Broadcom FY2Q26 Earnings & More (0604)

1. Broadcom FY2Q26 Earnings ① Overview

• Core Source

“Broadcom achieved record revenue, operating income, and free cash flow in Q2, driven by accelerated growth in AI semiconductor revenue and strong operating leverage.”

“Q2 AI semiconductor revenue grew 143% year-over-year to $10.8 billion, fueled by increased demand for custom AI accelerators and AI networking, surpassing our own forecast.”

“Why does the AI backlog of $30B far exceed revenue of $10.8B? — Customers need to align power infrastructure, HBM, and wafers simultaneously → advance orders are essential. This is planned lead-time management, not a supply shortage. No cancellations. Visibility has extended from FY27 to FY28 three months ago — proof that this is grounded in real demand.”

“AI networking revenue mix expanded from 30% in the prior quarter to 40% this quarter, contributing to AI semiconductor growth.”

“Operating margin reached a record high of 67.3%, up 200bp year-over-year. Adjusted EBITDA margin was 69%, exceeding consensus and demonstrating strong operating leverage.”

• Expected Impact

Broadcom’s FY2Q26 results contain numerous data points showing that AI semiconductor demand is strengthening both quantitatively and qualitatively.

In terms of revenue structure, total revenue for Q2 grew 48% year-over-year to $22.19 billion. Semiconductor Solutions revenue was $15.0 billion (+79% YoY), with AI semiconductor revenue of $10.8 billion growing +143% YoY to account for roughly half of total revenue. Non-AI semiconductor revenue of $4.2 billion (+6% YoY) also signals cyclical recovery. The expansion of AI networking mix from 30% to 40% quarter-over-quarter is a key data point. Broadcom’s AI semiconductor revenue is composed of two pillars — XPU (custom AI accelerators) and AI networking (switches, routing, DSP) — and as AI cluster scale grows, networking demand to connect thousands of chips explodes structurally. The expanding AI networking mix therefore signals that the infrastructure build-out cycle has entered a full-scale execution phase.

On profitability, the record operating margin of 67.3% and free cash flow of $10.3 billion representing 46% of revenue demonstrate that operating leverage is working powerfully, even against the mix headwind of lower-margin semiconductor revenue gaining share versus higher-margin software.

The most notable metric is the AI semiconductor backlog of $30 billion, approximately 2.8 times the quarterly shipment of $10.8 billion, meaning several quarters of future revenue are already effectively locked in. On the earnings call, management explained that this backlog does not represent double-ordering or excess purchasing, but rather planned advance orders placed because customers must align power infrastructure, HBM procurement, and wafer supply simultaneously. The extension of demand visibility from FY27 to FY28 — which occurred this quarter — corroborates that this backlog is grounded in real end-demand.

Broadcom FY2Q26 Earnings ② Q3 AI Revenue Guidance Below Consensus

• Core Source

“Q3 AI semiconductor revenue guidance of $16.0 billion, representing growth of more than 208% year-over-year, came in 7% below consensus.”

“Broadcom guided for full-year AI semiconductor revenue of $56 billion, implying that second-half AI semiconductor revenue will double versus the first half — 3% below the annual consensus of $57.5 billion.”

“This appears to reflect that Broadcom did not raise its long-term AI revenue outlook. Broadcom is maintaining its guidance that FY27 AI revenue will exceed $100 billion. However, Broadcom’s Hock Tan CEO noted that the AI segment has the potential to exceed this guidance. The decision not to raise the guidance appears to reflect a conservative stance, not concerns about market share loss or data center construction delays.”

“Since TPU supply discussions are progressing beyond 2027 into 2028, the judgment is that consensus itself was formed too high rather than reflecting a demand slowdown. In fact, Broadcom’s second-half consensus was sharply raised after last quarter’s earnings and had been gradually coming down since.”

“We expect Broadcom’s AI revenue to grow to $125 billion in FY27, exceeding guidance, and to $190 billion in FY28.”

