Broadcom is being etched into every AI cluster on earth. The fear is that its own customers learn to etch it out.
AI silicon revenue hit $10.8B in a single quarter — up 143% — with a bookings backlog past $30B and a line of sight to $100B+ by fiscal 2027. Yet the stock sits 27% below its June high because the market is pricing one question: does the custom-silicon boom run through Broadcom’s design IP and networking fabric, or do the six hyperscalers it depends on eventually design it out? Five analyst lenses, three scenarios, four time horizons.
Gray line = Broadcom’s actual price into today (10-for-1 split adjusted): the April ’25 tariff dip, the run to a $495 high on Jun 2 ’26, then a ~25% post-earnings slide to $360.45. The early-2025 trough predates the trailing 52-week window, so the marked 52-wk low is $270 (late summer ’25). Colored paths = synthesized scenario midpoints, probability-weighted (base 45% · bear 30% · bull 25%). Wall Street 12-month consensus ≈ $515 (range $437–$582, “Strong Buy” from 51 of ~57 analysts).
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.
Five analyst lenses, five answers
The same blowout quarter supports wildly different conclusions depending on which framework you trust. Each lens below is a synthesized expert perspective with its own 12-month target.
The Arms Dealer
AI silicon at $10.8B/qtr grew 143%, with in-quarter bookings above $30B against $10.8B shipped — demand Hock Tan calls “insatiable.” Six hyperscaler XPU customers (Google, Meta, OpenAI, Anthropic + two more), a $56B FY26 AI guide rising to $100B+ in FY27, all at a record 67% operating margin. Whoever wins the model race, they buy their custom silicon and networking fabric here.
The Cash Compounder
Record $10.3B free cash flow in one quarter (46% of revenue), 15 straight years of dividend hikes, a fresh $10B buyback, and a VMware software annuity throwing off high-margin, sticky recurring revenue. But at ~$1.7T the trailing FCF yield is only ~1.8% — a quality machine priced for a lot of future. Rare growth, rich price.
The Concentration Trap
A handful of customers drive the whole AI line. Google is bringing in MediaTek as a second TPU source — Broadcom’s TPU share could fall from ~95% to ~65% by 2028 (Macquarie, now Neutral). The price embeds a ~$115B FY28 free-cash-flow target that stress-tests closer to $67–83B. Once hyperscalers can design without you, the take gets squeezed.
The Design-Win Flywheel
Switching costs are brutal: once a hyperscaler co-designs an XPU on Broadcom’s IP and advanced packaging and wires the cluster with Tomahawk/Jericho, it’s locked for the chip’s multi-year life. Networking is ~40% of AI revenue and even harder to replace, and Alphabet’s TPU-development contract runs to 2031. Second-sourcing dilutes share; it doesn’t dislodge the incumbent.
Priced For Perfection
Shares made an all-time-high $481 close on Jun 2, then fell ~25% on an in-line (not raised) guide — a tape that punishes anything short of a beat-and-raise. Now mid-range and near the 200-day average. On absolute multiples it’s expensive (~35x this year’s EPS), but PEG is ~0.5 — cheap once you credit the growth. A coin-flip until execution re-rates it.
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 $360.45 — note that every target, even the lone Hold, sits well above it.
Sell-side 12-month targets — a selection of the ~57 firms covering Broadcom; the full consensus is ≈ $515, about +43% above today, with 51 Buy / 6 Hold / 0 Sell. The dashed line marks today’s $360.45: even Macquarie’s cautious Hold ($437, cut after Google’s second-source move) sits ~21% above the price. Zero Sell ratings is itself a signal — the Street sees the pullback as a valuation reset, not a broken thesis. Firms, ratings, and targets illustrative of publicly reported calls.
Where the traces lead
Synthesized scenario midpoints (mid-year). Returns shown vs. today’s $360.45. These are illustrative frameworks, not predictions with certainty — the five-year spread hinges almost entirely on whether the custom-silicon TAM keeps compounding and whether Broadcom keeps its share of it.
