Haverhill Financial Research · October 10, 2026
Summary: Haverhill price target $430 (12 months)
Broadcom has become the main supplier of custom AI chips to the largest tech companies, and the share price has not caught up with what that does to earnings. At $361.54 the stock trades at about 21 times our fiscal 2027 earnings estimate. Our 12-month price target of $430 (through October 2027) sits about 19% above the current price, so we rate the shares Buy.
Broadcom filings through the Q3 FY2026 results release; Haverhill Financial Research model; price as of Nasdaq close Oct 9, 2026
AI is now most of the business. AI chips were 56% of revenue in fiscal Q3 and are guided to about 62% in Q4, growing more than 200% a year.
The demand is contracted, not hoped for. Broadcom reported $164.6 billion of signed future revenue in May, about 30% of it due within a year.
Our view: the growth is real and the price is reasonable for it. The risks are a few very large buyers and thinner margins as Broadcom ships whole AI systems rather than just chips.
Investment thesis in brief
Our thesis: the custom AI chip business is growing faster than the market gives it credit for, and Broadcom turns almost half of every revenue dollar into free cash.
What we like. The largest AI buyers want chips designed around their own software, and Broadcom is the company they hire to build them. Those projects take years, which locks in the customer for each generation. The rest of the company, mostly VMware software, throws off steady cash that funds the dividend and pays down the debt from buying VMware.
What the market may be underrating. Broadcom's AI revenue went from $4.1 billion in fiscal Q1 2025 to a guided $21.7 billion in Q4 2026, more than five times in under two years. Investors still value the stock at about 21 times our 2027 earnings, below many slower-growing chip companies.
What holds us back. A handful of buyers account for most of the AI revenue, so one customer pausing would show up fast. Margins are also slipping as Broadcom ships full server racks with lower-margin parts inside.
How does the business actually make money?
Broadcom sells two things: chips that move and process data inside large data centers, and software that runs corporate data centers. Chips were 70% of revenue in fiscal Q3 2026 and software 30%.
Who pays, and for what. The largest AI builders buy custom accelerators (XPUs) designed to their spec, plus the networking chips that link thousands of them. That was 56% of fiscal Q3 2026 revenue, and each chip takes years to co-design, so switching suppliers means starting over on a new generation. Phone, storage and broadband makers buy wireless, storage, broadband and industrial chips, about 14% of revenue, held in place by long design cycles and few qualified suppliers. Large companies and governments buy VMware Cloud Foundation, mainframe and security software, the remaining 30%; VMware runs inside most corporate data centers, and replacing it is a multi-year project.
Companies publicly tied to Broadcom's custom AI chips include Google, Meta, ByteDance, OpenAI and Anthropic. Broadcom does not break out revenue by customer.
Broadcom quarterly results releases, Q1 FY2025 to Q3 FY2026, and Q4 FY2026 guidance
Why the money is sticky. At May 3, 2026 Broadcom had $164.6 billion of firm, multi-year contracts not yet billed, and expected about 30% of it, roughly $49 billion, within 12 months. In the same quarter it signed a long-term contract for custom AI accelerators. After buying VMware in 2023, Broadcom moved its customers from one-time licenses to subscriptions, which turned software into recurring revenue.
What do the numbers say?
Revenue grew 86% in a year to $29.6 billion in fiscal Q3, with AI chip revenue up 221% to $16.7 billion. Profits more than tripled, and Broadcom converted 46% of revenue into free cash. The one soft spot is gross margin, which is falling as AI systems take a bigger share.
Three things investors should know.
Growth is accelerating, not slowing. Revenue grew 29% year over year in fiscal Q1, 48% in Q2 and 86% in Q3, and Q4 is guided to 93%.
Margins are trading down for scale. Adjusted gross margin fell from 78.4% to 75.0% as Broadcom ships more complete systems. Operating margin still rose, because spending grows far slower than sales. Q4 guidance implies about 66%.
Software is a steady second engine. VMware and the other software lines grew 29% in Q3 and run at higher margins than chips. They fund most of the $2.60 annual dividend.
Broadcom quarterly results releases, Q1 FY2025 to Q3 FY2026
Debt. Total debt was $59.4 billion at Aug 2, 2026 against $24.0 billion of cash, most of it from the 2023 VMware purchase. At the current pace of free cash flow, about $55 billion a year, net debt could be paid off in under a year. Broadcom has instead kept returning cash: $8.45 billion of buybacks and about $9.3 billion of dividends in the first nine months of fiscal 2026.
Check the figures: Q3 FY26 results · Q2 FY26 results · Q1 FY26 results · Q4 and FY25 results · Q2 FY26 Form 10-Q
Competition, moat and management
Broadcom's edge is that it builds what each AI giant designs, at a scale few chip companies can match, while Nvidia sells everyone the same chip.
Who it competes with. Nvidia and AMD sell a ready-made AI chip with software developers already know; Broadcom's answer is that a custom chip costs the buyer less per unit of work at very large scale. Marvell, MediaTek and Taiwan design houses compete on price and ties to specific buyers, while Broadcom offers more complete designs, from chip to networking, at the leading edge. The buyers' own chip teams want full control but still need Broadcom's design tools, packaging know-how and networking. In software, public cloud, Nutanix and Red Hat pitch cheaper or newer alternatives to VMware.
Where the moat is real. Custom AI chips are designed together with the customer over two to three years, and each generation builds on the last. Broadcom also makes the Ethernet switch chips that connect AI clusters, so it can sell the whole data path. Few competitors can do both at the latest manufacturing nodes.