• Expected Impact

Broadcom’s Q3 total revenue guidance of $29.4 billion (+84% YoY, 3% above consensus) was solid, yet the AI semiconductor revenue guidance of $16.0 billion falling below consensus (approximately $17.2–18.0 billion) triggered an after-hours stock decline of 10–14%. The full-year AI semiconductor revenue guidance of $56 billion also came in below the consensus of $57.5 billion. The two core disappointments for the market were: first, the Q3 AI guidance number itself; and second, the absence of an upward revision to the FY27 AI revenue target of “over $100 billion.”

However, IB analyses converge on the view that this decline reflects a process of digesting overly elevated expectations, not demand deterioration. One key reason Q3 guidance came in below consensus was the clarification that the Anthropic contract structure is chip-only sales rather than full rack-scale sales — which the market had previously assumed. Rack-scale includes networking equipment and components beyond chips, implying higher unit values, but Broadcom formalized this quarter that it supplies only the chips it develops. This reduces near-term revenue size but is actually positive for margins.

The assessment that maintaining FY27 guidance of “over $100 billion” reflects management’s conservative tendency rather than demand decline is the dominant view. Deutsche Bank expects Broadcom to achieve $125 billion in FY27 and $190 billion in FY28, raising its target price from $430 to $515. JPMorgan forecasts over $150 billion in FY27 and approximately $300 billion in FY28, raising its target to $580. Citi stated it “views this price correction as a better buying opportunity” and maintained its target price of $500.

The broader context more important than the guidance numbers themselves is revealed by the fact that total Q3 revenue guidance ($29.4 billion) exceeded consensus. Even with AI semiconductor guidance missing, non-AI semiconductor guidance (+12% YoY expected, 7% above consensus) and infrastructure software ($8.9 billion, +31% YoY, 17% above consensus) both significantly exceeded consensus, lifting total revenue. This demonstrates that Broadcom’s revenue base is diversified beyond AI semiconductors alone.

3. Broadcom FY2Q26 Earnings ③ Google TPU Supply Diversification Acknowledged

• Core Source

“Google TPU long-term contract — is supply exclusive, and what is the likelihood of diversification? — [Answer] Some sourcing diversification acknowledged given Google’s explosive growth in AI computing consumption. However, Broadcom’s contract is a long-term commitment of ‘very substantial dollar value.’ Relationship continuity secured through technology and IP differentiation. The overall demand growth is faster than the pace of new supplier entry.”

“Broadcom and Alphabet have already been collaborating on semiconductor development and production for 10 years, and Alphabet’s TPU has been developed through to the 8th generation. However, Broadcom’s Hock Tan CEO mentioned in this earnings release the possibility that Alphabet could diversify its supply chain.”

“Broadcom, together with Apollo and Blackstone, is building the AI XPV platform, with plans to deploy more than 20GW of compute capacity for AI frontier labs by 2028.”

“Anthropic is approximately 3.5GW, OpenAI is 1.3GW. Meta MTIA’s initial ramp is also included, and purchase orders of $6 billion have already been secured from two additional customers.”

“Google-related market share competition risk exists, but the possibility of the absolute revenue base being impaired is limited.”

• Expected Impact

The most significant near-term driver of the stock decline from this earnings call was Hock Tan’s formal acknowledgment of the possibility of Google diversifying its TPU supply. Broadcom and Google have co-developed TPUs for 10 years, with 8 generations now released. What the market feared is the possibility of Marvell being adopted as a second supplier for Google TPUs.

However, two structural counterarguments hold when this issue is analyzed structurally.

First, overall demand growth is outpacing the pace of supply diversification. Management acknowledged that “some sourcing diversification is inevitable given Google’s rapidly growing AI computing consumption,” while simultaneously emphasizing that Broadcom “continues to provide technology and execution capability far superior to any alternative.” The logic is that even if a new supplier takes some volume, the impairment of Broadcom’s absolute revenue base is limited, because Google’s AI computing demand itself is expanding extremely rapidly.