1 Year
Mid-20272 Years
Mid-20283 Years
Mid-20295 Years
Mid-2031▸ Bull case — show the assumptions & math
▸ Base case — show the assumptions & math
▸ Bear case — show the assumptions & math
Revenue, capex, free cash flow & debt ($B)
Where the money actually goes. Broadcom is fabless — capex barely registers — so nearly half of revenue converts to cash. The bull and bear both live in the gap between the revenue bar and the debt bar VMware left behind.
Revenue nearly triples across the window — the FY24 jump is the VMware acquisition, the FY26E surge is AI silicon — while free cash flow (olive) marches from ~$18B to a projected ~$42B. Capex (clay) is almost invisible: a fabless model that turns ~45% of revenue into cash. The catch is the slate debt bar: VMware added ~$30B, peaking near $68B; it’s edging down (a $3B bond buyback in Jun ’26) but still roughly 1.5× annual FCF. Figures from Broadcom filings; FY2026E is illustrative. Debt is gross principal; FCF is annual.
EPS path underpinning the targets ($)
The price targets aren’t pulled from the air — each is an EPS estimate times an exit multiple. Here’s the non-GAAP earnings ladder the scenarios are built on.
Adjusted (non-GAAP) EPS — the clean view; reported GAAP swings on stock-comp and acquisition accounting. Gray = reported ($4.87 in FY24, $6.82 in FY25), olive = estimates that roughly double by FY27 on the AI ramp, then grow at a decelerating clip. The base case’s ~$30 of FY31 EPS at a ~27× exit multiple ≈ the $810 base-case 5-year target — this is the ladder underneath those prices. Estimates illustrative.
The business is compounding
Q2 FY26, year-over-year — read these against a stock that fell ~25% the day after it printed them.
Every line is deeply green — revenue +48%, EBITDA +52%, EPS +55%, semiconductors +79% — with the AI frontier (clay) compounding at 143% and guided to +200% next quarter. Yet the stock trades ~25% below its June high. That gap is the bull’s entire case in one chart: the business is accelerating; the multiple compressed on a guide that was merely reiterated, not raised. Software growth is the steady annuity underneath. Frontier figures per company guidance.
Bull vs. Bear
The entire valuation argument compresses into one disagreement: is Broadcom the indispensable design layer of the AI build-out, or a concentrated toll-taker its own customers are learning to bypass?
THE BULL CASE
- Demand is “insatiable.” AI silicon hit $10.8B in Q2 (+143%), with in-quarter bookings above $30B against $10.8B shipped — a backlog JPMorgan sees topping $150B in 2027.
- Guidance is staggering. $56B AI revenue in FY26 (~+180%), $16B guided for Q3 alone, and a reiterated $100B+ for FY27 — off a base that didn’t exist three years ago.
- Profitability scales with it. Record 67% operating margin and 69% adjusted EBITDA even as revenue nearly doubles — extraordinary operating leverage.
- A cash machine with a moat. $10.3B quarterly FCF, a $10B buyback, 15 straight dividend raises, plus a sticky VMware software annuity.
- Design-win lock-in. Six hyperscaler XPU customers; once a chip is co-designed on Broadcom IP and wired with Tomahawk/Jericho, it’s captive for years. Networking is ~40% of AI revenue.
- The Street is unanimous. 51 Buy / 6 Hold / 0 Sell, consensus ~$515 (+43%); firms raised targets into the post-earnings sell-off.
THE BEAR CASE
- Customer concentration is the whole risk. A handful of hyperscalers drive the AI line; Google is adding MediaTek as a second TPU source, and Broadcom’s TPU share could fall ~95% → ~65% by 2028 (Macquarie).
- Priced for near-perfection. ~$1.7T embeds a ~$115B FY28 free-cash-flow target that stress-tests to $67–83B — still huge, but well short of what the multiple assumes.
