Broadcom's AI guidance falls short of expectations, triggering a pullback: ASICs lead GPU in growth, AI computing enters a structural shift cycle

On June 3, 2026, Broadcom, the leading AI custom chip company, released its highly anticipated Q2 financial report that drew widespread market attention: revenue of $22.187 billion, up 48% year-over-year; AI semiconductor revenue of $10.8 billion, up 143%, reaching a historical high; and adjusted EPS of $2.44, exceeding expectations. All three core metrics—total revenue, earnings per share, and AI revenue—outperformed market consensus.

However, after the close in U.S. Eastern Time on June 4, AVGO’s share price plunged by about 13%. The after-hours quote fell to around $418, triggering a drop of more than 10% in the entire Philadelphia Semiconductor Index and causing the chip sector’s market capitalization to evaporate by over $1 trillion in a single day. What the market rewarded was not “good performance,” but “exceeding high expectations.” In the seven trading days before Broadcom’s earnings release, its stock price surged by more than 15% and its market cap expanded by about $300 billion. The capital market had already priced it as a “perfect expectation.” When management provided Q3 AI chip sales guidance of $16 billion—though the year-over-year growth rate was over 200%—it still came in below Wall Street’s $17.2 billion consensus. In addition, management refused to raise the 2027 fiscal year AI revenue target—“exceeding $100 billion”—which remained unchanged. This combination happened to trigger the “downside switch” most sensitive to overvalued stocks. Trading volume surged to 79.9 million shares that day, about 214% of the average daily trading volume over the past three months.

Q2 Earnings Recap: The Past Was Above Expectations; the Future Is Below Expectations

According to Broadcom’s official earnings report and management’s remarks during the June 3, 2026 earnings call, the second-quarter figures are as follows:

• Total revenue of $22.187 billion, up 48% year-over-year, slightly above market expectations of $22.13 billion;
• Adjusted earnings per share (non-GAAP) of $2.44, up 54% year-over-year, higher than the expected $2.40;
• AI semiconductor revenue of $10.8 billion, up 143%. Of this, custom AI accelerators (XPU/ASIC) contributed $5.6 billion, and AI network chip revenue was $2.8 billion;
• New AI semiconductor bookings in the period exceeded $30 billion, significantly higher than the actual shipments of $10.8 billion;
• Overall adjusted EBITDA profit margin of 69%, exceeding the company’s prior guidance of 68%;
• Infrastructure software revenue of $7.178 billion, up 9% year-over-year, slightly below market expectations of $7.32 billion.

For Q3 revenue and AI business guidance, Broadcom expects overall revenue in the third quarter to be about $29.4 billion, up 84% year-over-year, above market expectations of $28.6 billion; non-GAAP operating margin is expected to remain at the 67% level. But the AI chip business guidance most sensitive to the market was $16 billion—below the $17.2 billion average market expectation compiled by Bloomberg, a difference of about $1.2 billion, or a shortfall of roughly 7%. During the call, management further confirmed that full fiscal year 2026 AI semiconductor revenue is expected to reach $56 billion, up about 180% year-over-year. However, it also failed to match market expectations of around $57.7 billion.

The long-term target of AI semiconductor revenue exceeding $100 billion for fiscal year 2027 was not adjusted. For investors who had already priced in an “early upward revision,” management’s failure to proactively raise expectations directly triggered a sell-off. When asked why gross margin was under pressure, CFO Kirsten Spears said that as shipments of custom AI chips such as TPU increase, overall gross margin will narrow from 77% in the second quarter to about 74%. Fundamentally, this is driven by changes in business structure rather than an efficiency deterioration.

Separating from the basic fundamentals behind the short-term stock price volatility, it is worth sorting through the customer collaboration details disclosed in the earnings report. On the Google side, in April, multiple-generation TPU and AI network long-term development supply agreements were signed, covering more than 1 gigawatt of compute capacity in 2026 and 5 gigawatts starting in 2027. The TPU access agreements associated with Anthropic involve a compute deployment of over 1 gigawatt in 2026 and 5 gigawatts starting in 2027. OpenAI’s agreement to go into production by the end of 2026 includes a deployment commitment of 1.3 gigawatts in January 2027, as part of a total 10-gigawatt cooperation. For Meta, Broadcom expects to deploy 3 gigawatts of compute capacity related to MTIA XPU before 2028, with the first 1 gigawatt order beginning delivery in the second half of 2027. The long-term partnership relationships with six core custom chip customers have not wavered; outstanding orders have extended into multiple subsequent fiscal years.

