
Broadcom's AI Chip Revenue Hits $16.7B as Nvidia Pressure Mounts
Broadcom's AI chip revenue hit $16.70B in fiscal Q3, up 221% YoY, with Q4 guidance near $21.7B as hyperscalers shift from Nvidia GPUs to custom silicon.
- By
- Grace Kim
- Filed
- Channel
- AI & Compute
- Read
- 4 min read
By Grace Kim
Broadcom booked $16.70 billion in AI semiconductor revenue for its fiscal third quarter, up 221% year over year and 54% sequentially, as hyperscalers accelerate a shift toward custom silicon while Nvidia shares trade under pressure in the $160s. Barron's reported Thursday that Broadcom stock gained ground from Nvidia's decline. The market, in other words, is catching up to what regulatory filings have indicated for months.
The investment case centers on a simple commercial reality. When a frontier AI lab decides an off-the-shelf GPU costs too much, Broadcom frequently receives the design work. With Nvidia facing selling pressure, that dynamic now commands investor attention.
What did Hock Tan say about custom chips?
CEO Hock Tan made the argument directly on the September earnings call: "The lesson here is when you co-develop a chip that is optimized for your particular large language model workloads, you will outperform any GPU."
The claim rests on delivered products, not just roadmap promises:
- Broadcom shipped OpenAI's first custom accelerator, codenamed Jalapeno.
- The company began production shipments of TPU version 8i for Google, the Alphabet unit.
- Broadcom signed a long-term agreement to deliver TPUs worth multi-tens of billions of dollars annually over the next several years.
Management says custom XPUs often run workloads at less than half the cost of GPUs. Hyperscalers prioritize savings of that magnitude, and the arbitrage between merchant GPUs and co-designed accelerators now anchors Broadcom's AI franchise.
How fast is the AI business compounding?
Fourth-quarter guidance calls for approximately $21.70 billion in revenue, a 236% increase year over year. Management also states it has secured supply for about $115 billion of AI revenue in fiscal 2027 and sees a path to $230 billion in fiscal 2028.
Cash generation supports the expansion. Free cash flow reached $13.67 billion, equal to 46% of revenue, and cash on hand grew to $23.98 billion. Broadcom repaid $5.6 billion of longer-maturity debt during the quarter.
The dividend tells its own story. The quarterly payout of $0.65 per share marks the 15th consecutive annual dividend increase. At a share price of $364.36, the yield stands near 0.71%, so the income case depends on how quickly the payout keeps rising.
Is the valuation defensible?
Broadcom expects to exceed $30 in earnings per share in fiscal 2028. The current price works out to roughly 12 times that figure, a multiple the company's own forecast supports if execution holds.
Management has beaten EPS estimates for nine consecutive quarters. That track record lends weight to the fiscal 2028 projection, though it remains a management target rather than a confirmed result.
Why does Arista Networks lose the comparison?
Arista Networks is the obvious alternative play on AI networking. Its third-quarter revenue rose 58.45% year over year, the fastest growth since its IPO. The valuation keeps investors away: a P/E of 74, a free-cash-flow yield of 1.63% and no dividend.
Share performance diverged sharply this year. Arista gained 57.8%, while Broadcom added 5.85%.
The supply chain relationship matters more than the price gap. Arista's own 10-K discloses that it relies on Broadcom as its predominant merchant silicon vendor for switching chips. When Arista sells a switch, Broadcom has usually already sold the chip inside it.
What are the risks?
Customer concentration tops the risk list. Broadcom serves six frontier-model customers, and management describes only four of them as able to fund themselves.
The financing structure adds exposure. Broadcom set up financing vehicles for Anthropic and OpenAI that may carry residual-value guarantees, and the CFO gave no maximum exposure figure.
Margins are compressing at the edges. Gross margin should fall to about 73%, down from 78% a year ago, reflecting the cost of the custom-silicon business mix. Operating margin nonetheless reached a record 67.9%, and the Google agreement runs for years, which underpins the thesis despite the concentration risk.
Can the custom silicon run extend?
Tomahawk 7 has taped out as the industry's first 200-terabit-per-second Ethernet switch. Management expects AI networking revenue to grow as fast as XPUs, which would add a second growth engine alongside custom accelerators.
The next earnings report arrives December 9, 2026. As long as hyperscalers keep designing their own chips, and as long as Nvidia's pricing keeps giving them a reason to, Broadcom's custom silicon franchise retains its commercial foundation.
Source: 247wallst.com
More from Grace Kim
Show full bio
Market editor covering people, launches and day-to-day practice in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.
981 articles
Related articles
broadcom-semiconductor-revenue-up-68-as-analyst-cites-nvidia-opening
Broadcom Semiconductor Revenue Up 68% as Analyst Cites Nvidia Opening
broadcom-s-ai-revenue-grew-221-custom-silicon-may-outgrow-gpus-c59ba5d1
Broadcom's AI Revenue Grew 221% — Custom Silicon May Outgrow GPUs
broadcom-s-custom-silicon-unit-emerges-as-ai-chip-powerhouse-fc62c3c2
Broadcom's Custom Silicon Unit Emerges as AI Chip Powerhouse
broadcom-rallies-5-to-363-99-as-traders-rotate-out-of-nvidia
Broadcom Rallies 5% to $363.99 as Traders Rotate Out of NVIDIA