Where it is thin. The largest buyers have every reason to bring more design in-house or to play suppliers against each other. VMware's price increases after the acquisition pushed some customers to look for exits, which caps how long that business can grow.
Management. CEO Hock Tan has run the company since 2006 and built it through acquisitions, including Broadcom, CA Technologies, Symantec's enterprise business and VMware. The pattern is consistent: buy a business with sticky customers, cut costs hard and use the cash to pay down debt and raise the dividend.
What is it worth?
We value Broadcom at $430 per share over the next 12 months (to October 2027), a probability-weighted blend of three scenarios. That is about 19% above today's price, which supports a Buy.
| Assumption | Bear | Base | Bull |
|---|---|---|---|
| Fiscal 2027 revenue | |||
AI chip revenue Q4 FY26 guide: $21.7B a quarter | $80B | $110B | $150B |
Other chip revenue Wireless, storage, broadband | $16B | $17B | $18B |
Software revenue VMware and other software | $35B | $37B | $38B |
Total revenue | $131B | $164B | $206B |
| Earnings | |||
Adjusted operating margin Q4 FY26 guide implies ~66% | 61% | 63% | 64% |
Adjusted EPS ~4.94B diluted shares | $13.10 | $16.94 | $21.62 |
| Valuation | |||
P/E paid in Oct 2027 Today: ~21x our base case | 20x | 25x | 28x |
| Value per share | $262 | $424 | $605 |
| Probability weight | 25% | 50% | 25% |
| Probability-weighted value $429 | Price on Oct 9 $361.54 | Haverhill price target $430 +19% |
Our method. Broadcom is profitable and its earnings are growing fast, so we value it on earnings:
Forecast fiscal 2027 revenue (November 2026 to October 2027) by business: AI chips, other chips and software.
Apply an adjusted operating margin, then take out interest and tax to get adjusted net income, and divide by about 4.94 billion diluted shares.
Apply the price-to-earnings multiple we think investors will pay in October 2027, when they will be looking at fiscal 2028.
Weight the three cases 25% bear, 50% base and 25% bull.
We use adjusted (non-GAAP) earnings because they exclude amortization from the VMware purchase, which does not use cash. GAAP earnings are about 20% lower.
Why these numbers. Our base case of $110 billion in AI revenue assumes quarterly AI sales keep rising from the $21.7 billion guided for Q4, but at a much slower pace than this year's. That is consistent with the $164.6 billion contract backlog. The 63% margin allows for more lower-margin rack systems in the mix. A 25x multiple is a discount to the fastest-growing AI names, reflecting customer concentration.
What the price implies. Each turn of the multiple is worth about $17 a share on our base-case earnings, and each extra dollar of earnings about $25 at 25x. At $361.54 the stock prices in about $14.50 of fiscal 2027 earnings at 25x, or our base-case $16.94 at about 21x. Either way the market is assuming growth slows sharply after this year.
Today's valuation, for reference. At $361.54, Broadcom's equity is worth about $1.72 trillion, or about $1.76 trillion including net debt. That is roughly 31 times fiscal 2026 adjusted earnings, which we estimate at about $11.60 a share.
What are the real risks?
The biggest risk is not that AI demand disappears, but that a few very large buyers slow their orders at the same time. We rank the risks by how much they would move our value per share.
A big AI buyer pauses. Most AI revenue comes from a handful of companies. One delaying a chip generation by two quarters would take billions out of a year's revenue.
The AI spending cycle turns. If the largest companies cut data center budgets together, our bear case of $80 billion in fiscal 2027 AI revenue comes into play, and the multiple would fall too.
Margins slide on rack systems. Adjusted gross margin already fell 3.4 points in a year. Each point of operating margin is worth about $7 a share at our base-case multiple.
Buyers design more in-house. The largest customers keep building their own chip teams and could hand Broadcom a smaller share of each new generation.
A Taiwan or export-rule shock. Broadcom's leading chips are made in Taiwan, and US limits on chip sales to China could widen.
VMware customers leave. Price increases after the deal pushed some customers toward other platforms; renewals in 2027 will show how many go.
CEO succession. Hock Tan has shaped the company for two decades, and the deal-driven strategy depends on him.
Risks 1 and 2 drive most of the gap between our bear and base cases, and both show up first in the quarterly AI revenue line.
What would change our mind?
Our Buy rests on one idea: contracted AI demand keeps growing into fiscal 2027 while the stock is priced for a sharp slowdown. These are the specific signals that would prove that wrong in either direction.
Signals that would move us to Hold or Sell
AI revenue stalls. Two quarters in a row with AI revenue flat or down from the prior quarter.
Backlog shrinks. The contracted backlog, $164.6 billion in May, falling rather than growing.
Margins break lower. Adjusted operating margin below 60% in any quarter.
The price runs past about $480 with no change in our earnings view.
Signals that would make us more bullish
The fiscal 2027 outlook points above $130 billion of AI revenue. That would move our base case toward the bull case.
A sixth large AI customer signs a multi-year contract.
Margins hold near 66% even as rack systems grow.
Catalysts to watch
The next big test is the fiscal Q4 report in mid-December, when Broadcom usually frames the year ahead.
Broadcom reporting calendar; dates are Haverhill estimates
Sources and disclosures
Every company figure in this report comes from Broadcom's own filings and releases, linked below. The analysis, model and price target are Haverhill Financial Research's own work.
Disclosures. Haverhill Financial Research is a publisher of general financial commentary. This report is not personalized investment advice and does not consider any reader's financial situation; readers should do their own research. Price targets are estimates of value, not guarantees. The author of this post does not hold a position in AVGO. Data as of October 9, 2026.