- In-line, not raised. The stock fell ~25% after a beat because Tan reiterated rather than lifted the $100B FY27 target — sentiment is fragile.
- The XPV wildcard. A $35B Apollo/Blackstone financing vehicle turns Broadcom into a credit bet exposed to customer solvency and grid capacity, not just chip demand.
- In-sourcing pressure. Every hyperscaler is building its own silicon team; second-sourcing today can become design-it-out tomorrow, squeezing take rate.
- AI-capex cyclicality. If the data-center build-out digests even briefly, a $16B-per-quarter AI run-rate is a long way to fall from.
Risk map — likelihood × impact
Where each risk sits, not just how big it is. The hot upper-right — likely and high-impact — is the valuation re-rating; the genuinely existential items (a customer-solvency shock, an AI-demand air pocket) are low-odds tails that would reprice the stock overnight.
- Semiconductor cyclicality
- Google TPU share erosion
- Multiple de-rating
- EU antitrust probe
- AI margin-mix dilution
- VMware pricing backlash
- AI-capex digestion
- Hyperscaler in-sourcing
- XPV customer-solvency shock
- AI-demand air pocket
Multiple de-rating
At ~35× this year’s EPS, any wobble in the AI narrative compresses the multiple fast — the June sell-off on an in-line guide was a preview.
AI-capex digestion
Hyperscaler data-center spend pauses to digest; a $16B-per-quarter AI run-rate has a long way to fall if orders slow.
Hyperscaler in-sourcing
Customers build internal silicon teams and route around Broadcom’s design IP, compressing both volume and take rate.
Google TPU share erosion
MediaTek as a second TPU source cuts Broadcom’s share of its largest custom-chip customer — ~95% toward ~65% by 2028.
XPV customer-solvency shock
The $35B Apollo/Blackstone financing structure exposes Broadcom to a customer default or a stalled multi-gigawatt build.
AI-demand air pocket
A broad “AI bubble” unwind hits orders across all six customers at once — low odds, but it reprices the platform overnight.
AI margin-mix dilution
Custom-silicon hardware carries lower gross margin than legacy chips and software; a mix shift dilutes the blended margin.
VMware pricing backlash
Aggressive VMware repricing drives enterprise defections, eroding the sticky software annuity that anchors the quality case.
Semiconductor cyclicality
Non-AI chip lines (broadband, wireless, storage) remain cyclical and can drag the blended growth rate.
The jargon, decoded
Hover the dotted terms in the metrics, or scan the desk’s working definitions here.
- Custom AI accelerator (XPU / ASIC)
- A chip designed for one customer’s exact AI workload (e.g. Google’s TPU). Broadcom co-designs and manufactures these — cheaper per useful FLOP than a general-purpose GPU.
- Networking fabric
- The switch silicon (Tomahawk, Jericho) that wires thousands of chips into one cluster. ~40% of Broadcom’s AI revenue and harder to second-source than the accelerator itself.
- Second-sourcing
- A customer adding a rival supplier (e.g. Google adding MediaTek) to reduce dependence — it dilutes the incumbent’s share without fully replacing it.
- Design win / lock-in
- Once a chip is co-engineered on a vendor’s IP, switching mid-life is prohibitively costly — the source of Broadcom’s multi-year revenue visibility.
- Adjusted EBITDA
- Operating profit before interest, tax, depreciation and amortization, adjusted for stock comp and one-offs. Broadcom’s ran 69% of revenue in Q2.
- Free cash flow
- Cash from operations minus capex — the fuel for dividends and the $10B buyback. ~$10.3B in Q2 alone.
- FCF yield
- Free cash flow ÷ market cap. ~1.8% here: rich, because the price already credits years of AI growth.
- Exit multiple
- The P/E assumed at the end of the forecast. Multiply it by projected EPS to get a target price.
- Prob-weighted
- Each scenario’s price × its probability, summed into a single expected value across bear, base and bull.