Broadcom: Sideways for Nearly Two Months After Earnings, Is Now a Buying Opportunity?

Why does the stock price remain under pressure despite strong financial results?

In June, Broadcom delivered an earnings report with almost no obvious weaknesses: FY2026 second-quarter revenue grew 48% year-over-year, AI semiconductor revenue grew 143% year-over-year, and AI revenue growth guidance for the next quarter further exceeded 200%.

However, the capital market's reaction was completely the opposite. Following the earnings release, Broadcom's stock price tumbled 13%, continued to pull back in June, and entered range-bound trading in July; as of July 28, it closed at $380.91, representing a cumulative decline of approximately 21% compared to its pre-earnings level.

The issue lies not in the performance, but in expectations. The market had previously bet that Broadcom would continuously raise its long-term AI targets and quickly realize revenue from new customers beyond Google. However, after the earnings release, management maintained its FY2027 AI semiconductor revenue target of over $100 billion without providing more aggressive new guidance, meaning the strong financial results failed to trigger new upward revisions in expectations.

Supply share for Google's TPU is also an important factor behind the recent pressure on the stock price. Following MediaTek's entry into Google's supply chain, the market began to worry that Broadcom's share would rapidly drop to 50% or even be phased out over time.

Morgan Stanley believes this concern is overly aggressive. While it is true that MediaTek has become a second supplier, Broadcom is still likely to maintain an approximate 80% TPU share in the long run. The market's direct interpretation of a 'new supplier' as 'Broadcom will be replaced on a large scale' may have overestimated the pace of market share loss.

Over the past two months, while the stock price was still digesting valuations, the core business continued to advance. The Google TPU roadmap remained stable, OpenAI's custom XPU project entered its next phase, and demand for Tomahawk 6 outstripped supply.

This constitutes the most noteworthy divergence in expectations at present: the stock price has stalled, while fundamentals continue to forge ahead.

We believe the post-earnings consolidation is providing a better entry window. Broadcom remains a high-expectation, high-volatility stock, but the current price has already priced in some of the optimism, while there is still potential for future earnings to be revised upward.

What kind of company is Broadcom? Where does its revenue growth come from?

Broadcom's financial reports disclose only two official segments:

1. Semiconductor Solutions
2. Infrastructure Software

In the second quarter of FY2026, Semiconductor Solutions revenue was $15.0 billion, accounting for 68% of group revenue, while Infrastructure Software revenue was $7.2 billion, representing 32% of group revenue. Within this, AI semiconductor revenue reached $10.8 billion. Back-calculating from company disclosures, non-AI semiconductor revenue was approximately $4.2 billion.

To understand the business more clearly, we have broken down the revenue into five research tracks:

The semiconductor business has established a structure of 'AI-driven growth supported by traditional businesses'

Broadcom's semiconductor revenue can be directly divided into two parts:

• AI Semiconductors: $10.8 billion, accounting for 49% of group revenue and 72% of semiconductor revenue
• Traditional Semiconductors: $4.2 billion, accounting for 19% of group revenue and 28% of semiconductor revenue

AI semiconductors include custom AI chips and AI Networking, which are currently the primary drivers of incremental revenue. Traditional semiconductors cover wireless, broadband, storage, and industrial chips; although they grow at a slower pace, they have longer product lifecycles and more stable customer relationships.

Growth in AI semiconductors primarily comes from three directions:

• Google's TPU continues to iterate, with Broadcom expected to maintain its position as the primary supplier;
• OpenAI and other new customers are gradually entering the validation and mass production stages;
• Tomahawk 6 is in short supply, making AI Networking an independent source of growth.

Traditional semiconductors, on the other hand, play the role of stabilizing revenue. In July, Apple expanded its multi-year partnership with Broadcom, covering custom silicon products and advanced wireless connectivity technologies; storage, broadband, and traditional networking businesses also have room for cyclical recovery.

The software business serves as Broadcom's high-margin cash flow foundation

Infrastructure Software revenue in the second quarter was $7.18 billion, up 9% year-over-year, accounting for 32% of group revenue. The bulk of this comes from VMware, with the remainder consisting of enterprise software assets such as CA and Symantec.

Among them, VMware is the main body of the software business. VMware Cloud Foundation integrates server virtualization, storage, networking, and cloud management into a single platform, which large enterprises can use to build and manage private clouds. Since a large number of core applications run in VMware environments long-term, customer migration requires reconfiguring systems, data, and security architectures, resulting in high switching costs.

