Broadcom doubles the roadmap, then starts financing the check
First-ever explicit $115B/$230B AI revenue targets for FY27/FY28 at ~68% operating margins; a new financing vehicle to fund the labs that buy the silicon.
AVGO · Earnings Call · 2026-09-02
The roadmap went from vague to very specific
Hock Tan typically deals in trajectories, not targets. Three months ago he would only say fiscal 2027 would “very easily exceed $100 billion” of AI revenue and would not put a number on 2028. This quarter he did both. Fiscal 2026 AI revenue guidance was nudged up to ~$58 billion (up 186%), and for the first time management gave hard multi-year numbers: ~$115 billion in fiscal 2027 and ~$230 billion in fiscal 2028, with earnings per share on track to “exceed $30” in fiscal 2028.
In 2027, we have secured the supply to again double AI revenue to approximately $115 billion. Our demand actually exceeds this outlook and we will work to improve supply.
That language — “we have secured the supply,” “we will work to improve supply” — signals a genuine change in emphasis from the June call, where the bottleneck talk was about wafers and substrates. Now the binding constraint is no longer chips but the physical and financial scaffolding around them. The framing is a direct escalation of the compute capacity theme that now dominates Broadcom’s own keyword board (momentum 277, rank #1), and it echoes the customer journeys he walked through: Google’s long-term TPU agreement (multi-tens of billions of dollars annually), Anthropic ramping from 1 GW to an incremental 10 GW by 2028, OpenAI’s Jalapeño heading toward >5 GW, and Meta’s MTIA across three generations. The detail per gigawatt is strikingly consistent — roughly $20–$30 billion of content per GW, sustained as each generation grows in power and ASP but fewer chips fit in a GW.
We believe with a pretty high degree of confidence we will ship $350 billion of AI semiconductors to these customers in the next 2 years.
The vending machine and the “geniuses in Outer Mongolia”
What makes the roadmap credible is also what makes it new and slightly uncomfortable. Two of the six customers — the two fastest-growing — don’t yet have the cash flow to buy this much silicon, so Broadcom stood up the AI XPV platform with Apollo and Blackstone, closing a first $35 billion tranche in June for Anthropic’s first gigawatt.
“You have 2 geniuses in the middle of Outer Mongolia, say, and they need to go to college to fulfill where they want to … our XPUs enable cost-effective, sustainable growth at multiples of their infrastructure costs.” — Hock Tan, President and Chief Executive Officer · 2026-09-02
Amy Tiener is careful to frame the backstops as “modest residual value guarantees … we view as low risk,” but the strategic shift is unmistakable: Broadcom no longer merely sells into well-funded hyperscaler capex — it is co-arranging the capital formation of its own demand. That is a genuine business-model pivot, and it puts the balance sheet squarely in the conversation. Net debt sits at roughly $47 billion, and the off-balance-sheet guarantees — if they grow deal by deal — may eventually pressure the leverage profile that the company has worked hard to trim. As Hock put it, every gigawatt of compute they fund can become ~$30 billion of annualized revenue for the lab, which he calls “a hell of a business model.”
Margin math: watch the bottom line, not the gross line
There is also a louder structural trade-off this quarter. The Semiconductor Solutions segment’s gross margin was corrected mid-call to ~67%, and rising HBM/memory content in XPUs is now an open topic. Consolidated gross margin guides down from 78% a year ago to ~73% in Q4, yet operating margin is guided roughly flat at ~66%. Amy makes the case directly:
“As the proportion of AI revenue accelerates in Q4 … it impacts our margin. … Regardless, we expect Q4 operating margin to be approximately 66%, flat from a year ago, because our strong revenue growth drives substantial operating leverage.” — Amy Tiener, Chief Financial Officer · 2026-09-02
The balance is real, but the optics are getting harder to ignore, especially with the stock sitting ~24% below its June 2 high of $481.57 and down ~13% over the past 90 days. Notably, the last 30 days of tape have been conspicuously unfriendly to exactly the parts driving this quarter: high bandwidth memory, Frontier model optics, and the broad “AI data center” complex were among the tape’s largest decliners. Against that backdrop Broadcom delivered record free cash flow of $13.7B (46% of revenue) and paid down $5.6B of debt — yet the tape is clearly asking when the AI semiconductor surge stops being a promise stock and starts paying for itself. The same “Frontier model” phrasing surfaced in Credo’s (CRDO) report earlier this week (its keyword board ranks Frontier model second), so the read-through is shared, not idiosyncratic — the whole AI-infrastructure group has de-rated in tandem.
Enterprise becomes a second avenue
One quiet but notable counterpoint: for the first time in recent memory, software is being positioned as the way Broadcom reaches enterprises with AI without making them direct buyers of XPUs. Q3 infrastructure software revenue was $8.8B, up 29% with ARR growth of 15%, and Hock pointed to VMware Private AI Cloud as the vehicle for enterprises to run AI workloads on-prem. “Enterprise consumption of AI is opening a new opportunity for our infrastructure software business.” That is a softer echo of his earlier, sharper thesis that enterprises should consume tokens from the few frontier labs rather than own AI infrastructure — now the company monetizes that enterprise wave via software margins rather than silicon.
Looking at the financial trajectory underneath all this, the operating-leverage argument is the one the numbers keep supporting: operating income hit a record $20.1B in the quarter, with operating expenses growing only 22% against 127% segment revenue growth, and the interest coverage ratio is roughly 13.9x even after the debt paydowns. The question was never whether the model works quarter to quarter; it’s whether the buyers — and the balance sheet behind them — can keep writing checks big enough to justify a roadmap that just doubled, twice.