Broadcom (AVGO) on Solana

Broadcom CEO Hock Tan flatly dismissed concerns that the growing debate over moderating AI development would affect his company's business. When asked whether the discussion had altered his semiconductor demand outlook, Tan replied, "No, not in the least," maintaining that compute infrastructure spending would remain "very strong" and "very durable" as the industry shifts from model training toward inference and commercial deployment.

Broadcom joined Nvidia in pushing back against calls to slow AI development, with both companies arguing that competitive pressures and the accelerating rollout of commercial AI applications would sustain robust chip demand regardless of regulatory or philosophical debates about development pace. Despite Tan's confident tone, Broadcom shares fell 4.8% on the day the broader AI slowdown discussion gained traction, reflecting investor uncertainty about whether executive optimism can hold against a shifting policy narrative.

Broadcom's latest quarterly results showed $16.7 billion in AI semiconductor sales — 73% of which came from custom XPU shipments — with management guiding current-quarter AI revenue to $21.7 billion, a 236% year-over-year gain. The company raised its full-year FY2026 AI revenue guidance to $58 billion and projected consecutive doubling through FY2027 ($115 billion) and FY2028 ($230 billion), driven by custom accelerator demand from hyperscalers including Google Cloud, Meta, OpenAI, and Anthropic.

Analysts reviewing the Broadcom-versus-Nvidia question post-earnings note a valuation anomaly: Broadcom trades at roughly 46x trailing and 31x forward earnings — a premium to Nvidia's 28x trailing and 23x forward multiples — despite Broadcom's higher projected AI growth rates. The comparison highlights Broadcom's differentiated position as a custom-silicon supplier rather than a merchant GPU vendor, though the valuation gap leaves less obvious room for multiple expansion relative to its rival.

Broadcom's growing custom AI chip business — designing bespoke accelerators for hyperscalers such as Meta, whose Iris chip was co-developed with Broadcom — relies on TSMC as its manufacturing backbone. The arrangement highlights a structural dynamic in the AI chip market: as Broadcom takes share from NVIDIA's merchant-GPU model by offering tailored silicon, the foundry that makes those chips benefits regardless of which architectural approach wins. TSMC reported record August revenue of NT$514.81 billion, up 53.3% year-over-year, reflecting surging leading-edge demand that encompasses both custom and merchant AI chip production.

Analysts characterize TSMC as "one of the more durable beneficiaries of the custom-chip trend" precisely because it serves Broadcom, NVIDIA, AMD, and other designers simultaneously. Hedge fund ownership of TSMC shares rose to 249 funds in Q2 from 234, while Broadcom saw a modest dip to 170 funds from 173. Concentration risk remains a watch item for Broadcom — its custom AI revenue is tied to a small number of hyperscaler relationships — while TSMC faces its own headwinds in capital intensity, geopolitical exposure, and the long-term possibility of customers internalizing foundry capabilities.

Amazon CEO Andy Jassy confirmed the company will raise its capital expenditure to $220 billion in 2026, up from a prior $200 billion plan, and noted that even this expanded budget will fall short of satisfying AI infrastructure demand. Jassy described demand extending into 2028 as "striking," framing the accelerated spending as a response to genuine customer need rather than speculative buildout. For Broadcom, a supplier of networking switches, routers, and custom ASIC design services to hyperscalers, commitments of this scale from a top-tier cloud customer directly support order visibility for data center silicon and interconnect products.

Broadcom's custom chip business — which helps large customers including cloud providers design application-specific integrated circuits tailored to AI workloads — stands to benefit as Amazon continues expanding its in-house silicon program alongside third-party infrastructure procurement. The sustained multi-year spending signal from Jassy reduces uncertainty about demand duration, which has been a key concern for investors assessing the longevity of the AI infrastructure cycle that has driven Broadcom's recent revenue growth.

Broadcom reported fiscal Q3 2026 revenue of $29.59 billion, up 86% year-over-year, with adjusted EPS of $3.32 beating the $3.24 Wall Street consensus. AI-related semiconductor revenue was the standout, more than tripling to $16.7 billion — a 221% annual increase — driven by custom accelerator shipments for OpenAI (the Jalapeno chip), Meta (MTIA inference accelerator), Anthropic, and ongoing processor supply to Google. Net income tripled to $13.09 billion. CEO Hock Tan highlighted that Broadcom expects to ship tens of billions of dollars worth of processors to Google annually for several years, with Anthropic planning a 5-gigawatt TPU8i deployment next year and OpenAI targeting a 1.3-gigawatt Jalapeno rollout in 2027.

Despite the beat, shares fell nearly 2% after hours because Q4 guidance of $34.8 billion came in slightly below the $35.03 billion analyst forecast. Tan projected AI revenue would double to roughly $115 billion in fiscal 2027 and double again to around $230 billion in fiscal 2028, underpinned by commitments from the hyperscaler labs. CFO Amie Thuener also disclosed that Broadcom is evaluating contingent residual-value guarantees to help two leading AI labs bridge the gap between current cash flows and the large upfront capital their infrastructure programs require — an unusual financing posture that underscores how central Broadcom's custom silicon has become to next-generation AI buildouts.

