AVGO — Broadcom
The earnings kept moving after the stock stopped. The portfolio is still the annoying part.
Broadcom combines custom AI accelerators and networking with an extremely profitable infrastructure-software business built around VMware. At the current underwriting, standalone returns are around my hurdle rather than decisively above it. I already have a large direct AI-infrastructure sleeve, so another correlated allocation has to clear a higher bar.
Broadcom is a much more interesting investment than the chart alone explains.
Why I care about this one
Broadcom is a custom-AI-silicon and networking powerhouse attached to an extremely profitable infrastructure-software business.
The semiconductor side designs XPUs, Ethernet switching, optical interconnect and other critical data-center components. VMware gives it a second earnings engine with software margins that are frankly ridiculous.
The company is generating enough cash to pay down debt, buy stock and pay a dividend while the AI business scales.
The unresolved issue is no longer “is Broadcom good?” It is how much of the future AI build is already underwritten, how much financing risk sits behind that demand, and whether I need another position tied to the same infrastructure cycle.
Q3 FY2026 revenue reached $29.6B, up 86%. Semiconductor Solutions grew 127% to $20.8B. Infrastructure Software grew 29% to $8.8B. (Q3 Form 10-Q)
AI semiconductor revenue alone was $16.7B, up 221% year over year and 54% sequentially.
Broadcom expects $21.7B of AI semiconductor revenue in Q4 and roughly $58B for FY2026. Then management sees approximately $115B in FY2027 and $230B in FY2028.
Those are management outlooks, not my scenario assumptions.
They are also large enough that the five-year model has to let growth slow sharply after the next two years rather than pretending the current doubling cadence continues forever.
The company case is almost annoyingly easy. Q3 revenue grew 86%, AI semiconductor revenue grew 221%, management sees about $115B of AI semiconductor revenue in FY2027 and $230B in FY2028, and VMware gives the company a second high-margin cash engine. The harder questions are concentration and financing: the top five end customers are about half the business, Anthropic could become the largest compute customer, and Broadcom has agreed to provide up to $42B of financing tied to that infrastructure build.
Q3 FY2026 / Q4 guide
Primary sources: Q3 FY2026 results ↗ · Q3 FY2026 10-Q — SEC ↗ · Q3 FY2026 earnings-call transcript ↗ · Reuters — Broadcom / Anthropic financing ↗
Q3 FY2026 revenue reached $29.6B, up 86%, with AI semiconductor revenue up 221%. Management projects a substantial AI supply ramp through FY2028; the up-to-$42B Anthropic financing agreement adds counterparty and capital-risk questions. At roughly $624 FY2031 Base value, the standalone return is near my hurdle, while AI-infrastructure concentration still blocks new capital.
The $115B FY2027 and $230B FY2028 AI revenue path converts without customer-financing stress, VMware keeps producing exceptional software economics, customer diversity improves and repurchases begin to create clear per-share shrink.
AI deployments slip materially, Anthropic or another concentrated customer becomes a financing/counterparty problem, custom silicon loses share, software economics weaken, gross/operating margin compresses faster than modeled, or Broadcom increasingly uses its balance sheet to manufacture demand.
Why I passed
At the Oct. 6 reference price, the FY2031 Base value of roughly $624 implies around 11% annualized. Quality remains high, but the standalone return is near the hurdle and portfolio overlap still blocks capital.
A no-allocation decision is dated, not a permanent ban. New price, new evidence or a different portfolio context can reopen the work.
AI custom silicon/networking exposure was already represented by MRVL, ANET, MU and SKHY. The company could be phenomenal and still lose the next-dollar contest.
At $355.14, Broadcom's AI growth and VMware economics looked attractive, but I held zero shares because direct AI-infrastructure exposure was already roughly 18% of invested assets.
At the Oct. 6 reference price, the FY2031 Base value of roughly $624 implies around 11% annualized. Quality remains high, but the standalone return is near the hurdle and portfolio overlap still blocks capital.
Valuation & scenario math
The AI growth is increasingly concentrated among a small number of very large customers.
Broadcom’s 10-Q says its top five end customers represented roughly 55% of Q3 revenue. One semiconductor distributor alone represented 50% of reported revenue. (Q3 Form 10-Q)
Then there is Anthropic.
Reuters reported on Oct. 1 that Broadcom agreed to provide Anthropic with up to $42B of financing connected to infrastructure spending and TPU capacity. The instruments may be convertible into Anthropic equity, and Anthropic is expected to become Broadcom’s largest chip-design customer in 2027. (Reuters)
That may work beautifully.
It also means the bull case cannot simply count the revenue and ignore the capital structure supporting it.
The five-year model starts from roughly $105.9B of FY2026 revenue. Bear / Base / Bull assume 14% / 24% / 30% annual revenue growth, 46% / 55% / 59% terminal operating margins, +1% / flat / -0.5% annual share change and 15x / 22x / 26x terminal P/E. Those assumptions imply roughly $222 / $624 / $1,027. At the Oct. 6 reference close, Base offers around 11% annualized, with material sensitivity to the 2031 earnings multiple.
“My current view” is the output of my dated valuation framework, not a magical buy instruction. A BUY still has to beat the expected return of what I already own and the index before it gets actual money.
