Broadcom AI Chip Analysis: The ASIC vs GPU Repricing Behind AVGO's Earnings Surpassing Expectations but Plunging and the NVDA, MRVL Landscape

When a leading company that holds 70% of the customized AI chip market reports revenue and profit figures that surpass expectations, yet its stock plunges by 15% after hours and drags the entire semiconductor sector to erase over $1 trillion in market value in a single day—this reveals not a deterioration in business fundamentals, but a pivotal turning point marking a new phase of “expectation game” in the AI capital market.

After the market close on June 3, 2026, Broadcom (AVGO) released its Q2 earnings report that drew widespread attention. Total revenue reached $22.19 billion, up 48% year-over-year, exceeding analyst expectations of $21.13 billion; adjusted earnings per share (EPS) were $2.44, also above the expected $2.40. Revenue from AI semiconductors hit a record $10.8 billion, a 143% YoY increase. Yet, despite this seemingly stellar performance, AVGO’s stock plummeted over 13% the next day, triggering a 10.3% drop in the Philadelphia Semiconductor Index (SOX)—the largest single-day decline since March 2020’s COVID-19 outbreak—and the entire chip sector’s market cap evaporated by more than $1 trillion.

This is not a breakdown of business logic but a misalignment of market expectations. Understanding the logic behind this “unexpected plunge” is crucial for recognizing the structural shifts in the AI chip track, grasping the strategic divergence among the three major players (Broadcom, Nvidia, Marvell), and gaining insights into the ongoing capital rotation between crypto assets and AI stocks.

Deep Dive into Q2 Earnings: Where Are the “Hidden Worries” Behind the Surprises?

Financial Data Overview

| Indicator | Value | YoY / MoM | Market Expectation | | --- | --- | --- | --- | | Total Revenue | $22.19 billion | +48% YoY | $21.13 billion | | Adjusted EPS | $2.44 | +54% YoY | $2.40 | | AI Semiconductor Revenue | $10.8 billion | +143% YoY | Surpassed expectations | | Adjusted EBITDA Margin | 69% | Above guidance of 68% | 68.6% |

Broadcom’s Q2 results exceeded expectations across all key financial metrics. AI semiconductor revenue of $10.8 billion included $5.6 billion from customized AI accelerators (XPU/ASIC), up 140% YoY, and $2.8 billion from AI networking chips, up 60%. During Q2, new bookings for AI semiconductors exceeded $30 billion, significantly higher than shipments during the same period.

Management set Q3 AI semiconductor revenue guidance at $16 billion, representing over 200% YoY growth, and maintained the long-term goal of “exceeding $100 billion” in AI semiconductor revenue for FY2027. However, the Q3 guidance of about $16 billion was below Wall Street’s expectations of approximately $17.2 billion—this less than 10% shortfall became the core trigger for the market’s intense reaction.

Why did the stock plunge?

The market was not pricing in a “slowing of Broadcom’s AI business,” but rather “a lack of sufficient optimism.” The specific reasons can be summarized into three layers:

First, over-hedging valuation before Q2. In the seven trading days prior to the earnings release, AVGO’s stock surged over 15%, adding about $300 billion in market cap. The market had already priced Q3 AI guidance at a higher level; when actual data was slightly weaker—though still within the company’s usual conservative disclosure style—it still caused a negative shock after being “priced in” at an overly optimistic expectation.

Second, reference effect from competitor Marvell. Just a week before the earnings, Marvell had raised its AI ASIC revenue outlook. This comparison intensified investor disappointment over Broadcom’s failure to similarly upgrade its guidance.

Third, refusal to raise 2027 AI revenue targets. Although management stated during the call that the 2027 goal “remains very promising,” their refusal to upgrade to a higher range disappointed investors with high expectations.

Macro overlay: Strong Non-Farm Payroll report squeezes tech stock valuations

Beyond the structural negative feedback from the earnings report itself, the macro environment’s concurrent changes amplified the decline. During the same period, the US released June non-farm employment data that exceeded expectations, fueling market expectations that the Federal Reserve would maintain high interest rates. The high-rate environment, through elevated discount rates, directly lowered the present value of future cash flows for tech stocks, adding extra pressure on the previously soaring semiconductor sector.

This means the June decline in chip stocks was the result of a dual effect: “earnings expectation deviation” combined with “macro rate re-pricing,” rather than a signal of slowdown in AI infrastructure development.

Core Differentiation: Customized ASICs (AVGO/MRVL) vs General-Purpose GPUs (NVDA)

To understand why Broadcom remains regarded as a core supplier of AI infrastructure after the market volatility, one must start from the fundamental business model differences with Nvidia and Marvell.

