Goldman Sachs: Custom AI Chips to Rival GPUs by 2027—Here's Why $AVGO and $MRVL Matter

Goldman Sachs has issued a pivotal forecast for the artificial intelligence chip market: custom-built semiconductors, known as ASICs, will match the demand for GPU chips by 2027. This projection signals a fundamental shift in how the world's largest technology companies—including Google, Amazon, Microsoft, and Apple—will source their AI computing power. According to the investment bank's analysis, the market for these specialized chips will experience explosive growth, with Bloomberg Intelligence forecasting a 27% annual growth rate through 2033. Two semiconductor firms stand out as clear beneficiaries of this trend: Broadcom ($AVGO) and Marvell Technology ($MRVL), both of which are actively designing custom ASICs for major hyperscalers and positioned to capture significant market share in this emerging segment.

The Rise of Custom AI Chips: Market Dynamics and Growth Projections

The shift toward ASICs represents a strategic pivot by cloud giants seeking to reduce their dependence on NVIDIA and other traditional GPU manufacturers while simultaneously optimizing computing power and energy efficiency for their specific workloads. ASICs—application-specific integrated circuits—are purpose-built chips tailored to perform particular tasks with exceptional efficiency, making them ideal for the standardized, repetitive nature of AI inference and training operations at scale.

Goldman Sachs' timeline of 2027 for ASIC parity with GPUs carries significant weight given the firm's influence in technology sector analysis. The firm's projection is bolstered by Bloomberg Intelligence's more granular forecast:

• 27% annual compound growth rate (CAGR) projected through 2033 for custom AI chips
• Expected market maturation and adoption acceleration across hyperscaler networks
• Increasing cost pressures driving major technology companies to internalize chip design
• Superior power efficiency of ASICs compared to general-purpose GPUs for standardized AI workloads

This growth trajectory underscores a fundamental economics shift: as AI computing becomes commoditized and hyperscalers deploy identical models across millions of servers, the business case for custom silicon strengthens considerably. The one-time engineering investment in ASIC design becomes justified by years of operational savings at scale.

Broadcom and Marvell: Positioned at the Center of AI Chip Transition

Broadcom emerges as the dominant player in this evolving ecosystem, with Goldman Sachs forecasting the company will maintain a commanding 60% market share in the custom AI chip design space. This dominant position reflects Broadcom's established relationships with hyperscalers, deep expertise in high-speed interconnect technologies, and proven track record designing complex semiconductors for data center applications.

Both Broadcom ($AVGO) and Marvell Technology ($MRVL) are actively engaged with the world's largest cloud computing platforms:

• Google has been developing custom ASICs for AI workloads (TPUs) for several years
• Amazon is designing chips for AWS infrastructure optimization
• Microsoft is developing custom processors for Azure AI services
• Apple continues expanding its proprietary chip design capabilities

Marvell Technology occupies a strong secondary position, benefiting from its existing relationships with major cloud providers and its specialized expertise in storage, connectivity, and data center infrastructure semiconductors. The company's portfolio of high-speed interface and data center solutions positions it well to support the complex interconnect requirements of custom AI chips.

The competitive dynamics differ markedly from the GPU market, where NVIDIA ($NVDA) holds approximately 80-90% of the discrete GPU market share. In the emerging custom ASIC segment, the market structure is fundamentally more fragmented, with each hyperscaler potentially developing proprietary designs tailored to their specific infrastructure and algorithmic requirements. This fragmentation creates opportunities for chip design partners like Broadcom and Marvell, who serve as critical technology partners and intellectual property providers to these internal design efforts.

Market Context: The GPU Dominance Question and Infrastructure Economics

The Goldman Sachs forecast arrives amid ongoing debate about NVIDIA's sustainability in the AI chip market. While NVIDIA has captured an unprecedented share of AI infrastructure spending in recent years, the company faces structural headwinds:

• Hyperscaler concentration: The top five cloud providers (Google, Amazon, Microsoft, Apple, and Meta) account for the vast majority of AI infrastructure investment
• Internal design capability: Each major hyperscaler has invested billions building internal semiconductor design teams
• Economics at scale: The fixed cost of ASIC design amortizes dramatically when deploying millions of units across global data centers
• Customization advantage: ASICs designed for specific workloads can deliver superior price-performance compared to general-purpose GPUs

The semiconductor industry has witnessed similar transitions historically. Just as Intel saw its dominance challenged by custom chips in specific applications, NVIDIA's GPU dominance faces pressure from hyperscaler-designed ASICs. The critical distinction: NVIDIA retains optionality, as the company can pivot toward providing the software frameworks, libraries, and development tools that support AI workloads—a business with lower capital intensity but potentially higher margins.

Broadly speaking, the AI chip market is expanding sufficiently that growth of custom ASICs need not cannibalize GPU demand entirely. Bloomberg Intelligence's projection of 27% annual growth through 2033 suggests a market expanding to accommodate multiple architectures and suppliers. Nevertheless, the mathematical reality remains: as ASIC adoption rises from negligible levels today to matching GPU deployment by 2027, the GPU market's growth rate will inevitably decelerate.

Investor Implications: Why This Matters for Technology and Semiconductor Portfolios

For investors, the Goldman Sachs analysis carries several critical implications:

Semiconductor Design Leadership: Broadcom and Marvell represent leveraged plays on the custom AI chip transition without bearing the full design and manufacturing risk that hyperscalers themselves assume. Both companies benefit from recurring engineering and design partnerships that generate substantial recurring revenue from their hyperscaler customers.

Margin Expansion Opportunity: Design partnerships for custom chips typically command higher margins than selling standardized components. As Broadcom and Marvell shift their hyperscaler relationships from component suppliers to strategic technology partners, gross margins on these segments should expand.

Secular Growth Visibility: The 27% annual growth rate through 2033 provides exceptional clarity compared to mature semiconductor subsegments. For investors seeking exposure to long-duration secular growth trends, the custom AI chip segment offers compelling fundamentals.

Diversification from NVIDIA Concentration Risk: Portfolio managers overweight in NVIDIA ($NVDA) may view Broadcom and Marvell as beneficial diversification within technology infrastructure, providing similar secular AI exposure with different competitive dynamics and lower execution risk.

TSMC Implications: As the world's leading foundry, Taiwan Semiconductor Manufacturing Company ($TSM) represents the enabling infrastructure for ASIC manufacturing. The proliferation of custom chips benefits TSMC through increased wafer demand, even if specific designs fragment across multiple hyperscalers.

Looking Ahead: The 2027 Inflection Point

The convergence of ASIC adoption matching GPU demand by 2027 represents a critical inflection point for semiconductor industry structure. Goldman Sachs' forecast arrives with sufficient lead time for investors to reposition portfolios and for semiconductor companies to prepare for shifting competitive dynamics.

Broadcom's projected 60% market share in custom AI chip design validates the company's strategy of deep embedded partnerships with hyperscalers. The company's integrated portfolio—spanning high-speed interconnect, networking, and semiconductor design—positions it uniquely to capture value throughout the custom chip supply chain.

Marvell Technology enters this inflection point with strong fundamentals in data center infrastructure and growing hyperscaler partnerships. The company's expertise in the networking and storage challenges that custom AI chips must address creates a complementary business model to Broadcom.

For long-term investors in technology infrastructure, the emergence of the custom AI chip market represents a multi-year growth opportunity with visible catalysts, strong secular trends, and clear industry leaders. The window for establishing positions in beneficiary companies narrows as the 2027 inflection point approaches and market dynamics crystallize around Broadcom, Marvell, and the hyperscaler platforms they serve.