Second, Google dependency is structurally declining over this cycle. Looking at the 6-customer roadmap that management detailed on the earnings call, the weight of Anthropic (FY27~, approximately 5GW), OpenAI (production from end of FY26, 1.3GW in FY27), Meta (from H2 FY27, 3GW through FY28), and two undisclosed customers (already holding $6 billion in purchase orders) is growing rapidly. In particular, the 20GW+ AI XPV compute platform being built jointly with Apollo and Blackstone (first tranche approximately $35 billion) is a new revenue stream that supplies TPU-based computing capacity to frontier AI companies including Anthropic and OpenAI — a structural shift that dilutes single-customer Google risk.

JPMorgan estimates that the volume remaining for just these two customers (Anthropic and OpenAI) in FY28 is approximately 15GW, representing approximately $180–225 billion in value based on $12–15 billion of content per GW. Combined with Google, Meta, and additional large programs, JPMorgan sees FY28 AI revenue potentially approaching approximately $300 billion. Combining Broadcom’s approximately 70% market share in switching and routing silicon with the expectation that the AI networking market will grow at a CAGR of over 50% for the coming years, the Google TPU supply diversification issue is more reasonably interpreted as noise occurring atop an already-diversified customer portfolio rather than a factor that disrupts Broadcom’s structural growth trajectory.

4. Meta, AI Agent Monetization Accelerates — B2B and B2C Dual-Track Now Live

• Core Source

“The redesigned Meta Business Agent, previously known as Meta Business AI, provides functions to respond to customer inquiries and recommend products, and supports tasks to be handled within platforms such as WhatsApp.”

“The Business Agent Platform is a solution that goes beyond service support within Meta’s own platforms and can be linked with third-party platforms and solutions. Representative examples include integration with Shopify, Zendesk, and Shopee.”

“These capabilities will lead to expanded use of Meta’s subscription services by enterprise customers. Meta has the potential to generate billions of dollars in revenue through this.”

“Unlike the recently announced Meta Business Agent, which is centered on its own platforms, Hatch is positioned as a general-purpose agent that operates across the web and external services.”

“Zuckerberg is advancing Hatch as the core product of the Personal Superintelligence strategy. It is oriented toward an AI agent that understands users’ goals and continuously carries out tasks.”

• Expected Impact

Meta has entered the enterprise AI agent business in earnest with the global launch of Meta Business Agent, a comprehensive overhaul of the existing Meta Business AI. The core of this launch is placing agent functionality capable of actual task execution — including appointment scheduling, order processing, payment support, and sales response — on top of its own platforms such as WhatsApp and Instagram, going well beyond simple chatbot capabilities. Furthermore, by supporting integration with third-party platforms including Shopify, Zendesk, and Shopee, Meta has expanded into B2B infrastructure that operates even outside the Meta ecosystem — a decisive difference from previous products.

The significance of this business from a monetization structure standpoint is substantial. Meta’s current revenue structure is overwhelmingly concentrated in advertising. The introduction of a subscription model through enterprise AI agents opens a path to creating recurring, predictable subscription revenue (SaaS-type) at scale for the first time outside advertising. Canaccord forecast that this business has “the potential to generate billions of dollars in revenue” and maintained a Buy rating with a target price of $930. Mirae Asset’s market commentary also noted that “with hyperscalers recently expanding spending through debt rather than cash on hand, profitability has become significantly more important — and Meta’s monetization strategy is being positively received by the market.” In fact, on the same day, Meta’s stock rose +4.24%, virtually the only Mag-7 stock to advance while big tech broadly declined.

Meanwhile, separately from Meta Business Agent, the consumer AI agent ‘Hatch’ is also reportedly being considered for launch around July. Hatch will provide general-purpose agent capabilities including vibe coding, schedule management, and email sending, with a premium tier (Hatch Plus) under consideration at up to $199.99 per month — the same price point as ChatGPT Pro and Claude Max. While Meta Business Agent targets enterprise customers within its own platforms, Hatch is positioned as a general-purpose agent operating across the web and external services, making visible a dual-track strategy simultaneously building both enterprise-facing (B2B) and consumer-facing (B2C) AI agents. Zuckerberg has stated he is advancing Hatch as the core product of the ‘Personal Superintelligence’ strategy.

Combining these two products, Meta is in a phase of building AI agent monetization models in both enterprise and consumer directions atop its existing platform infrastructure of billions of monthly active users (WhatsApp, Instagram, Messenger). This could become the first large-scale case of a platform company converting AI from a mere feature enhancement tool into an independent billing unit, making it a potential new benchmark in the AI monetization competition within the market.

