The silicon is taped out through 2028. The customers’ funding is not.
In chip design, tape-out is the moment a design is frozen and sent to the fab — no more edits. Broadcom has now frozen a roadmap that takes AI semiconductor revenue from $58B this fiscal year to about $115B in FY27 and $230B in FY28, after a quarter that grew 86% at a 67.9% operating margin. Yet the stock sits 28% below its June high, because the market is pricing one question: is that order book locked, or is Broadcom lending its customers the money to place it? Six analyst lenses, three scenarios, four time horizons.
Gray line = straight segments between 22 dated price anchors (Dec 2024 to Oct 2, 2026; peak and low dots are intraday extremes). Colored paths = synthesized scenario midpoints at mid-year, probability-weighted bear 30% · base 45% · bull 25% — bear-tilted because of customer concentration, vendor financing, and a stock that has sold off twice on good prints. The clay circle is the Street’s 12-month mean target, $531.85 (S&P Global, 50 analysts, range $215.88–$715), plotted at the 1-year node. Axis is linear from $100.
Re-weight the scenarios
Those probabilities are a judgment call — so make them yours. Drag to set how likely the bear and bull cases are (base takes the remainder); the blended target below, the dotted line on the chart, and the prob-weighted row of the scenario cards all update live.
Six analyst lenses, six answers
The same fundamentals support wildly different conclusions depending on which framework you trust. Each lens is a synthesized perspective with its own 12-month target; the panel mean is $410, below the Street’s $531.85.
The Gigawatt Compounder
AI semiconductor revenue is guided $58B → ~$115B → $230B across FY26–28, with supply for FY27 already secured and demand above it. Anthropic plans 5 GW of TPU 8i in 2027 with line of sight to 10 GW more; OpenAI’s Jalapeno and Meta’s MTIA are shipping. Operating margin hit a record 67.9%. Conviction: high.
The Cash Counter
TTM free cash flow is $39.4B (consensus FY26E $49.2B) against a $59.4B debt stack that is being paid down and $24.0B of cash. The yield is thin at ~2.3%, so this lens pays for earnings, not for cash: FY28 consensus EPS of $25.86 at a 17.5x multiple. Conviction: medium-high.
The Circular Ledger
Top-5 customers were ~55% of Q3 revenue, up from ~45% a quarter earlier. Broadcom will lend Anthropic up to $42B against a $125.2B TPU lease while gross margin slides from 78% to ~73%. If funding or model scaling slows, FY28 EPS stalls near $17.50. Conviction: medium.
The Toll Gate
Custom silicon locks in through design: six XPU customers, multi-year gigawatt commitments, and AI networking up more than 2.5x. The crack is Google, which management said will likely use more than one supplier; a Marvell–Alphabet deal rattled the stock in August. Priced as a toll, not a growth rocket. Conviction: medium.
The Revision Gap
The stock is back near its Dec 31, 2025 close ($344.20) while trailing EPS is roughly 43% higher, and it trades below its 200-day average, inside a 50-day range of $339–$428. A re-test of the upper range needs only a rerating from 18.3x to 21x FY27E EPS of $19.39. Conviction: low.
The Credit Cycle
AI capex has become a credit story: a $60B chip-financing syndicate ($42B senior, $18B junior) is being marketed while oil sits above $108 and the Fed meets. A multiple on custom silicon is rate-sensitive, and Taiwan and export rules are the unpriced tail. FY28E trimmed to $24 at 15x. Conviction: medium.
Wall Street 12-month price targets
What the sell-side expects over the next year. Bars are sorted low to high; the dashed line is today’s $355.14 — nearly every target sits above it.
Eight recent firm targets (TipRanks via StockAnalysis, Sep 3–Oct 2, 2026) out of 50 covering the stock: 40 Strong Buy, 7 Buy, 3 Hold, 0 Sell. Median target $535. MarketBeat shows a $527.20 average, a small gap between aggregators. Only D.A. Davidson (Hold, cut $400 to $350 on Sep 3) sits near today’s price; the $215.88 low is not attributed to a named firm and is omitted from the bars.