12% / 15% / 18% required return to the five-year base case
| Band | Buy price | Model base center | Annualized return at entry |
|---|---|---|---|
| Starter | $353 | $622 | +12.0% |
| Normal | $310 | $622 | +15.0% |
| Aggressive | $273 | $622 | +18.0% |
The zone label uses the live quote when available; the model itself is anchored to the Oct 6, 2026 reference close at $376. The dollar bands come from the active base case. AVGO is already inside the normal standalone return zone. A buy band does not override the portfolio-overlap constraint.
The model starts with FY2026 revenue of roughly $105.9B, built from reported nine-month revenue plus the Q4 guide. It makes revenue growth, endpoint operating margin, share-count change and terminal P/E explicit. The 82% operating-income-to-net-income conversion is my simplifying assumption. The main point is to force the FY2027/FY2028 AI ramp to earn its way into per-share value rather than capitalizing the headline.
| Revenue today | $105.9B |
| Annual revenue growth | 24% |
| Revenue by Oct 2, 2031 | ~$309.5B |
| Endpoint operating margin | 55% |
| Modeled net income | ~$139.6B |
| Share-count path | flat |
| Endpoint diluted shares | ~4.94B |
| Endpoint EPS | ~$28 |
| Terminal P/E | 22x |
| Implied base price | ~$622/share |
The bridge is revenue → operating margin → modeled after-interest/tax income → share count → EPS → terminal multiple. The 82% conversion from operating income to net income is my simplifying assumption, not company guidance.
| Case | What has to be true | Target range | Total return* | Annualized return* |
|---|---|---|---|---|
| Bear | 14% revenue growth · 46% endpoint operating margin · 1% annual share growth · 15x terminal P/E. The AI ramp remains real but lands far below the current management trajectory, customer concentration and financing risk rise, mix compresses margins and dilution returns. | $206–$237 center ~$222 | -40.9% | -10.0% |
| Base | 24% revenue growth · 55% endpoint operating margin · flat share count · 22x terminal P/E. Broadcom compounds from the FY2026 revenue anchor without using management's FY2028 AI target as the center of the model. AI remains exceptional, VMware stays highly profitable, margins normalize below today's level and the share count is roughly flat. | $578–$665 center ~$622 | +65.5% | +10.6% |
| Bull | 30% revenue growth · 59% endpoint operating margin · 1% annual share shrink · 26x terminal P/E. Custom accelerators and networking scale across multiple frontier customers, VMware sustains exceptional economics, customer financing performs cleanly and repurchases begin to create modest per-share shrink. | $952–$1,095 center ~$1,023 | +172.3% | +22.2% |
*The static returns use the dated reference price. The annualized figure on the live page also refreshes from the current quote. For shorter model horizons, annualizing can make the number look ridiculous, so I care about total return and the underlying milestones at least as much.
Capital allocation
Great capital engine. The new financing loop deserves real scrutiny.
Custom AI silicon, networking and VMware show exceptional conversion of capital into growth and cash. FCF is large enough to support debt reduction, dividends and repurchases. I cut deployment for the up-to-$42B Anthropic financing exposure because customer financing can blur the quality of demand and put more balance-sheet risk behind a concentrated revenue base. SBC also keeps shareholder alignment below full credit.
Stage-relative, not some objective rating-agency grade. The number is there to force the judgment into the open. See the rubric →
What management is actually doing with the cash
Broadcom has an unusually strong record of turning R&D and giant acquisitions into cash. VMware is the current proof. The new question is customer financing: Anthropic's IPO filing disclosed up to $42B of Broadcom financing tied to infrastructure capacity. That can support very valuable demand, but it also mixes capital deployment, counterparty concentration and product sales in a way I want to watch closely.
What would make me reconsider
At the end of Q3, Broadcom had about $24.0B of cash and $59.4B of debt.
Free cash flow was $13.7B in the quarter. Broadcom paid down $5.6B of long-term debt in Q3, while the first nine months also included meaningful dividends and repurchases. (Q3 Form 10-Q)
So leverage is not the thesis problem it could have been after VMware.
Stock compensation is the small leak I keep watching. The company recorded roughly $6.3B of SBC through the first nine months of FY2026, and diluted shares are not yet shrinking as cleanly as the buyback headline might suggest.
What can break this
Q3 infrastructure-software revenue was $8.8B. Management said VMware ARR grew 15% and the segment produced roughly 84% non-GAAP operating margin. (Q3 call transcript)
That software engine is important for two reasons.
First, it throws off an absurd amount of cash.
Second, it means AVGO is not simply another pure semiconductor bet. The company still depends heavily on the same AI infrastructure spending cycle I already own elsewhere, but the software mix gives it a different earnings shape from MU, SK hynix, MRVL or ANET.
Different is not the same as uncorrelated.
I use a five-year operating model rather than one forward P/E.
The model starts around $105.9B of FY2026 revenue: reported nine-month revenue plus the Q4 guide. Management’s FY2028 AI outlook is a useful comparison, not a guarantee that the whole ramp flows through to 2031 earnings.
Base case:
- revenue compounds 24% annually to roughly $310B in 2031;
- operating margin settles at 55%;
- share count stays flat;
- 82% of operating income converts to modeled net income;
- terminal P/E is 22x.
That produces roughly $624/share.
Bear is roughly $222.
Bull is roughly $1,027.
At the Oct. 6 reference close, Base is around 11% annualized. Keep the Base operating outcome but use 20x instead of 22x and it drops toward 9%.
Even at 24% revenue growth, a 55% terminal margin and flat shares, the return is sensitive to what investors will pay for the mature business. The business can be phenomenal without the stock being a bargain.