5. Navitas Unveils GaN Power Solution in NVIDIA MGX Ecosystem

• Core Source

“Navitas unveiled a DC-DC power delivery board (PDB) at COMPUTEX 2026 that directly converts 800V to 6V. The key feature is the removal of the existing 48V intermediate bus converter (IBC) stage, improving system efficiency and reliability while enhancing space utilization within servers.”

“The board is equipped with 16 GaNFast power devices rated at 650V and 11mΩ, targeting maximum efficiency of 97.5%, switching frequency of 1MHz, and power density of 2,100W/in³. Its thickness is approximately 20% thinner than a smartphone, enabling very close placement to GPU boards and improving power delivery efficiency and instantaneous load response performance.”

“As AI data center power demand surges rapidly, power supply and conversion efficiency in megawatt-class AI server racks is emerging as the core bottleneck.”

“Navitas provides the GeneSiC silicon carbide (SiC) product family for supplying power from the grid to AI server racks, alongside GaNFast gallium nitride (GaN) technology responsible for high-efficiency power conversion from rack to GPU.”

• Expected Impact

The technical core of the 800V→6V DC-DC power delivery board unveiled by Navitas is the complete elimination of the 48V intermediate bus converter (IBC) stage from the conventional data center power conversion architecture. The previous data center server structure required passing through an intermediate stage (48V) when converting high-voltage external supply to the low voltage used by GPUs — by eliminating this stage, conversion losses are reduced, component count decreases, and space is freed. The peak efficiency of 97.5%, power density of 2,100W/in³, and thickness approximately 20% thinner than a smartphone are the results of this architectural innovation. The reduced thickness enabling very close placement to GPU boards further shortens power delivery paths, additionally improving instantaneous load response performance.

The reason this technology is attracting market attention is that the AI infrastructure power problem has now emerged not merely as a cost issue but as a physical bottleneck. As management noted on the Broadcom earnings call, one of the primary reasons customers are placing advance orders for AI semiconductors through 2028 is securing power infrastructure. Actually operating megawatt-class AI server racks requires alignment not just of chip supply but of the entire power supply and conversion efficiency chain. The CEO’s statement that “power supply has become the core challenge in building next-generation gigawatt-class AI factories” reflects this context.

Navitas’s participation in the NVIDIA MGX ecosystem means that Navitas’s power solution has been officially incorporated into the open modular AI infrastructure standard led by NVIDIA, opening a path where Navitas’s order opportunities structurally expand as MGX-based AI server rack deployment scales.

6. Apple Overhauls XR Roadmap, Scraps Vision Pro Line in Favor of Two Smart Glasses Products (Ming-Chi Kuo)

• Core Source

“The Apple XR headset and smart glasses roadmap I put together about a year ago is no longer a useful reference. For now, only two smart glasses products remain visible in the roadmap.”

“The major overhaul was signed off by Apple’s next CEO, John Ternus. Removing the Vision Pro line was the right call, as Apple shifts resources toward smart glasses with greater mass-market potential.”

“My latest supply chain checks suggest Apple’s display-equipped AR/XR smart glasses device, powered by optical waveguides, has slipped to 2029.”

“The display-less AI glasses, similar to Ray-Ban Meta, are still expected to ship in 2027.”

• Expected Impact

Ming-Chi Kuo published a post on X (@mingchikuo) on June 3rd stating that the Apple XR headset and smart glasses roadmap he released approximately one year ago is no longer valid, with only two products remaining in active development on Apple’s current roadmap. The development pipeline of head-mounted wearables has been dramatically reduced from the original seven products to just two.

The paths to launch for the two remaining products are as follows. First, display-less AI glasses with a form factor similar to Ray-Ban Meta, maintaining the 2027 shipment timeline. Second, display-equipped AR/XR smart glasses powered by optical waveguides, which have slipped to 2029. According to multiple outlets including 9to5Mac, the decision for this overhaul was made by John Ternus, who is scheduled to become Apple’s CEO on September 1, 2026, and who signed off on the large-scale reduction of the Vision product roadmap — this is the core of Kuo’s claim.