Where the road leads
Synthesized scenario midpoints (mid-year), each equal to next-twelve-month EPS times an exit multiple. Returns are vs. today’s $355.14. These are illustrative frameworks, not forecasts — outcomes hinge on whether the AI order book converts to cash.
1 Year
Mid-20272 Years
Mid-20283 Years
Mid-20295 Years
Mid-2031▸ Bull case — show the assumptions & math
▸ Base case — show the assumptions & math
▸ Bear case — show the assumptions & math
Revenue, capex, free cash flow & debt ($B)
Where the money actually goes. The bull and the bear theses both live in the gap between these four bars.
Revenue has nearly tripled since FY23 ($35.8B to a $106.0B consensus FY26E) and free cash flow is heading from $17.6B to $49.2B (consensus) while capex stays tiny: $1.25B over the last twelve months. Debt (balance-sheet short- plus long-term) peaked after the VMware deal at $67.5B and was $59.4B at Aug 2, 2026 against $24.0B of cash, so the buyback and dividend are funded by cash, not leverage. The bear’s worry is not capex but the off-bar item: a $42B customer loan that would sit on this balance sheet. FY26E revenue and FCF are consensus (S&P Global); the FY26E capex and debt bars are trailing/actual as of Q3.
EPS path underpinning the targets ($)
Each target is an EPS estimate times an exit multiple. Here is the earnings ladder the scenarios are built on, with management’s own FY28 claim marked.
Adjusted (non-GAAP) EPS: FY25 $6.82 reported; FY24 $4.87 derived from the reported +40% growth. FY26E $11.66 and FY27E $19.39 are S&P Global consensus; FY28E $25.86 is the LSEG consensus quoted by CNBC. The clay marker is Hock Tan’s statement that Broadcom is on target for over $30 of EPS (CNBC’s report ties it to FY28), which is 16% above consensus. FY29E–FY31E are the desk’s base-case extrapolation, not consensus. At $355.14 the stock is 18.3x FY27E and 13.7x FY28E consensus.
The business is still accelerating
Q3 FY26 (quarter ended Aug 2, 2026), year over year — read these against a stock 28% below its high.
Core P&L lines in olive, AI frontier lines in clay. Revenue +86% to $29.6B, free cash flow +95% to $13.7B, EPS +96% to $3.32, infrastructure software +29% to $8.75B. Management gave AI networking as “over 2.5x” and XPU shipments as “over 3.5x” year over year, shown here as minimums. Yet the stock is where it ended 2025. That gap is the bull’s case in one chart.
Bull vs. Bear
The valuation argument compresses into one disagreement: is the AI order book locked, or financed by Broadcom itself? Below, the wafer-map view of how few dies carry the yield.
Share of revenue from top-5 end customers · Q3 FY25 ~40% → Q2 FY26 ~45% → Q3 FY26 ~55% (company estimate via Motley Fool, Sep 27, 2026)
▲ THE BULL CASE
- The plan is guided, not guessed. AI semiconductor revenue $58B in FY26 (raised from $56B), ~$115B in FY27 with supply secured, $230B in FY28; Q4 AI guide $21.7B, +236%.
- Beat and raise. Q3 revenue $29.59B vs. $29.36B expected and EPS $3.32 vs. $3.24 (LSEG); Tan said $350B of AI chips ship to six XPU customers over two years.
- Operating leverage. Record 67.9% operating margin; $13.7B of free cash flow in one quarter and $39.4B over twelve months.
- Earnings outran the price. The stock sits near its 2025 year-end close while trailing EPS is up ~43%; at $355 it is 11.8x management’s >$30 FY28 EPS.
- Named, multi-gigawatt demand. Anthropic 5 GW of TPU 8i in 2027 plus 10 GW in sight, OpenAI’s Jalapeno, Meta’s MTIA, and tens of billions a year for Google.
- The Street is lopsided. 50 analysts, zero Sells, $531.85 mean target (+50%); nearly every firm target is above the current price.
▼ THE BEAR CASE
- Concentration is rising. Top-5 customers were ~55% of Q3 revenue vs. ~40% a year earlier, and Anthropic is expected to be the largest XPU customer in 2027.
