Broadcom Erased $300 Billion After Its Best AI Quarter

Broadcom’s Q2 FY2026 AI chip revenue hit $10.8 billion (up 143% from a year earlier) as the company beat earnings estimates across almost every metric. The stock still fell 12.59% on June 4, erasing over $300 billion in market cap after its Q3 AI chip guidance came in below the $17.2 billion analysts had expected.

Broadcom entered earnings at a record $495 per share and closed at $418.91. The stock has multiplied roughly ninefold since late 2022, when generative AI pushed hyperscaler spending on custom silicon into the tens of billions. That run set a bar high enough that beating consolidated revenue, earnings, and Q3 total revenue guidance wasn’t enough when the Q3 AI chip sub-line came in short.

A Beat on Almost Every Line

The quarter ending May 3 produced records across nearly every financial category. Total Q2 revenue hit $22.2 billion, up 48% year over year, per Broadcom’s official Q2 FY2026 earnings release. Non-GAAP earnings per share came in at $2.44 against a $2.40 consensus. GAAP diluted EPS hit $1.91 (up 85%), as GAAP net income climbed 88% to $9.31 billion.

Adjusted EBITDA (earnings before interest, taxes, depreciation, and amortization) reached $15.2 billion, equal to 69% of revenue. Free cash flow came in at $10.3 billion (46% of quarterly revenue), as profit margins widened even while the AI semiconductor operation scaled. AI bookings in the period exceeded $30 billion.

Hock Tan, President and CEO of Broadcom, named six core custom chip customers on the earnings call, including Google, Meta, Anthropic, and OpenAI. The board declared a quarterly dividend of $0.65 per share, payable June 30, per Broadcom’s quarterly earnings disclosure.

One segment fell short. Infrastructure software revenue, much of it from the 2023 VMware acquisition, landed at $7.18 billion, up 9% from a year earlier but $140 million below the $7.32 billion analysts polled by StreetAccount had estimated.

Q3 consolidated revenue guidance of $29.4 billion, implying 84% year-over-year growth, came in above the $28.53 billion consensus. The Q3 AI chip sub-line sat $1.2 billion short of the most aggressive analyst estimate.

The Guidance Gap That Triggered the Exit

Management held the full-year AI semiconductor guidance at $100 billion, unchanged from the target set earlier in the year, and the Q3 AI chip forecast landed below what analysts had modeled. A stock up nearly 40% heading into the print had priced in something larger on both counts.

Q2 semiconductor revenue from AI of $10.8 billion grew 143% year-over-year, above our forecast, driven by increasing demand for custom AI accelerators and AI networking.

Hock Tan, President and CEO of Broadcom, in the company’s Q2 FY2026 8-K filing with the SEC.

Two additional disclosures during the earnings call pushed the selloff deeper, per reporting by Barron’s. Tan acknowledged that Google, Broadcom’s most prominent design partner, was likely to source AI chips from multiple suppliers going forward. He also warned that rapid growth of AI semiconductor sales was beginning to compress overall gross margins. Kirsten Spears, CFO of Broadcom, guided Q3 adjusted EBITDA margin to 68%, stable despite the revenue mix shift but below the 69% posted in Q2.

A fourth factor shifted the revenue count itself. Broadcom had previously signaled plans to sell complete integrated AI systems, including racks bundled with networking hardware, for hyperscaler deployment. On the earnings call, the company pivoted to silicon only.

Stacy Rasgon, managing director at Bernstein, told Yahoo Finance that the original AI revenue targets had included rack-level shipments for Anthropic, “which they’re not doing anymore.” Broadcom described the pivot as margin-protective, keeping the company in its highest-margin silicon business and removing the lower-margin systems-integration work from the revenue plan.

Nvidia Closed Higher That Same Session

Nvidia (Nasdaq: NVDA) closed at $218.66 on June 4, up 1.94%, as Broadcom shed over 12%. In the quarter ending April 26, Nvidia posted revenue of $81.6 billion, up 85% year over year, with data center revenue of $75.2 billion (a 92% gain). The company trades at roughly 25 times forward earnings, per Nvidia investor relations data.

Broadcom builds custom AI accelerators (application-specific integrated circuits, or ASICs) engineered for each client’s specific model workloads. Nvidia designs general-purpose GPUs (graphics processing units) and sells them broadly across the market. Most of the largest cloud operators in Broadcom’s custom chip program simultaneously run Nvidia GPUs for research and training workloads, so the same AI capital budget produces revenue at both companies on different parts of the compute stack.