Fundamental Market Positioning Divide

• Nvidia (NVDA): A general GPU provider, selling not just chips but a complete cluster-level solution including GPUs, CUDA software stack, and networking interconnects (Spectrum-X / Quantum-X).
• Broadcom (AVGO): A custom ASIC (XPU) designer, deeply customizing inference-specific chips for large-scale clients, also providing high-end AI network switches. It does not directly compete with Nvidia in the end-market GPU space but builds differentiated dedicated compute solutions for customers.
• Marvell (MRVL): A second-tier custom AI chip vendor, mainly serving clients like Amazon (Trainium) and Microsoft (Maia), and through strategic Nvidia investments, establishing an important position within the NVLink-compatible interconnect ecosystem.

Market share and customer structure comparison

Based on comprehensive analyses from multiple industry sources in the first half of 2026:

| Company | Custom AI ASIC Market Share | Key Clients / Projects | AI Revenue Guidance (2026) | | --- | --- | --- | --- | | Broadcom | 55%-60% (overall ASIC) / 64% (AI server ASIC) | Google TPU, Meta MTIA, OpenAI, Anthropic, ByteDance | About $56 billion (full FY2026) | | Marvell | 13%-15% (overall ASIC) | AWS Trainium, Microsoft Maia | About $11 billion (AI ASIC) | | Nvidia | Not applicable (general GPU market ~70% share) | Major large-scale clients | Significantly higher than above (entire GPU market) |

Broadcom and Marvell together control roughly 95% of the custom AI ASIC design market, forming an effective duopoly. Broadcom holds about 70% of this share, making it the industry leader.

A particularly noteworthy structural change: Marvell secured some design orders for Google’s next-generation TPU in 2026, breaking the previous exclusive supply arrangement with Broadcom. Meanwhile, Nvidia’s strategic investments in Marvell to strengthen its interconnect ecosystem give this second-largest ASIC vendor additional strategic backing, accelerating its influence in the market structure.

Structural benefits of inference era for ASICs

Market research indicates: By 2026, custom ASIC shipments are expected to account for 27.8% of the AI chip market, with a YoY growth of 44.6%; in contrast, general-purpose GPUs will grow about 16.1% YoY. The growth gap reflects a structural shift where AI inference workloads are tilting from training to inference—an environment characterized by continuous, high-frequency, cost-sensitive scenarios. Large-scale clients, motivated by long-term TCO optimization, have a strong incentive to adopt dedicated ASICs over expensive general-purpose GPUs.

Broadcom’s management disclosed during the call that they have secured six core large language model (LLM) platform clients—including Google, Meta, OpenAI, Anthropic, and two unnamed clients. OpenAI plans to deploy customized XPUs in the second half of 2026 through 2027, committing to a total of 10 GW of compute capacity, with 1.3 GW of that in the initial deployment; Meta aims to deploy 3 GW of MTIA XPU compute capacity by the end of 2028, with the first 1 GW order starting delivery in the second half of 2027. These deployment plans provide clear, quantifiable mid-term revenue expectations for Broadcom’s custom AI chip business.

AI Networking: Broadcom’s “Invisible Fortress”

Beyond custom ASICs, AI networking chips constitute another key differentiator for Broadcom. In Q2 2026, AI networking chips contributed nearly 40% of AI revenue; management expects this proportion to fall back to about 30% after XPU accelerators ramp up.

The core product Tomahawk 6 is the world’s first 102.4 Tb/s switching chip, built on 3nm process with 200G SerDes, and is currently the only mass-produced switch chip of this level, capable of supporting over 1 million XPU clusters. Nvidia’s Spectrum-X solution, while generating around $10k annually, still lags in product maturity and deployment progress compared to Tomahawk 6.

Financial Panorama Comparison

| Dimension | Broadcom (AVGO) | Nvidia (NVDA) | Marvell (MRVL) | | --- | --- | --- | --- | | Revenue Scale | ~$55.8 billion annually (including software) | Leader in AI GPU market (~$70+ billion in 2025) | About $8 billion in 2025 (total) | | Gross Margin | 60%+ (semiconductors) + high-margin software | GPU gross margin over 70% | Around 60% target | | Software Ecosystem | VMware (~$27 billion annual software revenue) | CUDA ecosystem (not independently monetized software) | No independent software layer | | Moat Type | Customization + network chips + software bundling | GPU performance + CUDA ecosystem | Customization + interconnect optical communication + Nvidia ecosystem binding |

Crypto vs AI Stocks: Capital Rotation Accelerates

Data Evidence: Capital Flows Out of Crypto Assets

Coinshares data shows that as of late May 2026, weekly net outflows from digital asset ETPs totaled about $1.47 billion, with roughly $1.32 billion from Bitcoin products. Market monitoring by Gate.io indicates that over the past six months, the overall crypto market cap shrank by approximately $1.16 trillion, while major AI company financing totaled around $140 billion.