Broadcom is set to report fiscal Q3 2026 earnings on September 2 after market close, with analysts forecasting revenue of $29.4 billion — an 84% year-over-year increase — and EPS of $3.24, representing 92% growth. AI chip revenue is expected to reach approximately $56 billion for the full year 2026, driven in part by a custom AI chip developed with OpenAI, codenamed "Jalapeño," with two additional generations already in development. Networking products currently account for 40% of Broadcom's AI chip revenue, though that share is expected to decline as custom processor demand accelerates.

Investors are watching whether the results confirm Broadcom's positioning in the expanding optical networking market, which Goldman Sachs projects could grow from $15 billion in 2026 to $154 billion by 2028, with demand for co-packaged optics already outpacing supply. Despite that growth trajectory, Broadcom trades at roughly 19x forward earnings — a meaningful discount to the Nasdaq-100's 24x multiple — a gap analysts note could compress if the company delivers on its AI revenue forecasts and reaffirms its earnings growth outlook of approximately 70% for the current and next fiscal years.

BMO Capital Markets initiated coverage of Broadcom (AVGO) on August 21, 2026 with an Outperform rating and a $455 price target — representing over 25% upside from the stock's price near $368 at the time — citing the company's position as the "leading AI supplier in custom ASIC (XPU) and networking." The initiation arrived as shares had dropped roughly 7% over five trading days and sat about 25% below their 2026 high of $495, though BMO's target remains more conservative than the Street's average of $506.29 on a Strong Buy consensus.

Broadcom reported fiscal Q2 revenue of $22.19 billion, up approximately 48% year-over-year, and management has guided toward more than $100 billion in AI chip revenue for fiscal 2027 — a figure CEO Hock Tan has repeatedly underscored on earnings calls. With shares trading near 35 times forward earnings and a next earnings report scheduled for September 2, 2026, BMO's initiation frames the recent pullback as an entry point ahead of what it expects to be continued AI-driven demand for Broadcom's custom silicon and networking infrastructure.

Broadcom is negotiating a fresh debt package potentially reaching $100 billion, structured as a roughly $60–70 billion senior secured tranche and a $30 billion junior tranche channeled through a special purpose vehicle. The financing is designed to fund AI chip arrangements for customers including Anthropic, extending a June partnership with Apollo Global Management and Blackstone that committed $35 billion to expand Anthropic's computing capacity as part of a push to enable more than 20 gigawatts of AI compute by 2028. Broadcom also signed a five-year deal with Alphabet covering future TPU chip generations and networking equipment, reinforcing its role as the preferred ASIC supplier for hyperscalers seeking custom alternatives to general-purpose GPUs.

The scale of the debt raise tracks Broadcom's accelerating AI revenue base: AI semiconductor sales reached $10.8 billion in the second quarter alone, putting the business on roughly a $43 billion annualized run rate. Management has projected more than $100 billion in AI chip revenue by 2027, a figure that would exceed the company's total 2025 revenue of $64 billion and suggests Broadcom is positioning itself not just as a chip supplier but as a capital intermediary for large-scale AI infrastructure deployment.

Stanley Druckenmiller's Duquesne Family Office and Dan Loeb's Third Point fully exited their Broadcom positions in the most recent 13F filings, selling 195,955 and 50,000 shares respectively, while simultaneously adding to Taiwan Semiconductor. Druckenmiller increased his TSM stake to 589,680 shares, and Third Point lifted its holding to 460,000 shares. David Tepper's Appaloosa ran against the grain by opening a new 150,000-share AVGO position, though Appaloosa's much larger TSM bet — 1.65 million shares worth roughly $788 million — signals the same underlying rotation.

The shift comes despite Broadcom reporting strong Q2 FY2026 results: revenue of $22.2 billion (+48% year-over-year) and AI semiconductor revenue of $10.8 billion (+143%), with Q3 guidance calling for $29.4 billion in revenue including $16 billion in AI silicon. Broadcom's stock is down roughly 8% over the past week and up about 14% year-to-date, while Taiwan Semiconductor has gained 41% in 2026 and 79% over the past year. Analysts note that concerns over customer concentration risk — including reports of Alphabet diversifying its custom-chip suppliers — and 230 basis points of margin compression are weighing on sentiment toward AVGO even as its AI revenue trajectory remains among the strongest in semiconductors.

Broadcom has established an off-balance-sheet chip financing vehicle that could accumulate up to $370 billion in senior debt by mid-2029. The structure allows customers to lease Broadcom's custom AI accelerator chips (XPUs) by backstopping their lease obligations, making it easier for hyperscalers and other large buyers to access Broadcom's AI platforms without committing full upfront capital.

The central risk is what happens if customers default. Because Broadcom effectively guarantees those lease obligations, widespread non-payment could push exposure back onto the company's own balance sheet — meaning the off-balance-sheet label may not hold in a downturn in AI spending. Analysts are flagging future quarterly disclosures on guaranteed lease exposure as the critical metric to watch as the vehicle scales toward 2029, with Broadcom's roughly $2 trillion market capitalization already pricing in a sustained AI infrastructure buildout.