The June 4 semiconductor selloff extended across the sector. Micron Technology fell more than 7%, ARM Holdings shed 4%, and AMD slipped 3%.

South Korea’s KOSPI index dropped 1.8%, with Samsung Electronics and SK Hynix falling 2% to 4%. In Taiwan, Hon Hai Precision and Wistron declined roughly 4% and 8%, respectively. Broadcom’s guidance carries that weight because it supplies AI chips to the largest cloud operators globally, making each quarterly disclosure a proxy for the pace of hyperscaler AI deployment.

Why Hyperscalers Keep Both on Their Vendor Lists

The Inference Shift

Custom ASIC server shipments are projected to grow 44.6% in 2026 against 16.1% for merchant GPU shipments, according to TrendForce AI accelerator market tracking, the Taiwan-based semiconductor research firm. ASIC-based AI servers are expected to reach 27.8% of the total AI server market this year, the highest share on record. Hyperscaler capital expenditure on AI infrastructure is estimated at $650 billion to $690 billion for 2026.

Evercore analysts, in research circulated in May 2026, described the environment as an “inference-led regime” where buying criteria have shifted from maximum compute throughput to cost-per-token, power efficiency, and total cost of ownership. At billion-chip-scale deployments, small efficiency advantages compound into substantial cost differences. Purpose-built silicon, designed for known workloads at scale, holds a structural advantage on those dimensions.

The Multi-Year Design Commitment

Broadcom is widely estimated to hold roughly 70% of the custom AI accelerator design market, anchored by its Google TPU silicon program and Meta’s custom chip development. The Meta partnership was extended through 2029. Anthropic, Google, and Broadcom signed a compute agreement covering approximately 3.5 gigawatts of AI capacity expected to come online from 2027, Yahoo Finance reported on the earnings call disclosure.

Each program is a multi-year co-design engagement. A custom accelerator is built around a specific hyperscaler’s model architectures and memory hierarchies, which means switching costs are embedded in the silicon’s design process. Moving off Broadcom silicon restarts a development cycle that typically takes two to three years from specification to production.

Several hyperscalers run both custom ASICs and Nvidia GPUs in parallel. Microsoft deploys its Maia 100 (a custom AI accelerator built for Azure) alongside Nvidia GPU platforms for training and experimental workloads. Both chip vendors collect from the same capital budget on different parts of the workload stack, which is how the dual-vendor approach became the default configuration in large-scale AI infrastructure by mid-2026.

Wall Street Raised Targets on the Dip

After Thursday’s close, multiple analyst teams raised their Broadcom price targets, The Motley Fool reported. About 93% of analysts covering the stock rate it a buy or strong buy.

• Jefferies raised its target from $500 to $550
• KeyBanc Capital Markets lifted its target to $575
• JPMorgan moved its target to $580

Consensus analyst estimates put Broadcom’s 2026 revenue at $106 billion, scaling to $166 billion in 2027. The 2028 consensus sits at $218 billion, while JPMorgan projects roughly $300 billion for the same year, The Motley Fool reported, reflecting a more aggressive assumption about the pace of hyperscaler AI buildout.

After the drop, Broadcom trades at a price-to-earnings-growth (PEG) ratio near 0.55, based on forward earnings growth estimates. A PEG below 1.0 typically indicates a stock whose earnings are growing faster than its current price reflects.

Daniel Newman, CEO of the technology research firm Futurum Group, said the Q2 report “feels more like an expectations reset than thesis damage.” Newman cited Broadcom’s plan to deliver 10 gigawatts of AI compute capacity as still on schedule, with additional capacity planned for 2028. Broadcom’s CEO described AI demand on the call as “simply insatiable”; Futurum’s read was that the guidance figure reflects timing in chip shipment schedules, with the underlying order pipeline intact.

Broadcom enters the current quarter with its $100 billion full-year AI chip target the only guidance the company left unchanged Wednesday.

Disclaimer: This article is for informational purposes only and does not constitute financial or investment advice. References to specific securities, analyst price targets, and forward guidance involve inherent risks and uncertainties. Past performance is not indicative of future results. Readers considering investment decisions should consult a qualified financial adviser. Figures reflect data available as of publication date.