This roadmap overhaul carries significance on two market dimensions. First, from a competitive landscape perspective, Apple has effectively abandoned the direction of high-priced, enclosed XR headsets and shifted its strategic center of gravity toward the AI glasses category where Meta is already establishing market leadership with Ray-Ban Meta. This could be a factor that short-term strengthens Meta’s competitive advantage in the smart glasses market. Second, from a supply chain perspective, the slip of display-equipped glasses with optical waveguides to 2029 suggests that yield and cost challenges for this technology remain unresolved, meaning that the timing of benefits for optical waveguide-related component and materials companies may be delayed.

This announcement, coming ahead of WWDC 2026, is likely to serve as an occasion for the market to recalibrate expectations regarding Apple’s spatial computing strategy overall. Meanwhile, Mark Gurman has mentioned that a thinner and lighter successor headset remains on Apple’s development roadmap — partly conflicting with Kuo’s information. With multiple information channels painting different pictures, the final contours of the actual roadmap are expected to become clearer through the WWDC announcement.

7. Palantir × Google Cloud, Comprehensive Integration Partnership Established

• Core Source

“Palantir Technologies Inc. today announced a multi-tiered partnership with Google Cloud, enabling first-class integrations across Google Cloud platforms and making Palantir available on Google Cloud Marketplace.”

“This includes two-way data federation between BigQuery and Foundry, building on existing support of zero-copy virtual table integration, as well as the two-way semantic exchange between Google’s Knowledge Catalog and Foundry’s Ontology.”

“By uniting BigQuery and Gemini with Palantir’s Foundry and AIP, we’re giving joint customers a secure, unified foundation to run their most complex, high-stakes workflows at scale.” (Satish Thomas, VP Applied AI & Platform Ecosystem, Google Cloud)

“Our partnership with Google Cloud marries the years of investments that customers have made into Google’s Knowledge Catalog, BigQuery, and Cloud Storage with the operational force of Foundry and AIP, and enables them to unleash Gemini alongside their Ontology-powered AI strategy.” (Akshay Krishnaswamy, Palantir Chief Architect)

• Expected Impact

Palantir officially announced a multi-tiered strategic partnership with Google Cloud on June 4th. The core of this partnership consists of three technical integration pillars.

First is two-way data federation between BigQuery and Foundry. Building on the existing zero-copy virtual table integration support, data federation between the two platforms operates bidirectionally. This means Google Cloud customers who have already built data assets in BigQuery can leverage Palantir Foundry and AIP as-is, without moving or replicating data. The burden of data migration — one of the biggest barriers to Palantir adoption — is eliminated.

Second is the two-way semantic exchange between Google Knowledge Catalog and Foundry Ontology. Ontology is a core concept of the Palantir platform that connects enterprise data to real business operational entities (assets, organizations, processes, etc.) in a form that AI can understand and act upon. When Google’s Knowledge Catalog and Ontology are integrated, the data context that enterprises have defined in Google Cloud can flow directly into Palantir’s operational AI layer.

Third is deep Gemini–AIP connectivity. Palantir’s AI Platform (AIP) can now directly leverage Google’s Gemini model within enterprise AI workflows, and through the Ontology, Gemini can be embedded into the enterprise operational context.

The press release cites industrial company Eaton as a real-world use case. Through the combination of Foundry, AIP, Ontology, and Gemini, Eaton built production workflows that transform engineering documentation into intelligent operational assets, resulting in faster quote generation, improved engineering precision, reduced workload, and enhanced customer responsiveness.

This partnership also carries significant meaning from the standpoint of Palantir’s market penetration strategy. Making Foundry and AIP available through the Google Cloud Marketplace means Palantir can leverage Google’s existing cloud customer base directly as a distribution channel, accessing Google Cloud’s large enterprise customer network without an independent sales cycle. For Palantir, this can serve as a catalyst for accelerating commercial customer expansion and AIP adoption; for Google Cloud, it strengthens a differentiated enterprise solution portfolio combining data analytics and AI operations.