The Three-Layer Logic Behind the Plunge: An Expectation Gap, the Reference Frame, and the Resonance of the Macro Environment

At the surface level, the logic is that Q3’s single-quarter AI revenue guidance fell below market consensus. But the deeper reasons are threefold.

First, valuation preloading is overloaded. In the seven trading days before the earnings report, AVGO’s cumulative rise exceeded 15%, and its market cap expanded by about $300 billion, already pricing in Q3 AI performance surpassing expectations. After actual data landed in the company’s consistently conservative forecast range—even if it still represented strong growth—it could no longer support the elevated valuation. CFRA Research semiconductor analyst Angelo Zino’s core judgment is that the after-hours stock decline mainly reflects the market’s extremely high expectations before the earnings, rather than a real deterioration in the business.

Second, the reference effect from competitors. In its May 27, 2026 earnings report, Marvell Technology directly raised its AI ASIC revenue outlook, and on top of that, NVIDIA CEO Jensen Huang publicly backed Marvell at COMPUTEX 2026, saying it has the potential to disrupt a trillion-dollar market cap. Marvell’s stock surged by more than 30% on the single day after the earnings. Inevitably, the market produced a comparative judgment: when a competitor raised guidance, the leading company Broadcom chose not to adjust.

Third, the resonance effect of the macro environment. During the earnings release period, the U.S. May Consumer Price Index rose to 4.2% year-over-year, the highest since April 2023, strengthening expectations that the Federal Reserve will maintain a hawkish stance. The S&P 500, the Dow Jones, and the Nasdaq all faced downward pressure at the same time. The synchronized declines among semiconductor peers such as NVIDIA and Micron indicate that this is a collective adjustment of overvalued tech stocks rather than a single event targeting Broadcom.

The impact also extended across the entire semiconductor value chain. Intel and AMD widened their decline over two consecutive trading days. Meanwhile, Broadcom’s trading volume surged to 79.9 million shares, as investors are systematically re-examining the overly high AI growth assumptions. Macquarie downgraded Broadcom’s rating from “outperform the broader market” to “neutral,” citing that Google is increasing its efforts in self-developed AI chips, which could erode Broadcom’s share in the ASIC market in the long run. This concern triggered the market to reprice the risk of customer concentration for Broadcom. Marvell’s progress in custom AI chips also increases competitive tension in this duopoly-style landscape.

Interpreting AVGO’s Post-Earnings Market Trajectory

After the stock fell continuously from the historical high of $495 before the earnings, as of June 11 it had retreated about 24% from the peak. On June 10, it dropped another 5.12% in a single day, closing at $372.1 with 38.19 million shares traded—up 3.15% from the previous trading day, indicating that positions continued to loosen. On June 12, AVGO rose 3.62% during the day to $385.57. This reflected that some near-term selling pressure was temporarily released, but it had not yet returned to the pre-earnings price range.

If Q3 delivers year-over-year AI semiconductor growth of 200% or more, and management subsequently discloses more customer contract details to improve visibility, the market may reprice its long-term compound growth outlook. If competitive pressure shows up in later earnings through changes in market share or slower revenue growth, and if the risk of customer concentration continues to rise, valuation will face persistent downward pressure. Investors will track the actual AI semiconductor growth rate in the early September 2026 quarterly report, announcements about customer order expansions, and progress on Google’s self-developed chip projects as core observation points to judge the trend.

From a Single Event to Industry Structure: The Logic Behind the Repricing of the ASIC Track

Stepping away from the earnings-driven stock-price fluctuations of a single company, Broadcom’s “plunge after an above-expectations earnings report” reflects a broader industry signal: the AI chip track is evolving from a single GPU narrative into a dual logic of GPU and ASIC operating in parallel. The core driver is the change in AI workload structure. Deloitte data shows that in 2023, inference workloads accounted for about one-third of total AI workloads, rising to two-thirds by 2026. Inference scenarios demand far higher efficiency per dollar and per watt of compute than training scenarios—this is precisely the design goal of customized ASICs. Taking Google TPU as an example, in specific matrix operation scenarios, an ASIC’s performance per watt can reach 3 to 5 times that of general-purpose GPUs.

