Broadcom's latest quarter provided exactly what Bulls wanted to see. Clean beats, higher guides and most importantly, its custom AI Accelerator (XPU) strategy is transforming marquee outlook into paid customers. Revenues rose 22% to $15.95 billion, with the adjusted EPS reaching $1.69 due to heavy lifting by AI semiconductors and VMware. Management has also revealed a $10 billion new production order from its fourth XPU customer, increasing confidence in the sudden AI lamp from fiscal year 2026. With the stock being split to an all-time high, the market votes that Broadcom's AI Plus Networking Plus software formula is not only resilient, but it's compounded.
In Scorecard vs. Expected Value, the Broadcom heading number cleared the bar. Revenues of $1595.2 billion broke the consensus of $15.826 billion. The adjusted EBITDA landed at $10.7 billion (67% margin) ahead of the 66% guide. As for the segment, the semiconductor solution rose to $9.2 billion with AI. Infrastructure software rose to $6.8-67.9 billion with support from VMware Cloud Foundation Momentum. AI semiconductor revenue jumped to $5.2 billion, 63% year-on-year, bringing past executives' $5.1 billion signs. In the October quarter (FQ4), Broadcom is expected to lead revenues to approximately $17.4 billion (conflict-$17 billion), and AI semiconductor revenues are expected to accelerate to $6.2 billion. Total margin was 78.4% in the third quarter, with CFO commentary pointing to a modest sequential step-down in Q4 of the mix (XPU and seasonal wireless).
The main drivers were simple and highly consistent with the paper. First, custom AI accelerators: Broadcom gains wallet share by co-designing silicon that compresses the distance from “prospects” to “revenue” and maps it to hyperscale workloads. Second, networking: With Tomahahawk 5/6 switches and Jericho4 fabric routers, Broadcom is located in a large AI cluster at the chokepoint. It is a gating factor, and the maturation of the ecosystem in Ethernet is seduction. Third, VMware:VCF 9.0 is building bridges for enterprise customers to build containerized on-prem AI, creating cross-currents that support software mixing and gross profits. The $110 billion integrated backlog (an AI-skinned backlog for each management) is visible.
Outlook has been tightened and improved where it matters. For FQ4, management is looking at around $17.4 billion in revenue, with the semi-finals at around $10.7 billion (AI is $6.2 billion) and around $6.7 billion in software, up until mid-year, but it's almost flat and down 1% as customers consume VCF 9.0. The non-AA semi-finals are stable at around $4.6 billion with double-digit consecutive bumps, but management has yet to call a V-shaped recovery. The bigger swing factor is 2026: the fourth additional eligible XPU customer, currently posting production orders, “significantly” increasing the company's AI revenue outlook and +60% views of the previous quarter. Several sell-side shops quickly moved their price targets to $382-400, raising revenue trajectories for the FY26 AI. It also flagged the commitment of CEO Hock Tan until 2030 as a useful breakthrough in the playbook.
$10 billion AI order for that “secret”: Broadcom checked its size and timing (2nd FY26 focused delivery) rather than the name. Multiple reports and analyst notes point to Openai as a possible customerconsistent with prior coverage of Broadcom's collaboration to diversify beyond merchant GPUs. It's not just the dollar numbers that are worth noting. That's cadence. Moving from prospects to qualified customers on billions of dollars production orders over this timeframe is unusually fast in custom silicon programs, reinforcing cases where Broadcom can convert remaining XPU leads into revenue over the next 12-24 months. We will also examine the papers that show that the AI value chain is branching. Merchant GPUs are custom silicon for speed, cost/performance at hyperscale.
The impact on Nvidia is subtle. In recent years, Broadcom's victory will not change Nvidia's 2025-26 advantage. Nvidia was led to a $7 billion consecutive revenue growth rate, surpassing Broadcom's overall quarterly AI revenue of 3-4 times the Networking Scale's framing per Morgan Stanley. Take-outs have fewer “today's stock losses” and “tomorrow is progressive TAM and selective replacements.” If Openai and peers optimize some of their inference or specialized training using custom XPUs, Nvidia's mix can shift in margins as the overall pie grows. Broadcom also supplies network piping that sews GPU clusters, so even if you compete in the accelerator layer, you will still be part of Nvidia's ecosystem. Broadcom's risks are the flip side of opportunity. Multisourcing is still the norm, allowing ASIC schedules to slip and export controls to plan. But mathematics is persuasive when even 80% of management stories prove.
Financially, Broadcom continues to print cash and increase profitability. Operating profit for the third quarter was $10.5 billion. Free cash flow was $7 billion (44% of revenue). Dividends continue to hum ($0.59 per share this quarter), inventory is carefully constructed ahead of the lamp, leaving liquidity ($10.7 billion in cash vs. $66.3 billion in total debt). The mix slightly lowers the total margin in Q4, but the operating leverage remains intact given the software contribution and structurally high XPU content.
What to see next: (1) Create a signal in any color of the fourth XPU customer's platform milestone and remaining prospect colors. (2) Adoption of networking products as cluster size and Ethernet vie for more AI spine and leaf sockets. (3) Moneyization of VMware as businesses roll into VCF 9.0 and pilot Private-Cloud AI. (4) Non-AAI semi-lead-through until 2025 – Study is fine. Acceleration is the opposite. As the stock is the highest ever, this argument will center around how suddenly the slope of FY26–27 AI is, and whether the execution will continue to walk according to expectations.
Conclusion: Broadcom offers AI and VMware fixed beat and Rays, paired with a $10 billion XPU order of catalysts that upgrade the medium-term growth arc. The Nvidia fortress has not collapsed, but the map has changed at the edges, and Broadcom has drawn some new lines. Anyway, I agree with the tape.