Once software product development is completed, new subscriptions and renewals do not require a synchronous increase in investment in wafers, packaging, or production equipment. Contract terms are longer, and revenue is more predictable than chip orders.

Therefore, Broadcom's current business structure can be summarized as:

AI semiconductors drive growth, traditional semiconductors stabilize revenue, and infrastructure software improves the predictability of profits and cash flow.

Driver 1: Google defends its core business, while new customers lift the growth ceiling for ASICs

Broadcom's ASIC logic can be summarized into three questions: why do customers need custom chips, which customers are adopting them, and whether Broadcom can maintain its competitive advantage.

Why are large AI customers starting to increase their use of ASICs?

The advantage of GPUs lies in their high versatility, allowing them to quickly support different models and tasks. ASICs, on the other hand, are designed around specific workloads, offering relatively lower flexibility but further optimizing performance, power consumption, and cost.

For large customers such as Google, Meta, and OpenAI, ASICs primarily solve three problems:

• Reducing the computing cost per token;
• Increasing the computing power per unit of electricity;
• Reducing reliance on a single GPU supplier.

This economic value is particularly evident in the inference phase. Training is typically concentrated in the model development phase, whereas inference demand continues to grow with the number of users, invocation frequency, and penetration of AI applications. The more stable the task and the larger the scale, the stronger the customer's incentive to deploy dedicated chips.

Large customers will not simply replace GPUs with ASICs; they are more likely to form a portfolio:

GPUs handle general and rapidly changing tasks, while ASICs take on the largest, most repetitive workloads with the highest cost pressures.

J.P. Morgan pointed out after communicating with Broadcom management in July 2026 that leading customers are accelerating tape-out schedules for next-generation XPUs and developing more chip versions for different inference scenarios. Management expects that for some frontier model developers, the shipment ratio of XPUs to GPUs could approach 50:50 by 2027.

Which customers are currently using Broadcom's ASICs?

Google is Broadcom's most mature customer. The two parties have co-developed multiple generations of TPUs, with the next-generation TPU v9 still progressing as planned, and Broadcom expects to continue holding the majority of the supply share over the long-term partnership.

The biggest market debate currently is how much share MediaTek will capture after entering Google's supply chain.

Morgan Stanley's base case is that Broadcom can still maintain about an 80% TPU share in the long run, with MediaTek more likely acting as a secondary supplier. Entering the supply chain does not equate to being able to quickly replace Broadcom: after a chip completes tape-out, it still needs to secure HBM and advanced packaging capacity, pass system validation, and run stably in large-scale clusters.

Therefore, what investors truly need to focus on is:

Whether the rate of share decline is faster than the growth rate of Google's overall TPU demand.

As long as demand for Gemini, AI Search, Google Cloud, and external TPUs continues to expand, even if Broadcom's share declines slightly, related revenue is still likely to grow.

Beyond Google, new customers are gradually entering the validation and mass-production phases:

What are Broadcom's advantages compared to its competitors?

Competitors in the custom chip market include Marvell, MediaTek, and also customers' in-house chip teams.

Broadcom's advantage lies not just in completing chip designs, but in rapidly transforming customer-proposed architectures into products ready for mass production and large-scale deployment.

This capability is primarily reflected in three aspects:

• Design experience. Broadcom has already partnered with large customers across multiple generations of products, possessing extensive experience in developing complex AI chips;
• Network synergy. The company simultaneously owns switching chips, high-speed interfaces, and interconnect products, enabling it to deliver computing and networking solutions together;
• Mass production and supply chain capabilities: Broadcom can secure HBM and advanced packaging resources in advance, and complete system-level verification and large-scale delivery.

Customers define models, workloads, and chip architecture, while Broadcom is responsible for chip implementation, network connectivity, and mass production delivery.

What Broadcom actually sells is not just a chip, but the ability to transform a customer's AI chip design into a large-scale computing system.

Driver 2: AI Cluster Expansion; Networking Could Be the Next Source of Upside Surprises

GPUs and ASICs can be understood as production equipment in AI factories, while the network is responsible for transmitting data between these devices.

When cluster sizes are small, the impact of network congestion is limited. However, once clusters expand to tens or even hundreds of thousands of accelerators, if data transmission speeds cannot keep pace, expensive computing chips will sit idle, and the utilization rate of the entire cluster will drop. Therefore, when customers increase AI computing power, they also need to simultaneously upgrade switches, network interface cards (NICs), and optical connectivity equipment.

