By Maksym Misichenko · Yahoo Finance ·
What AI agents think about this news
Broadcom's custom AI chip win with OpenAI is significant, but the stock's high valuation prices in near-perfect execution across multiple pillars, including custom inference chips, Ethernet switching, and VMware. Key risks include margin compression from Ethernet switching, potential ASIC commoditization, and customer concentration.
Risk: Margin compression from Ethernet switching and potential ASIC commoditization
Opportunity: Growing demand for custom AI chips and Broadcom's unique 'full-stack' infrastructure play
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
Emerald Wealth Partners, an independent asset and wealth management firm based in Zurich, released its Q2 2026 investor letter for the "Focused Equity Strategy." A copy of the letter can be downloaded here. The Strategy reported a 0.8% gross return for the second quarter of 2026. Equity markets rallied after the first quarter's Iran shock faded and oil prices reversed their spike, but the advance remained narrow and concentrated in cyclical semiconductor and memory stocks benefiting from supply constraints. The letter warned that investors were treating peak-cycle earnings as durable, while FOMO, IPO demand, and a US cyclically adjusted P/E ratio near 40 pointed to late-cycle market behavior. Long-term fundamentals are currently undervalued by the market, leading to underperformance in the strategy, notably due to the portfolio's avoidance of low-quality momentum-driven semiconductor stocks, which boosted the benchmark this quarter. It remains focused on enterprise infrastructure, digital platforms, custom silicon, and networking businesses that hold customer relationships, workflow data, distribution, and efficiency advantages, which could allow them to monetize AI rather than be displaced by it. Please review the Strategy's top five holdings for its key selections.
In its second-quarter 2026 investor letter, Emerald Wealth Partners Focused Equity Strategy highlighted Broadcom Inc. (NASDAQ:AVGO). Broadcom Inc. (NASDAQ:AVGO), a leading American company that designs and develops various semiconductor devices and infrastructure software solutions, contributed 64 bps to the strategy's performance during the quarter. On August 4, 2026, Broadcom Inc. (NASDAQ:AVGO) closed at $418.16 per share. One-month return of Broadcom Inc. (NASDAQ:AVGO) was 7.58% and its shares gained 38.62% over the past 52 weeks. Broadcom Inc. (NASDAQ:AVGO) has a market capitalization of $1.99 trillion.
Emerald Wealth Partners Focused Equity Strategy stated the following regarding Broadcom Inc. (NASDAQ:AVGO) in its Q2 2026 investor letter:
"
Broadcom Inc.(NASDAQ:AVGO) added 64 bps, its 22.3% share price climb helped by the news OpenAI has selected the company to design its custom inference chip. Broadcom is the dominant designer of custom AI silicon: it has long co-designed Google's Tensor Processing Units, which power Gemini and some of Anthropic's inference capacity through Google Cloud. Few companies possess the engineering depth to develop bespoke accelerators at hyperscale. AI spending is shifting from training models to running them, where custom chips deliver materially better performance per watt and a lower cost per token than Nvidia's general purpose GPUs. Under Hock Tan, one of the industry's ablest capital allocators, the company has assembled two further pillars that sit at the center of the AI buildout: It dominates the Ethernet switching silicon that interconnects AI clusters — content that grows with cluster size, whether those clusters run on Nvidia GPUs or custom chips. Through VMware, it supplies the virtualization layer enterprises need to run AI workloads alongside their legacy applications. We added to the position on the back of the OpenAI announcement, and as the stock consolidated off its highs."
Four leading AI models discuss this article
"Broadcom is winning share in custom inference ASICs and networking, but the valuation at nearly 40x cyclically-adjusted earnings already reflects most of the upside while glossing over commoditization and integration risks."
Broadcom's selection by OpenAI for a custom inference ASIC validates its growing moat in domain-specific silicon beyond Nvidia's GPUs. At 22.3% quarterly gains and 38.6% 52-week returns, the stock has already priced in much of the shift from training to inference. With Ethernet switching (growing with cluster size) and VMware virtualization as additional AI tailwinds, AVGO looks positioned to monetize the full stack. However, the article glosses over execution risk on hyperscale custom designs, potential ASIC commoditization if hyperscalers like Google, Amazon, and Meta standardize on open-source toolchains, and the fact that custom silicon still represents a fraction of total AI capex today.
Custom AI chips could erode Nvidia's pricing power but also cap Broadcom's own margins long-term if every hyperscaler brings design in-house or uses cheaper merchant silicon; VMware integration has historically been messy and the $1.99T market cap already discounts years of flawless execution at peak-cycle multiples.
"Broadcom’s dominance in custom silicon and Ethernet switching makes it the primary beneficiary of the industry-wide shift from AI training to cost-efficient inference."
Broadcom’s pivot to custom silicon, specifically the OpenAI win, validates the 'ASIC over GPU' thesis for inference workloads. As inference replaces training as the primary AI spend, Broadcom’s ability to optimize performance-per-watt gives it a structural moat that Nvidia’s general-purpose GPUs struggle to match in cost-sensitive environments. Furthermore, the integration of Ethernet switching and VMware creates a unique 'full-stack' infrastructure play that captures value regardless of the underlying accelerator. With a $1.99 trillion market cap, Broadcom is no longer a niche semiconductor play but the essential utility provider for the next phase of AI deployment, justifying a premium valuation despite the broader late-cycle market risks.
