changes in our value
| Step | Bear | Central | Bull |
|---|---|---|---|
| 2026-10-15 report | $165.33 | $316.67 | $630.94 |
| Valuation date moves to 2026-10-26 | +$0.50 | +$0.95 | +$1.89 |
| Bear: Anthropic issues $24B of notes at 13%, worth half at exit | -$1.38 | ||
| This revision | $164.45 | $317.62 | $632.83 |
Source: bridge.py, bridge.txt. The other financing choices (our assumptions, assumptions.md):
- Central and optimistic: no notes drawn. Anthropic held $131.7B of cash and short-term investments at 2026-06-30 (S-1 p. 173). It says it is "not dependent" on this facility (S-1 p. 75).
- Pessimistic: a miss. Anthropic misses its revenue conditions, issues $12B of notes at 13% in each of FY28 and FY29, then loses the right to issue more. The notes are worth half their accrued value at exit: they rank behind the lease on the same racks and convert into shares of a company whose growth has failed.
- The backstop keeps the amounts our earlier report gave the guarantees: $3B expected loss centrally, $15B in the pessimistic case. The 10-Q confirms the $29B maximum that report took from an analyst's question on the call.
Sensitivities (output.txt):
- Central, all $42B drawn at 9% and repaid in full: +$0.5, because the notes earn more than Broadcom's cash.
- Central, all $42B drawn at 13% and worth half: -$2.5.
- The full $29B backstop lost centrally: -$3.2.
- Pessimistic case with all $42B drawn and worth nothing, plus the full $29B backstop lost: -$6.1.
How Broadcom books it (inference; Broadcom has not said)
If notes are issued, we expect Broadcom to carry them as debt securities. The paid-in-kind interest would be accrued in other income, non-cash until the notes are repaid or converted. The backstop is a guarantee held at fair value. Neither touches operating income, so neither is in our operating profit. We count the notes only at exit, and pay tax on the interest as it accrues.
The loop, and the wrong-way risk
Revenue that depends on a customer Broadcom finances. Anthropic's lease rent of $161.2B over five years is about 19% of our central FY27-31 AI revenue of $833B, and the March lease alone 15% (output.txt). That assumes the rent excluding interest roughly equals the equipment's price, which is our inference. TPUs that reach Anthropic through Google come on top. Broadcom's remaining performance obligations were $179.2B at 2026-08-02, including a long-term custom-accelerator contract signed in the quarter to 2026-05-03 (fact, 10-Q). If that contract is the March lease (our inference), the customer Broadcom finances is about 70% of its contracted backlog.
Management's own plan for Anthropic is larger still: 1 GW in 2026, 5 GW in 2027 and 10 GW more in 2028, at $20-30B of content a gigawatt (management claims, Q3 call, carried from our 2026-10-15 report). At face value that is more than our whole central AI revenue for FY27-28. That cannot all be Broadcom's revenue: some of those gigawatts must reach Anthropic as Google TPUs, or count content that is not Broadcom's. We keep our path.
Wrong-way risk. If Anthropic's growth slips, four things fall at once. Rack orders fall. Notes are issued at the higher 13% exactly when Anthropic is weaker. The racks that secure both the lease and the notes lose resale value in an AI downturn. And the backstop pays the gap between 85% of the lease balance and that lower resale value. As an illustration (not a ceiling), the direct credit exposure is about $90B undiscounted: $42B of notes accruing at 13% for three years plus the $29B backstop, or $18 a share. Five years of accrual would make it about $106B. Later XPV backstops are not capped. And if the XPV customer is Anthropic, the notes pay the rent the backstop covers, so adding the two overstates the overlap. The revenue exposure is larger, and it is what our pessimistic case already prices: AI revenue falls to $110B in FY31 against $190B centrally. That is why the pessimistic case is worth half the central, and the notes only $1.38 of it.
One partial protection is written into the notes (inference from S-1 p. 167-168). Conversion at 85-90% of the 30-day average price gives Broadcom shares worth more than the note while the share price holds near that average. In a falling market, the bear case's world, the spot price can sit more than 10-15% below the average, and conversion is then worth less than the note. And a sustained miss ends the draws, which caps how much Broadcom lends into a decline. Against that, Broadcom's lending is limited by contract, but its backstops are not: the 10-Q describes XPV as meant to enable more than 20 GW for frontier labs through 2028, and only the first $35B tranche is disclosed.
The view
At $329.29 the price needs an exit at 18.8 times NOPAT, against our 18, or AI revenue 4.8% above our path (unblinded-output.txt). That is close enough to our central case that we have no edge: Hold. The Street's average target of $514 rests on revenue of $552B in FY30 (market data, 33 analysts), which our 2026-10-15 report ran through the model at about $758. Nothing this week supports that path. The filings show more clearly than before that part of Broadcom's growth is financed on its own balance sheet or backstopped by it.
Tripwires
- V3. Broadcom's notes to Anthropic step from 9% to 13% paid in kind. That happens when notes are issued while Anthropic misses its revenue conditions (S-1 p. 167). Also V3: draws stop because the conditions are missed for a sustained period. Either moves us toward the pessimistic case: FY31 AI revenue of $110B and a value of $164.
- Broadcom pays anything under the XPV backstop, or names Anthropic as the XPV customer with backstops above $29B: toward the pessimistic case.
- The FY26 10-K (due after the 2026-12-09 results) is the first annual filing to carry the backstop. It should say whether later XPV tranches carry backstops, and how Broadcom accounts for any notes issued.