From scale data, TrendForce expects AI ASIC chip shipment growth of about 44.6% in 2026, while GPU growth is about 16.1% over the same period. Goldman Sachs forecasts that ASIC’s share of the AI chip market will rise to 40% in 2026 and exceed 45% in 2027, almost splitting the market evenly with GPUs. Counterpoint Research expects AI ASIC market shipments in 2027 to reach three times the 2024 level; in 2028, they could exceed GPUs with a volume of more than 15 million units. Regarding market size, institutions such as Goldman Sachs believe the AI ASIC market will be about $30 billion in 2026, increasing to $700–$800 million in 2027, capturing nearly half of the industry. The paradigm shift from general-purpose GPUs to customized ASICs has become a new main investment line that cannot be ignored in AI hardware.

In terms of competitive landscape, the ASIC market has formed a “Broadcom and Marvell” dual-leader structure. Broadcom holds about 70% of the custom AI chip market, with a customer roster including six core hyperscale customers such as Google, Meta, OpenAI, and Anthropic. Marvell focuses on customized projects like Amazon Trainium and Microsoft Maia, and through strategic investments from NVIDIA totaling several hundred million dollars, it has established an important position within the NVLink-compatible interconnect ecosystem. MediaTek, which operates in the mobile SoC field, has also entered the arena. Recent forecasts indicate that its AI ASIC business will contribute about $2 billion in revenue in Q4 2026, expanding to several billion dollars in 2027, targeting 10% to 15% share of the AI ASIC market by 2027. Qualcomm is accelerating its data center custom chip product line based on its AlphaWave acquisition. The ASIC supply-side landscape is shifting from a duopoly toward a multi-player contest.

Gate TradFi: A New Cross-Asset Investment Channel From Crypto to Stocks

Against the backdrop of accelerating changes in the competitive landscape in the AI chip space, investors need not only to follow developments of core targets such as Broadcom, Marvell, and NVIDIA, but also to seek more efficient, lower-barrier cross-asset allocation tools.

On June 1, 2026, Gate officially launched real stock trading services, becoming one of the first exchanges in the industry to directly integrate access to the U.S. stock market within a crypto platform. As of June 2026, Gate TradFi has launched more than 10,000 real stocks and ETFs, fully covering the five major mainstream exchanges: NYSE, Nasdaq, NYSE Arca, NYSE American, and BATS. The product lineup spans a wide range of assets—from technology giants to industry-theme ETFs. Users do not need to exchange currencies, make cross-border remittances, or open additional brokerage accounts. They can directly buy with USDT in their Gate account with one click. Gate’s U.S. stock spot trading fee is as low as 0.023%, and the VIP threshold is only $2,000. Investors holding real stocks enjoy all company action rights such as cash dividends, stock dividends, rights issues, stock splits, and bonus shares; dividend income is automatically credited to the Gate account.

The recent launch of the Direct IPO (IPO Access) service further opens doors for users to participate in early growth opportunities of high-growth companies. The first featured project is SpaceX. Users can submit intention subscription requests before the company officially lists, with payments supported in USDT. After listing, shares are directly distributed to users’ Gate stock accounts. This feature completes an end-to-end investment product ecosystem covering Pre-IPO, IPO subscription, and stock trading.

For crypto-native users who prefer round-the-clock trading, Gate also offers two tokenized stock products: xStocks and Ondo Stocks, each linked 1:1 to the prices of real stocks, enabling uninterrupted 7×24 trading that is not constrained by traditional stock market trading hours closures. Users can choose flexibly between the two products according to their configuration needs, achieving unified account management for traditional finance and digital assets.

Conclusion

The intense market reaction triggered by Broadcom’s Q3 AI guidance of $16 billion falling below market expectations should not be interpreted simply as a sign that AI infrastructure investment has peaked. A more accurate conclusion is that the capital market is shifting from “enthusiastically chasing high growth across the board” to “verifying the quality of growth one step at a time,” and the marginal expectation at the time is becoming the key pricing variable. The AI custom chip track’s own high-speed growth trend has not reversed. With inference workloads continuing to expand and hyperscale cloud customers seeking optimal TCO solutions, the structural share of the ASIC market is rising on a clearly defined slope.

The real stock trading service recently launched on the Gate platform integrates U.S. stocks into a unified account ecosystem. Combined with innovative features such as Direct IPO, it is building a comprehensive channel for investors covering core AI-themed assets and cross-asset allocation. After Broadcom’s expectation correction, the market will pay more attention to the actual growth rates of its AI business in subsequent earnings, new customer expansion progress, and changes in competitive positioning by rivals. How these factors evolve will determine the next direction of the valuation center of gravity for the ASIC dual-leader. Investors should closely track the realization of actual Q3 growth, customer order expansions, and shifts in the competitive landscape to find an allocation pace that matches their own risk preferences amid high volatility.