Broadcom's Q3 guidance falls short of expectations at $1.2 billion, down over 13% after hours, AI narrative "cooling"?

Author: Ada, Deep Tide TechFlow

After the market close on June 3rd Eastern Time, Broadcom announced its fiscal year 2026 second quarter results ending May 3, 2026. From absolute numbers, this is a record-breaking quarterly report. Revenue of $22.19 billion, up 48% year-over-year, the highest single-quarter growth since January 2017; adjusted EPS of $2.44, surpassing analyst consensus expectations of $2.40. However, the market's focus was not on Q2, but on Broadcom's AI chip revenue guidance for Q3, which is $16 billion, up over 200% year-over-year, but nearly 7 percentage points below the sell-side consensus of $17.2 billion. This gap, along with the company's software business slightly below expectations, triggered a sharp market reaction.

Q2 performance nearly perfect, AI semiconductor revenue grows for 13 consecutive quarters

According to Broadcom's official disclosure, Q2 AI semiconductor revenue reached $10.8 billion, up 143% year-over-year, exceeding the company's previous guidance of $10.7 billion given in March. CEO Fuyang Chen stated in the earnings release that this quarter's growth was driven by "dual demand for custom AI accelerators and AI networking."

Looking at the business segments, semiconductor solutions revenue was $15.01B, up 79%, accounting for 68% of total revenue; among which, AI semiconductor revenue now accounts for 72%, non-AI semiconductor revenue was $4.2 billion, up 6%, with order backlog exceeding $6 billion, showing cyclical recovery. Infrastructure software (i.e., VMware) revenue was $7.18B, up 9% year-over-year, consistent with the company's guidance but below StreetAccount's surveyed estimate of $7.32 billion, a gap of about $140 million.

Profitability was also strong. Adjusted EBITDA reached $15.2 billion, representing 69% of revenue, setting a new record; free cash flow was $10.26 billion, 46% of revenue; cash balance at the end of the quarter was $19.63 billion, an increase of $5.4 billion from the previous quarter.

Q3 guidance exceeds expectations, but AI semiconductor revenue "misses by $1.2 billion"

Broadcom's Q3 guidance projects total revenue of $29.4 billion, up 84% year-over-year, above the consensus estimate of $28.54 billion; among which, semiconductor revenue guidance is $20.5 billion, up 124%. However, the AI semiconductor revenue guidance is $16 billion, compared to the sell-side consensus of $17.2 billion aggregated by LSEG, a shortfall of 7%; relative to some more optimistic buy-side expectations, the gap is even larger.

More critically, Fuyang Chen did not raise the AI chip revenue guidance for FY2026 during the call. According to CNBC, he reiterated that "the company expects this momentum to continue into FY2027, and maintains the AI semiconductor revenue target of over $100 billion." Bernstein analyst Stacy Rasgon commented that the underperformance in Q3 AI revenue guidance was dragging down Broadcom's stock price.

Adding Q1 ($8.4 billion), Q2 ($10.8 billion) actual revenues and Q3, Q4 forecasts, Broadcom's total AI chip sales for this fiscal year are expected to be around $56 billion, still about $1.6 billion below the analyst consensus of $57.6 billion.

Post-market decline over 13%, options market has already priced in intense volatility

Broadcom's post-market reaction was highly volatile. After the earnings release at 4 p.m. Eastern Time on June 3, AVGO initially dropped about 5%; as guidance details were disclosed during the call, the decline continued, with a drop of over 15% at one point after hours, ultimately closing down 13.78%. Based on the pre-earnings closing price of about $479, the market cap evaporated by more than $270 billion in a single day.

It is noteworthy that the capital markets had already been preparing for significant post-earnings volatility. Multiple media outlets cited that the options market had priced in about 7.8% of single-day volatility for Broadcom's post-earnings move, significantly above the historical average. This pricing reflects investors' dilemma: Broadcom's stock had already rebounded over 60% from the March lows before earnings, with nearly a 40% increase since 2026, and its valuation (around 90 times P/E) was far above the semiconductor industry average of about 69 times.

Due to this valuation concern, the market's implicit threshold for Broadcom's earnings was "full-outbeat expectations"; any guidance below a "blowout" performance could trigger profit-taking.

AI networking revenue share to fall from 40% to 30%

For the A-share optical module sector, a statement from Fuyang Chen during the call about AI networking business might be more damaging than the overall AI guidance.

According to Yahoo Finance, citing the call, Chen confirmed that this quarter, AI networking accounts for "close to 40%" of AI semiconductor revenue, but he also said that this proportion is expected to "normalize over time to around 30%, rather than remaining near 40%."

This is the first time Broadcom's management has explicitly outlined a path for the decline in AI networking's share. AI networking (including Ethernet switch chips, optical transceivers, and connection chips) is the core revenue source for China's A-share optical module leaders such as InnoLight, New Easight, and Tenda Communication. All three companies' stock prices have surged significantly this year, with a combined market value once exceeding Moutai. InnoLight's forward P/E is about 66 times, Tenda Communication's is 139 times, with valuations based on expectations of sustained high-speed growth in AI networking.

Fuyang Chen's latest statement suggests that even if AI compute demand remains robust, the networking segment's share may peak first. If buy-side investors accept this signal, the previous valuation premium of A-share optical module leaders could face direct scrutiny.

Ripple effect: Marvell declines after hours, Asian AI supply chain under pressure today

The guidance effect from Broadcom has already begun to spill over. Marvell's stock fell about 9% after hours, narrowing to around 6% decline before press time. Other companies in the AI networking/connection space, such as Astera Labs and Credo Technology, also came under pressure after hours. Notably, Marvell surged 32% in one day on June 2 after Nvidia CEO Jensen Huang called it "the next trillion-dollar company"; on June 3, the stock continued to rise 3.73%, but the after-hours pullback indicates that the previous day's Nvidia premium was being largely realized.

For Asian markets, two key points today: first, whether the A-share optical module leader "Yizhongtian" can absorb Fuyang Chen's comments on the declining network share; second, whether Korean chipmakers SK Hynix, Samsung Electronics, and other HBM suppliers are affected by the overall cooling of AI narratives. Considering that InnoLight's trading volume on June 2 exceeded half of the entire A-share sector's daily turnover, sector sentiment reactions could be amplified.

However, the earnings report itself does not negate the long-term prosperity of AI compute demand. During the call, Fuyang Chen again described AI chip demand as "difficult to satisfy," and reaffirmed the goal of over $100 billion in AI chip revenue in FY2027. UBS and other institutions also adopted a "buy on dips" approach after similar declines following Broadcom's earnings last December. Whether this correction marks a narrative turning point or is just a routine profit-taking of high-valuation stocks remains to be seen, and will depend on subsequent earnings calls from leading companies and capital expenditure trends among large cloud providers.