Among these, switch chips are the core components of networking equipment, responsible for forwarding data at high speeds among GPUs, ASICs, and servers. Traditional data centers also use Ethernet, but AI clusters require a massive number of accelerators to collaborate simultaneously, demanding higher bandwidth, lower latency, and greater stability, which in turn requires higher-performance AI Ethernet switch chips.

Broadcom's Tomahawk series is the primary product in this market. The latest-generation Tomahawk 6 is designed for hyperscale AI clusters, offering a single-chip switching capacity of 102.4 Tbps, which is twice that of the previous generation.

As clusters expand from tens of thousands of chips to 100,000 or even larger scales, the demand for networking equipment is expected to grow faster than the number of accelerators, driven primarily by three factors:

• Increased number of connections: The more accelerators and servers there are, the more switch ports, NICs, and optical connections are required;
• Increased network tiers: Large clusters require more layers of switching equipment to connect tens or even hundreds of thousands of chips;
• Product specification upgrades: Networks are evolving from 800G to 1.6T, with higher-speed products offering higher value per port.

Broadcom currently leads the AI Ethernet switch chip market. J.P. Morgan estimates that the company's share in this market is approximately 70%; major frontier model customers are adopting Tomahawk 6, and demand currently outstrips supply.

The networking business has another major advantage: customers do not need to use Broadcom's custom chips. Even if a data center adopts accelerators from Nvidia or AMD, it may still purchase Broadcom's switching and connectivity products.

XPUs determine whether Broadcom can enter a customer's core AI projects, while Networking determines how much revenue the company can generate from the overall cluster expansion.

The market is currently more focused on Google TPUs and OpenAI XPUs, but with the expansion of cluster sizes, upgrades in network speeds, and Tomahawk 6 in short supply, AI Networking could become the business segment most likely to drive upward revisions in earnings forecasts over the coming quarters.

Driver 3: The Core Focus for VMware Is Whether Earnings Improvement Can Sustain

VMware is a leading global provider of enterprise virtualization and cloud infrastructure software. Its core products help enterprises manage servers, storage, networking, and private clouds within a single IT system. Because the core applications of many large enterprises have long run in VMware environments, customer migration costs are high, resulting in relatively strong renewal rates and revenue stability.

Broadcom completed its acquisition of VMware in November 2023. The core purpose of the acquisition was to expand its high-renewal, high-cash-flow infrastructure software business, reduce the group's dependence on the semiconductor cycle, and further penetrate the enterprise private and hybrid cloud markets.

Following the completion of the acquisition, Broadcom quickly streamlined products and channels, accelerated the transition to subscription models, and focused resources on large customers and VMware Cloud Foundation.

This strategy may lead to a decline in customer count, but it can also increase the contract value and earnings contribution of individual customers.

To evaluate VMware's performance in the future, the focus will be on three key areas:

• Whether large customers continue to renew;
• Whether the adoption rate of VMware Cloud Foundation can increase;
• Whether price increases and product bundling accelerate customer churn.

Currently, the greatest risk comes from small and medium-sized enterprise (SME) customers. They are more likely to transition to other virtualization platforms or public clouds, and are more sensitive to price increases. Large enterprises face higher migration costs, making their renewals relatively stable, but Broadcom still needs to prove that price hikes will not damage its customer base over the long term.

For investors, the significance of VMware lies in enhancing the stability of Broadcom's earnings. Although AI semiconductor revenue is growing rapidly, factors such as project timelines, customer capital expenditures, and supply chains can introduce volatility. In contrast, VMware provides more predictable profits and cash flows through long-term contracts and low capital requirements.

In the first half of FY2026, Broadcom generated approximately $18.3 billion in cumulative free cash flow. While VMware is not the sole source of this cash flow, its high-margin and low-capex profile provides vital support for the group's R&D, debt repayment, dividends, and share buybacks.

Therefore, VMware does not need to be a new high-growth business. As long as revenues remain stable and cost controls continue to improve, it can consistently enhance Broadcom's overall earnings quality.

Key Risks: Growth Realization Remains Uncertain

1. Customer Concentration Amplifies the Volatility of Google Projects

Broadcom's custom AI chip revenue remains highly dependent on Google. Even if total demand for TPUs continues to grow, Google introducing more suppliers, adjusting its purchasing rhythm, or delaying the deployment of a single generation of products could cause significant fluctuations in Broadcom's revenue recognition.