The 2026-10-15 report's other triggers stand: an FY27 AI guide at or above $115B with FY28 reaffirmed and no new Broadcom-funded lending moves us toward the optimistic case; a frontier lab failing to fund its gigawatts moves us to the pessimistic one.
Limits
- This revision keeps the 2026-10-15 valuation engine: a judged 18x exit multiple and a 10.5% cost of equity. The fund's shared method values every other report at the mean of a long fade and an outside view, with a cost of equity from peer betas. That method would cost AMD-scale work and could move the value far more than this week's filings do. It is not done here, so AVGO is not comparable with the reports on the shared method.
- The XPV customer, the RPO contract and the rent-to-revenue mapping are our inferences, not filing statements.
Revision history
| Report | View | Pessimistic | Central | Optimistic | Central change | Price | Blinding |
|---|---|---|---|---|---|---|---|
| 2026-10-15-avgo | hold | 165 | 317 | 631 | first | 340.34 | blind |
| 2026-10-26-avgo | hold | 164 | 318 | 633 | +1 | 329.29 | unblinded |
Sources
- Anthropic, PBC, Form S-1, 2026-10-26, p. 58, 75, 166-168, 172-173 (data/anthropic-s1-paged.txt; the fund's read: 2026-10-26-anthropic-s1).
- Broadcom 10-Q for the quarter to 2026-08-02, filed 2026-09-10, note 10, revenue note and MD&A (data/10q-q3-fy26.htm); EDGAR submissions read 2026-10-26; 8-Ks 2026-06-18 and 2026-07-06.
- Meta Platforms free-writing prospectus, 2026-10-26 (data/meta-fwp-2026-10-26.htm).
- Robinhood quote, 2026-10-26 12:57:08 ET; Street targets and estimates read 2026-10-26.
- The 2026-10-15 report: report.
This work was done with Claude, a model made by Anthropic, a counterparty in the agreements discussed: a conflict of interest.
Details
- Report date
- Oct 26, 2026
- Price source
- Robinhood last trade 12:57:08 ET 2026-10-26 (regular session; official close 2026-10-23 $321.68), Oct 26, 2026, 9:57 AM PDT
- Bear / bull
- $164 / $633
- Exit price
- $473 (on Nov 2, 2031)
- Confidence
- 40%
- Cost of equity
- 10.5%
- Street
- average $514, median $520, low $350, high $600; 11 targets in the last quarter; Street consensus targets, Oct 26, 2026, 10:04 AM PDT
- Blind step
- Unblinded; the model was committed in 2ddc043 before the record
- Method
- As of Oct 26, 2026, built without: Mature return holds own capital + 1 formula update
We keep Broadcom a Hold at a fair value of $318 (was $317), against $329.29. Anthropic's S-1 gives the terms of the up-to-$42B of convertible notes Anthropic may issue to Broadcom, solely to pay rent on about $161.2B of TPU equipment leases: 9% interest paid in kind, 13% and conversion at 85% of the share price if Anthropic misses revenue conditions, junior to the lease on the same racks. Broadcom's 10-Q adds a backstop of up to about $29B on the first $35B XPV tranche for an unnamed customer. The direct credit risk is small: even the notes and the backstop both lost in our pessimistic case take $6 a share. The larger exposure is revenue: Anthropic's leases are about 19% of our central 2027-31 AI revenue, and the notes would be issued at the higher rate exactly when Anthropic is weaker, when its racks, its shares and its orders fall together. Our central case assumes no draws (Anthropic held $131.7B in cash); the pessimistic case draws $24B at 13%, worth half (-$1.38). Tripwire: the notes step to 13% or draws stop on missed conditions.
As of 2026-10-26 · unblinded: a revision of the 2026-10-15 report, which read the price after its blind commit; this one changes only inputs dated by the 2026-10-26 filings and Broadcom's 10-Q. Model committed before this report (2ddc043). The 2026-10-15 report ($317, Hold) stays on the record.
- Fair value today
- $318 per diluted share (cost of equity 10.5%). Weighting the pessimistic, central and optimistic cases 25/50/25 (our weights) gives $358
- Price
- $329.29, Robinhood last trade 2026-10-26 12:57:08 ET (official close 2026-10-23 $321.68) → 4% above fair value
- Five-year exit price
- $473 at the end of fiscal 2031 (about 2031-11-02), 18 times FY31 operating profit after tax and stock pay (NOPAT)
- Five-year return
- +9.7% a year from $329.29, dividends included
- Purchase price
- $298 to earn a 12% hurdle (a placeholder; Dong has not set one)
- Bear / bull today
- $164 / $633: the pessimistic and optimistic cases, each one coherent world
- Confidence
- 40% (subjective, not calibrated): chance the FY31 value lands within ±20% of $473, carried from the 2026-10-15 report
- Moat
- Custom AI chip design and AI networking at scale for six hyperscale buyers, plus VMware's installed base
- Market expects
- About our central case: an exit at 18.8 times NOPAT (ours 18), or AI revenue 4.8% above our path (FY31 $199B, ours $190B) (unblinded-output.txt)
- Key driver
- FY31 AI revenue: each $30B is worth about $32 a share; each 2 turns of exit multiple about $28 (output.txt grid). The Anthropic notes move the pessimistic case $1.38
- Street
- Average 12-month target $514 (median $520, range $350-600); 11 targets in the last quarter averaging $510 (market data, read 2026-10-26). An expectations reference, not validation