src=’https://storage.ghost.io/c/5f/b9/5fb98cf7-67da-41b5-a184-b1b39752525f/content/images/size/w1384/2026/05/hashrate-index-banner-ai-asic-report-p2.jpg’ alt=’Hashrate Index AI ASIC Report Banner’>
Two major semiconductor design firms, Broadcom and Marvell, are the undisclosed powerhouses behind the majority of custom Artificial Intelligence (AI) Application-Specific Integrated Circuits (ASICs) developed for hyperscale cloud providers. Despite the public narrative of hyperscalers developing “in-house silicon,” a significant portion of this development relies on design services and intellectual property from these two entities. Broadcom leads this market, estimated to hold around 70% of the custom AI accelerator design services, while Marvell holds a strong second position, enabling over 80% of hyperscaler AI ASIC programs collectively.
This arrangement is fundamentally different from traditional hardware manufacturing. These design partners provide the architectural blueprints and specialized IP, with hyperscalers absorbing the substantial capital expenditure (capex) and manufacturing risks. This capital-light, design-centric model allows Broadcom and Marvell to achieve high gross margins, with Broadcom’s exceeding even NVIDIA’s. Their revenue visibility is further bolstered by multi-year committed customer backlogs, insulating them from the typical volatility of the semiconductor industry.
Key Takeaways
- Broadcom and Marvell are the primary design partners for custom AI ASICs used by major hyperscalers.
- Broadcom commands an estimated 70% market share in AI ASIC design services.
- The business model is capital-light, with customers bearing manufacturing and capex burdens.
- Broadcom’s gross margins are notably high, surpassing NVIDIA’s.
- Customer concentration poses the primary risk for both design partners.
The Two Independent AI Chip Models
The landscape of independent AI chip companies can be broadly categorized into two distinct business models, each requiring a unique analytical framework:
- Design Enablers: Companies like Broadcom and Marvell specialize in providing custom ASIC design services, intellectual property (IP), and networking silicon. They do not sell AI chips under their own brand for direct deployment in data centers. Their revenue is generated from the design process and licensing of their IP. Their competitive arena is primarily against other design enablers, not against chip manufacturers like NVIDIA or the hyperscalers they serve. This model is characterized by its capital-light nature and heavy reliance on large-scale cloud customers.
- Direct Competitors: Firms such as Groq, Cerebras, Etched, Tenstorrent, and Tensordyne develop and market their own AI chips. They directly challenge NVIDIA’s GPU dominance and, in some instances, compete with proprietary hyperscaler silicon. These companies operate on a more traditional semiconductor startup model, requiring significant capital investment, facing long development cycles, and managing the inherent risks of silicon validation and customer acquisition. Their success is often tied to demonstrable technical advantages and securing anchor clients.
Understanding these distinctions is crucial for accurate valuation and risk assessment. Applying the same evaluation criteria to a design enabler like Broadcom and a direct competitor like Tensordyne would lead to flawed conclusions due to their fundamentally different operational scales, competitive dynamics, risk profiles, and growth trajectories.
Broadcom: The Dominant Design Partner
Broadcom stands as the undisputed leader in the AI ASIC design enablement market. Its market share in custom AI accelerator design services is estimated at approximately 70%, an increase from previous figures. This dominance is amplified by its success in securing design wins with major AI players, including Google, Meta, OpenAI, Anthropic, and, more recently, Apple. This growing customer base, especially in the custom silicon space, signifies a significant shift in the AI hardware supply chain.
The financial performance of Broadcom’s AI semiconductor segment reflects this market leadership, exhibiting a doubling in year-over-year revenue for four consecutive quarters. Projections indicate even stronger sequential growth heading into the second quarter of fiscal year 2026, underscoring the accelerating demand for its design services and specialized IP.
Details can be found on the website : hashrateindex.com