The impact of this risk is not limited to market share. Custom chip projects typically involve large amounts and long cycles; thus, changes in a single customer or a single product generation can affect quarterly revenue, product mix, and profit margins. Current FY2027 earnings expectations already assume continuous growth of Google TPUs. If order momentum or the value of projects secured by Broadcom falls short of expectations, the market may simultaneously downgrade both revenue forecasts and valuation multiples.

2. Volume Production for New Customers May Be Later Than Expected

After projects like OpenAI's complete tape-out, they still need to undergo verification, packaging, and system deployment, meaning significant revenue contribution is still some time away. Jefferies expects its first-generation chip to start shipping by the end of 2026, but high-volume production may not occur until the second half of 2027. If the production timeline is delayed, revenue and earnings expectations for new customers in FY2027 may be revised downward.

3. AI Networking Growth Depends on the Penetration Rate of Ethernet Solutions

Broadcom benefits from AI cluster expansion and the upgrade from 800G to 1.6T, but competition in the networking market is equally intense. Nvidia is simultaneously pushing InfiniBand and its own Ethernet solutions, while increasing the bundling of computing, networking, and software through integrated systems. If customers increasingly adopt Nvidia's integrated architecture, the connection ratio of Broadcom's switch chips within individual clusters could come under pressure.

Valuation: Risk-Reward Ratio is Currently Balanced; Upside Depends on Continued Earnings Realization

Broadcom's current stock price is around $381. Based on the consensus estimate of approximately $19.51 for FY2027 adjusted EPS, this corresponds to a P/E multiple of around 19.5x.

This valuation is not considered aggressive, but the $19.51 EPS estimate itself already assumes continued growth of Google TPUs, revenue contributions starting from new customers, and sustained strength in AI Networking. Therefore, whether the current valuation is attractive still primarily hinges on whether FY2027 earnings can be realized.

Using the P/E valuation method with the consensus estimate of $19.51 as the Base Case, the Bear and Bull cases are revised downward and upward by 15%, respectively. Based on historical valuations, peer comparisons, and earnings certainty, valuation multiples of 20x, 24x, and 28x are applied respectively.

Bear Case: Project delays, valuation declines

If Google's share declines, mass production for new customers is delayed, and AI networking growth falls short of expectations, FY2027 EPS could be 15% below consensus, dropping to $16.58. Applying a 20x P/E multiple yields a target price of approximately $332, representing a downside of about 13% from the current share price.

Base Case: Core projects delivered as expected

The base case utilizes the consensus EPS estimate of $19.51. Assuming Google maintains its position as the primary supplier, new customers progress as planned, and AI networking continues to grow, applying a 24x P/E multiple yields a target price of approximately $468, representing an upside of about 23% from the current share price.

Bull Case: Earnings revised upward, valuation returns near historical average

If Google's deployment exceeds expectations, OpenAI and other customers scale up production faster, and AI networking continues to outperform, FY2027 EPS could be 15% above consensus, reaching $22.44. Applying a 28x P/E multiple yields a target price of approximately $628, representing an upside of about 65% from the current share price.

Conclusion and Investment Recommendations: Sideways consolidation is better suited for accumulating positions in tranches

Broadcom is currently not significantly undervalued, and future returns still depend on whether AI growth can materialize. However, the stock price correction following the earnings release has eased some valuation pressures, and the current risk-reward ratio has improved compared to the June peak.

If I were an investor, I would prefer to build positions in tranches rather than going all-in at once. Although the core investment thesis has not changed significantly, uncertainties remain regarding Google's TPU market share, the timing of volume production for new customers, and AI networking supply. Therefore, managing position sizing and pacing is more critical than simply predicting market direction.

From a fundamental perspective, Google's TPU remains the core anchor for near-term revenue, while new customers will determine mid-to-long-term growth potential, and AI networking could become a major source of earnings upgrades in the next phase. Meanwhile, uncertainties persist regarding Google's share, the timing of volume production for new customers, and the supply of Tomahawk 6. Consequently, the short-term stock price may continue to be influenced by project progress and shifts in market expectations.

The three most important indicators to watch going forward are:

1. Whether Broadcom can maintain its position as the primary supplier with a 70% to 80% share in Google's TPU project;
2. Whether OpenAI and other new customers can transition to volume production in the second half of 2027;
3. Whether the supply of Tomahawk 6 can improve and convert the backlog of demand into revenue.

We view the current sideways consolidation as an accumulation window for building mid-to-long-term positions in tranches. The base case already offers around 23% upside, and once concerns over Google's share subside and new customers alongside networking drive earnings upgrades, the stock price will possess even greater upside potential.