Broadcom Q2: Google's "new favorite" is a disruption, but the $29.4 billion Q3 guidance and orders lined up through 2028 are the fundamentals.

After Broadcom’s Q2 earnings report, the stock price once fell over 8% in after-hours trading.

This is one of the most unusual earnings nights in the past two years—revenue hit an all-time high, AI chip revenue soared 143% year-on-year, cash flow posted a 46% profit margin, every data point was the best in Broadcom’s history. Yet the stock price fell instead of rising. The reason is simple: the company has already risen too much by 2026, and the market is used to the “the better, the harder it falls” game rule. But if you strip away emotions and look only at the numbers, you'll find: the true weight of this earnings report isn't in Q2, but in the Q3 guidance and the first-ever fully disclosed power commitment lineup from six major customers.

Q2 Revenue is the Appetizer, Q3 is the Main Course

In the second quarter of fiscal year 2026 (ending May 3), Broadcom delivered revenue of $22.2 billion, up 48% year-on-year, exceeding analysts’ consensus of about $22 billion. Non-GAAP diluted EPS was $2.44, higher than the expected $2.40. Adjusted EBITDA reached $15.2 billion, accounting for 69% of revenue, a historic record. Free cash flow was $10.3 billion, 46% of revenue, annualized over $40 billion. All these are Broadcom’s highest levels ever, but the market has already priced them in.

What actually makes people reassess the company’s scale limit is the Q3 guidance: Broadcom expects Q3 revenue of about $29.4 billion, up 84% year-on-year and 32% from the previous quarter. In just one quarter, Broadcom’s revenue steps up another tier. Among this, the AI semiconductor Q3 guidance is $16 billion, up over 200% year-on-year and up 48% from Q2’s $10.8 billion. Before the earnings report, most institutions predicted Broadcom Q3 revenue in the $24–25 billion range, far below the actual guidance.

CEO Hock Tan wrote in the earnings statement: “Continued growth in demand for AI custom accelerators and AI networking drove Q2 AI semiconductor revenue above expectations. That momentum is continuing.”

$30 Billion Orders, Visibility Extended to 2028

If Q3 guidance led to a repricing of Broadcom’s short-term growth, the first quantitative disclosure of the six major customers’ roadmaps painted an auditable outline for Broadcom’s long-term growth.

In Q2, Broadcom shipped AI semiconductors worth $10.8 billion, but total customer orders received during the same period exceeded $30 billion—nearly triple the shipment volume. Such backlog isn’t due to supply shortage, but because customers are planning ahead: AI data center construction requires not only chips, but also power coordination, cooling systems, and cross-regional infrastructure, and these supporting preparations often take longer than chip deliveries. Customers place orders in advance to lock in future computing capacity. Therefore, Broadcom’s order visibility has extended from 2027 three months ago to 2028.

On the conference call, Hock Tan systematically listed for the first time the specific deployment plans of six key customers: Google will use over 1 GW of Broadcom TPU computing in 2026, with a long-term agreement covering multiple product generations signed in April; Anthropic will receive 5 GW of next-gen TPU computing starting 2027, with over 1 GW delivered ahead in 2026; OpenAI’s chips have been delivered, expected to start mass production by year-end and deploy 1.3 GW in 2027—part of the “10 GW before 2029” mega order; Meta commits to deploying 3 GW of MTIA XPU by the end of 2028, with the first 1 GW order received and delivery starting in H2 2027; two other unnamed customers have cumulatively placed $6 billion in orders, with shipments expected to begin by the end of 2026. When Goldman Sachs asked if such growth can continue, Hock Tan replied bluntly: “Demand for XPU and networking is simply endless.”

Google “Looking Elsewhere”—Broadcom Faces Calmly

Before the earnings release, the market’s biggest concern was Google. News that Marvell took over part of Google’s next-gen TPU design raised questions for months about whether Broadcom would lose its biggest customer. On the call, Hock Tan chose rare active candor: “We fully accept this fact: given the pace of AI computing consumption, it is completely foreseeable for customers like Google to have diversified supply sources.”

This is the first official confirmation from Broadcom that Google is seeking supplier diversification. But Hock Tan immediately provided a floor: the long-term agreement signed in April covers multiple generations of TPU and AI network products, “their committed amount to us is very substantial.” In 2026, Broadcom will supply Google over 1 GW of TPU computing—the quantitative answer itself.

Google’s diversification is real, but the direction is “find another besides Broadcom,” not “replace Broadcom.” With AI computing demand almost doubling annually, this logical difference is enough to support long-term cooperation. Broadcom’s strategy is to hedge the risk of market share ceiling with multi-generation contracts and deep technology binding, not rely on monopoly maintenance.

Profit Margin Declines, but Accounts Still Add Up

In Broadcom’s earnings report, gross margin is the only item that warrants a second look. Non-GAAP gross margin is 77.1%, down 230 basis points year-on-year. Q3 guidance drops further to about 74%. The underlying mechanism is simple: Broadcom’s two business lines, semiconductor gross margin about 70%, infrastructure software about 93%. As AI semiconductors’ share of total revenue rises from about 30% in Q2 FY2025 to 49% in Q2 FY2026, blended gross margin naturally falls.

But does this mean “the more you earn, the thinner the margin”? Data says no. Non-GAAP operating margin in Q2 actually rose 200 basis points year-on-year to 67.3%—operating expenses are relatively inflexible, so rapid revenue expansion produces effective operating leverage. EBITDA margin set a record at 69%, confirming this as well. CFO Kirsten Spears suggests investors “model semiconductor and software separately,” rather than judging total profitability by blended gross margin. She also announced retirement effective June 12th, with incoming CFO Amie Thuener as replacement.

$35 Billion, A New Chessboard

There’s another incremental update in the report: Broadcom announced a partnership with Apollo, Blackstone, and others to form the “AI XPV platform,” planning deployment of over 20 GW computing capacity by 2028, with an initial scale of $35 billion launched by Apollo. The logic is clear: frontier AI labs like Anthropic and OpenAI have strong computing needs, but their balance sheets cannot support infrastructure investment of hundreds of billions. Broadcom provides chips and tech, financial institutions bring capital, the platform delivers scalable computing to customers at the lowest cost and power consumption.

When analysts asked if Broadcom would provide a complete rack-level solution, Hock Tan responded firmly: “No, no, all chips.” CFO Kirsten Spears added: “Only chip business.” Amid wide enthusiasm for full-stack AI solutions, this focus is one reason Broadcom can maintain high operating margins. If the AI XPV platform proceeds as planned, Broadcom will become not just a chip supplier but a coordinator across the AI computing ecosystem.

Editor’s Conclusion

The after-hours decline is short-term valuation skirmish, not a judgement on the company’s operational reality. What truly matters from this Broadcom earnings report is the six major customers’ quantified roadmap in GW and USD, and the openly extended order visibility to 2028 for the first time.

Google’s “looking elsewhere” will remain a variable, but the $30 billion order backlog and combined commitments of more than 20 GW from six customers have built a solid base for Broadcom’s growth trajectory. The FY2027 AI semiconductor $100 billion+ target now sounds less like an ambitious forecast, and more like a signed client schedule.

Source: Research Beyond the Surface

Risk Warning and DisclaimerThe market has risks, and investments need caution. This article does not constitute personal investment advice, nor does it consider individual users' specific investment goals, financial situations, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article fit their specific circumstances. Investing accordingly is at your own risk.