AXT Inc. (a supplier of semiconductor substrates for AI data center chips) saw its stock surge over 5100% in 12 months, while Bitcoin and Ethereum fell nearly 40% during the same period—creating the largest divergence in history. This gap reflects not short-term trading but a structural reallocation of capital.

Institutionalization Trend: Prioritizing “Safe AI Exposure”

There is a clear capital tug-of-war effect between crypto ETFs and AI growth stocks. Multiple analyses suggest that funds previously heavily invested in digital assets are increasingly shifting toward semiconductor and AI infrastructure companies. This shift is not just “hot money switching” but driven by:

• AI infrastructure companies’ business models resembling the classic “capital expenditure → operating income” cash flow framework;
• Crypto assets being constrained by macro factors like high interest rate expectations and regulatory uncertainties;
• Institutional funds dynamically rebalancing between the two asset classes based on risk-adjusted returns.

Massive Spending: Structural Demand for AI vs Crypto

In January 2026, Meta announced that its AI-related capital expenditure would reach about $135 billion—almost double the previous year. Other cloud providers are also significantly expanding their AI infrastructure budgets. This massive spending provides a solid fundamental demand base for AI semiconductor companies (including Broadcom, Nvidia, Marvell).

Unlike crypto assets’ vulnerability to interest rate sensitivity, AI data center capital expenditure is “structural and necessary,” not optional, creating a structural driver for capital rotation toward AI.

Analyst Target Prices and Market Consensus

Ratings and Target Prices (Updated Post-Earnings)

After the earnings release, several Wall Street firms adjusted their target prices for AVGO:

| Firm | Rating | Target Price | Core Logic | | --- | --- | --- | --- | | Goldman Sachs | Buy | $525 | ~30% upside; strong AI ASIC + networking business, extended AI capex cycle | | Bank of America | Buy | $530 | Raised from $450; upgraded AI demand + Broadcom’s pricing power | | Morgan Stanley | Overweight | $485 | Raised from $470; valuation already reflects some AI premium, cautious recovery | | Wells Fargo | Buy | $545 | Dual drivers: AI semiconductors + VMware software cash flow | | Truist | Buy | $550 | AI ASIC growth exceeds market expectations; strong AI switch demand | | Evercore | Outperform | $582 | Highest target; revaluation of “core AI infrastructure supplier,” most optimistic scenario | | UBS | Buy | $500 | Structural growth confirmed, but short-term valuation somewhat high |

According to MarketBeat’s aggregation of 33 analysts, the current consensus rating for AVGO is “Moderate Buy,” with an average target price of $490.13. Goldman Sachs reaffirmed its Buy rating and raised the target to $525, believing the post-earnings decline offers about 30% upside.

Risk Variables

The current market divergence centers on three key uncertainties:

• Profit margin compression: Custom chip gross margins are lower than those of traditional network chips and software. Management’s Q3 guidance shows EBITDA margin dropping from 69% in Q2 to 68%, raising concerns this trend may continue.
• Intensified competition from Marvell: The diversification of Google TPU orders signals that Broadcom no longer has exclusive supply dominance in flagship ASIC projects for top clients. Nvidia’s strategic investments in Marvell also bolster its interconnect tech, increasing Marvell’s influence.
• Valuation re-rating in a high-interest-rate environment: The SOX index’s 10.3% single-day drop on June 5 indicates that even with strong fundamentals, over-anticipated AI chip stocks face systemic valuation corrections.

Conclusion

The market turbulence triggered by Broadcom’s Q2 earnings underscores a core conclusion: the long-term demand structure for AI chips remains robust, but the short-term capital market has entered a “expectation race” phase. Broadcom’s dominance in the custom ASIC market, with key clients locked in for years, and the mass production of Tomahawk 6 in AI networking chips—establishing a clear lead—these fundamentals remain unchanged despite the stock’s volatility.

What has truly changed is the market’s pricing rhythm—when AVGO surged over 15% in the seven trading days before earnings, when the custom ASIC market faced breakthrough competition from Marvell, and when capital continues to rotate from crypto into AI infrastructure—investors are no longer questioning “whether AI demand exists,” but rather “whether expectations are fully priced in.”

For market participants, understanding the structural differences among Broadcom, Nvidia, and Marvell in the compute paradigm is more valuable in the long run than focusing on a few hundred million dollars’ worth of quarterly guidance errors. The increasing penetration of custom ASICs in inference, the cluster deployment of AI network chips, and the long-term XPU compute agreements with large clients—these are the core drivers of Broadcom’s next growth phase.