- The loop. Broadcom will lend Anthropic up to $42B (convertible, about a third of its $125.2B TPU lease); a $60B syndicate is being marketed. Anthropic’s own filing flags conflicts, and it reported ~$4.6B of 2025 revenue and an operating loss above $8B.
- Margins dilute. Gross margin fell to 75% in Q3 and is guided to ~73% in Q4 vs. 78% a year ago as memory-heavy rack-scale XPUs ship.
- Good prints get sold. Shares fell 12.6% on Jun 4 when the AI forecast was not raised and 11.4% on Dec 12, 2025 despite beats; Q4 revenue guide of $34.8B was below the $35.03B LSEG expected.
- Anchor-customer risk. Tan acknowledged Google would likely use multiple suppliers, and a Marvell–Alphabet chip deal knocked shares ~6% in August.
- Sentiment is fragile. Amodei’s Sep 13 essay urging a slower frontier pace pushed AVGO down roughly 3–5% on Sep 14 (sources differ), though Tan said targets are unchanged.
Risk map — likelihood × impact
Where each risk sits over a 3–5 year view. The hot corner — likely and high-impact — holds one risk, customer concentration; the financing and digestion risks sit one row down, in “possible.”
- Flat software
- Margin dilution
- Multiple swings
- Few buyers
- Google rival
- Supply caps
- Capex pause
- Loan credit
- Taiwan shock
Customer concentration
Top-5 customers were ~55% of Q3 revenue; if Anthropic, Google or OpenAI pauses orders, FY27–28 volume breaks first.
AI capex digestion
Labs slow model scaling or run short of funding, and the $115B / $230B ramp slips a year or more.
Vendor-financing credit
A $42B loan to a loss-making customer puts Broadcom’s balance sheet behind its own revenue; defaults hit both lines.
Margin dilution
Rack-scale XPUs carry memory and pass-through cost; gross margin falls to ~73% in Q4 versus 78% a year ago.
AI valuation swings
Two 11–13% one-day drops on good prints (Dec ’25, Jun ’26) show how little room the multiple leaves.
Google multi-sourcing
Management said Google will likely use more than one chip supplier; a rival design win trims the anchor customer.
Supply bottlenecks
Demand exceeds the $115B plan, so wafers, memory and packaging set the ceiling, not orders.
Software stagnation
Infrastructure software is flat at ~$8.7B a quarter, so it dilutes growth rather than adding to it.
Taiwan / export shock
A cross-strait disruption or sudden export rule halts advanced-node supply; low odds, but it stops shipments outright.
The jargon, decoded
Hover the dotted terms in the metrics, or scan the desk’s working definitions here.
- XPU
- A custom AI accelerator designed with Broadcom for one customer (Google’s TPU, OpenAI’s Jalapeno, Meta’s MTIA). Six customers; 73% of Q3 AI revenue.
- Tape-out
- The step where a chip design is frozen and sent to manufacturing. After it, changes cost months and millions — hence the report’s metaphor.
- Non-GAAP EPS
- Earnings per share excluding acquisition amortization, stock compensation and one-offs. Q3: $3.32 vs. $2.68 GAAP.
- Free cash flow
- Cash from operations minus capital spending. $13.7B in Q3; $39.4B trailing twelve months.
- FCF yield
- Free cash flow ÷ market cap. About 2.3% on trailing cash flow — low, because the price already assumes growth.
- Vendor financing
- A supplier lending customers money to buy its product. Broadcom’s up-to-$42B Anthropic facility is the live example; it can lift volume and also concentrate credit risk.
- Mix dilution
- Gross margin falling because a lower-margin product (memory-heavy XPU racks) grows faster than the rest, even while dollars of profit rise.
- EV/EBITDA
- Enterprise value ÷ operating cash earnings. On Q3’s annualized adjusted EBITDA (~$81B, desk estimate) it is about 21x; management does not report a Q3 figure in the release.
- Exit multiple
- The P/E assumed at the forecast date. Multiply it by next-twelve-month EPS to get a target price.
- Prob-weighted
- Each scenario’s price × its probability, summed into one expected value across bear, base and bull.