The news
Broadcom earnings for the quarter ended August 2, 2026, reported by Broadcom Inc. (AVGO) on September 2, showed revenue up 86% from a year earlier to $29.6 billion, led by AI semiconductor revenue that rose 221% to $16.7 billion.
President and CEO Hock Tan said demand for the company's custom AI accelerators and networking chips "continues to be very strong." AI semiconductor revenue grew 54% from the prior quarter, and Tan said Broadcom expected it to reach $21.7 billion in the fiscal fourth quarter, up 236% from a year earlier. The company guided total fourth-quarter revenue, for the period ending November 1, 2026, to about $34.8 billion, a 93% increase, with non-GAAP operating income of about 66% of revenue.
Semiconductor solutions revenue rose 127% to $20.84 billion, or 70% of the total. Infrastructure software revenue grew 29% to $8.75 billion. GAAP net income was $13.09 billion, up from $4.14 billion a year earlier, and non-GAAP diluted earnings per share were $3.32, compared with $1.69. Chief Financial Officer Amie Thuener said the quarter set records for revenue, operating profit and free cash flow.
Cash from operations was $14.2 billion. After $0.5 billion of capital spending, free cash flow came to $13.7 billion, or 46% of revenue. Broadcom declared a quarterly dividend of $0.65 per share, payable on September 30 to shareholders of record on September 21, according to its Form 8-K.
On the earnings call, according to a transcript published by Investing.com, Tan said six XPU customers (XPU refers to custom AI processors) were speeding up adoption. He said Broadcom delivered Ironwood TPU v7, the seventh version of the tensor processing unit (TPU), in high volume to Anthropic and Google, shipped OpenAI's first-generation custom accelerator, and expected production shipments of Meta's MTIA accelerator. He said AI networking revenue rose more than 2.5 times from a year earlier and that Broadcom had secured supply to double AI revenue to about $115 billion in 2027. Thuener said the company expected fourth-quarter gross margin of about 73%, down from 78% a year earlier, as a rising mix of custom accelerators with more memory content dilutes margins.
Investing.com reported that non-GAAP earnings per share topped an analyst estimate of $3.21 and revenue beat a $29.25 billion estimate. Broadcom shares had closed the regular session on September 2 down 0.6% at $367.47 and were up 0.21% at $368.01 in after-hours trading at the time of that report.
The numbers
- Q3 revenue
- $29.6 billion (up 86%)
- Q3 AI semiconductor revenue
- $16.7 billion (up 221%)
- Semiconductor solutions revenue
- $20.84 billion (up 127%)
- Infrastructure software revenue
- $8.75 billion (up 29%)
- Non-GAAP diluted EPS
- $3.32 (vs $1.69)
- Free cash flow
- $13.7 billion (46% of revenue)
- Q4 revenue guidance
- About $34.8 billion (up 93%)
- Q4 AI semiconductor revenue forecast
- $21.7 billion (up 236%)
Why CEOs should care
For technology buyers, the result shows where AI computing capacity is being built. Tan named Google, Anthropic, OpenAI and Meta as customers for Broadcom-built custom accelerators, which means more of the AI capacity enterprises rent may run on custom silicon rather than general-purpose graphics processors (GPUs). Ask your cloud providers which accelerators your workloads will land on, how portable your models and code are across chip types, and what capacity they can commit to. Tan also cited memory supply as a constraint and said Broadcom is building its own chip substrate capacity in Singapore, so reserve capacity for 2027 projects early.
CFOs should read the software line alongside the chip line. Infrastructure software revenue grew 29%, while Broadcom cited 15% growth in annual recurring revenue on the call. Companies running VMware should benchmark renewal quotes, model multi-year cost scenarios, ask whether longer commitments buy price protection and price out alternatives before contracts come due.
Boards and CISOs should treat AI supply concentration as a risk topic. Tan's customer list shows several of the best-known AI model developers relying on custom chips from the same designer. Directors can ask management how dependent key AI products are on one cloud or chip platform, and CISOs evaluating the VMware Private AI Cloud offering Tan highlighted on the call should weigh the data-control benefits of on-premises AI against deeper lock-in to one vendor's stack.
The bigger picture
By Tech CEO Daily's calculation from Broadcom's figures, AI semiconductors made up more than half of third-quarter revenue, and the fourth-quarter forecast would take that share above 60%. The company tied its fourth-quarter gross margin forecast of about 73% to that rising AI share, according to the call transcript. Tan's comments also point to a broader change: large AI developers are designing their own accelerators with partners such as Broadcom rather than relying on off-the-shelf processors alone.
What happened next
On September 9, Broadcom filed its Form 10-Q, which put remaining performance obligations (contracted revenue not yet recognized) at about $179.2 billion as of August 2, with about 25% expected to be recognized within 12 months. The filing said that total includes a long-term custom AI accelerator contract signed in the quarter ended May 3. A September 5 Insider Monkey article on Yahoo Finance reported that Truist Securities analyst William Stein cut his price target from $550 to $520 while keeping a Buy rating, and that the fourth-quarter revenue forecast fell short of a $35.03 billion consensus.
On September 11, Bloomberg reported that EU antitrust officials were questioning cloud providers about Broadcom's VMware licensing changes, according to SDxCentral. The dividend is payable on September 30, and the fiscal fourth quarter closes on November 1, when the $21.7 billion AI forecast will be tested.