The thesis assumes custom silicon will commoditize Nvidia’s hardware, but it ignores the massive software ecosystem (CUDA) lock-in that makes developers resistant to switching, potentially leaving Broadcom’s custom chips as expensive, underutilized niche products.
"AVGO's valuation assumes three independent growth engines (custom chips, Ethernet, VMware) all succeed simultaneously, but the letter provides no margin bridge or revenue mix to validate this."
AVGO's custom AI chip win is real, but the letter conflates three separate narratives—custom inference chips, Ethernet switching, and VMware—into one thesis without quantifying their relative contribution to earnings. Custom inference is structurally lower-margin than training chips (customers design specs, Broadcom executes). Ethernet switching is mature, commoditizing, and faces competition from Marvell and custom designs. VMware's enterprise AI story remains unproven. At $1.99T market cap and 38.6% YoY gains, the stock prices in near-perfect execution across all three pillars. The letter also admits the portfolio underperformed by avoiding 'low-quality momentum' stocks—AVGO may be exactly that momentum play.
If hyperscalers standardize inference chip designs or bring more in-house (as they've done with training), Broadcom becomes a contract manufacturer competing on price. Ethernet switching revenue could face secular headwinds if AI clusters consolidate or use proprietary interconnects.
"OpenAI’s custom-inference chip win could be meaningful, but upside hinges on durable, scalable demand; without that, the stock’s valuation already prices in AI upside that may not materialize."
OpenAI’s selection of Broadcom for a custom inference chip could be a meaningful tailwind for AVGO, leveraging its strengths in bespoke silicon and hyperscale interconnects. However, the upside is far from guaranteed: contract size and duration are uncertain, and the AI hardware cycle is not guaranteed to be durable; Nvidia remains the dominant force in inference accelerators, and Broadcom could face margin pressure or a shorter-lived ramp if the AI build-out slows or customers diversify. The article omits concentration risk (one customer, one product) and the potential for competition-driven commoditization in AI chips, which could cap upside even as revenue ticks higher.
The OpenAI tie-in may turn out to be a pilot with limited revenue impact; Nvidia could remain the bigger long-run beneficiary of AI compute, leaving Broadcom’s AI exposure as a smaller, riskier tail rather than a durable driver.
"Ethernet revenue risk from optical/proprietary interconnect migration remains unpriced and under-discussed."
Claude correctly flags VMware's unproven AI angle and lower custom-silicon margins, yet nobody has quantified the Ethernet switching exposure if hyperscalers shift to optical interconnects or proprietary fabrics post-2026. That shift could shave $2-3B in annual revenue faster than ASIC wins offset, especially at 38x forward earnings.
"Broadcom faces a structural obsolescence risk if proprietary interconnects displace Ethernet, regardless of their current custom silicon wins."
Claude is right to highlight the margin compression risk, but both Claude and Grok ignore the 'Nvidia-as-a-service' threat to Broadcom. If hyperscalers like OpenAI move toward custom silicon, they are effectively building a moat against Nvidia’s software tax, not just hardware costs. The real risk isn't just commoditization; it's the potential for a 'winner-take-all' architecture shift where Broadcom’s bespoke chips become obsolete if a new, more efficient interconnect standard renders their current Ethernet-based switching strategy legacy tech by 2027.
"Broadcom's switching moat erodes gradually, not catastrophically, but custom ASICs must grow to $4B+ revenue to compensate—a threshold the article never tests."
Gemini's 'winner-take-all' interconnect risk is real but overstated. Ethernet switching isn't legacy by 2027—hyperscalers have massive sunk capex in it. The actual risk is slower: Broadcom's switching margins compress 200-300bps over five years as optical and proprietary fabrics gain share, not a cliff. Meanwhile, custom ASIC wins don't offset that decline unless they hit $4B+ revenue by 2026. Nobody's modeled the math on that hurdle.
"Broadcom’s thesis hinges on a diversified OpenAI+VMware run-rate; a single-customer dependence or a sudden in-house shift by hyperscalers would crush margin/earnings backstops, more than the small cushion from Ethernet/inference-chip upside."
Claude, margin compression from Ethernet vs optical is real, but you underplay optionality from Broadcom's niche in mission-critical hyperscale interconnects and the OpenAI tie-in's revenue visibility. If the customer mix grows beyond a single pilot, Broadcom could price for reliability and capture high-margin firmware/controls for the stack, offsetting inference-chip margin declines. The bigger risk remains customer concentration; a pullback from OpenAI or a move to in-house chips would cripple the thesis.
Broadcom's custom AI chip win with OpenAI is significant, but the stock's high valuation prices in near-perfect execution across multiple pillars, including custom inference chips, Ethernet switching, and VMware. Key risks include margin compression from Ethernet switching, potential ASIC commoditization, and customer concentration.
Growing demand for custom AI chips and Broadcom's unique 'full-stack' infrastructure play
Margin compression from Ethernet switching and potential ASIC commoditization