The biggest names in technology have decided that buying AI chips off the shelf is no longer enough, and the shift toward custom AI chips is now showing up in the financial results of the companies that design and manufacture them. Broadcom, Marvell Technology and Taiwan Semiconductor Manufacturing are emerging as the clearest beneficiaries, according to reporting by Crypto Briefing and The Motley Fool.
Alphabet, Microsoft, Amazon and other large tech companies are spending hundreds of billions of dollars on this expanding market, The Motley Fool reported, as they deploy more advanced models and AI agents. Custom processors can be fine-tuned to run specific AI models faster and more efficiently than general-purpose GPUs for some tasks, which is why the hyperscalers are investing in silicon built around their own workloads.
Why hyperscalers want custom AI chips
For years, AI data centers relied almost exclusively on GPUs for processing and training, according to The Motley Fool. That is changing as companies that run the same kind of AI task billions of times look for chips tuned precisely for that task. The processors in question are application-specific integrated circuits, or ASICs, which are built for one job rather than general-purpose computing.
Many of these companies rent out cloud computing and AI services to customers, so the performance of their silicon directly affects the products they sell. Alphabet's Google Cloud business spans cloud computing and Gemini Enterprise subscriptions, and that segment's second-quarter sales jumped 82% to $24.8 billion, the fastest growth of any Alphabet revenue line, The Motley Fool reported. To power those services, Alphabet relies on its own custom-designed Tensor Processing Units.
The market data points the same direction. Custom AI ASIC shipments are projected to grow about 44.6% year over year in 2026, and custom ASICs are expected to capture roughly 27.8% of the AI server market by 2026, according to Crypto Briefing. TrendForce data cited by The Motley Fool shows custom processors growing an estimated 45% this year, against 16% growth for GPUs.
Broadcom's custom AI chips backlog
Broadcom accounts for an estimated 60% to 70% of custom AI ASIC design services work as of October 2026 and sits on a backlog of roughly $73 billion in AI-related contracts, according to Crypto Briefing. That combination gives the company an unusually clear view of future revenue.
A few months ago, Alphabet signed a multiyear deal with Broadcom for custom processors through 2031, The Motley Fool reported. The financial terms were not disclosed, but the agreement is helping accelerate Broadcom's custom processor sales. Broadcom's reported AI semiconductor revenue reached $16.7 billion in fiscal Q3 2026, up 221% year over year, according to Crypto Briefing.
Broadcom's leadership has told investors that AI semiconductor revenue should hit $56 billion this year before climbing past $100 billion next year, according to The Motley Fool. The Motley Fool also noted that Broadcom's stock trades at a forward price-to-earnings ratio of 19, which it described as cheaper than the broader tech sector, and said Broadcom could have the most potential of the three companies because its sales and earnings are growing faster than those of TSMC and Marvell.
Marvell's Google deal and Nvidia investment
Marvell designs custom silicon behind Amazon's Trainium and Microsoft's Maia AI processors, according to The Motley Fool. That matters because Amazon and Microsoft operate the two leading cloud computing platforms, and Microsoft's Azure recently surpassed $100 billion in annual revenue for the first time on rising AI compute demand.
Google accounts for much of Marvell's optimism. The two companies have built a partnership worth a potential $120 billion in cumulative agreements as of August 2026, according to Crypto Briefing. Marvell expects its custom business revenue to exceed $12 billion by fiscal 2029 and projects that figure could reach about $30 billion by fiscal 2031.
Speaking to investors in October 2026, Marvell laid out revenue goals of $70 billion to $90 billion by fiscal 2031, Crypto Briefing reported. Marvell also recently secured a $2 billion investment from Nvidia, which gives Nvidia's customers access to Marvell's custom processors, according to The Motley Fool. The arrangement shows how intertwined the AI processor market has become, with custom chips working alongside traditional GPUs.
TSMC gains from custom AI chips either way
Roughly 73% of processors worldwide come out of TSMC's fabs, along with an estimated 90% of the most advanced chips, custom or otherwise, according to The Motley Fool. That position means the foundry benefits regardless of how the contest between GPUs and custom AI chips plays out.
In August 2026, TSMC posted revenue 53% higher than a year earlier, lifted by orders for AI-related wafers, according to Crypto Briefing, which noted that the figure reflects demand across the AI chip spectrum rather than custom designs alone. The company's second-quarter sales rose 36% to more than $40 billion, and non-GAAP diluted earnings increased 77% to $4.31 per American depositary receipt, The Motley Fool reported.
Nvidia's share, concentration risk and what to watch
The rise of custom processors comes alongside a projected reduction in Nvidia's share of the AI server market from around 70%, according to Crypto Briefing, though that shift is unfolding inside a market that is growing quickly. Nvidia gains when Marvell's custom chips ride inside its network offerings, The Motley Fool reported, yet the spread of custom AI chips could eventually eat into Nvidia's market share.
Traditional AI chip companies are not standing still. AMD recently won a deal with Meta worth up to $60 billion and holds AI compute agreements with OpenAI and Anthropic, and demand for its processors lifted second-quarter sales 50% to $11.5 billion, The Motley Fool reported. Intel, despite a 188% share price gain over the past year and investments from the U.S. government and Nvidia, is expected by analysts to grow sales just 15% next year, and its foundry business lost $2.1 billion in the second quarter.
The risks for the custom chip winners center on concentration. When a market depends on a small group of very large customers, the loss of even a single relationship would sting, and much of Marvell's forecast rests on agreements described as potential rather than locked in, according to Crypto Briefing. The signals to track are hyperscaler capital spending plans, any change to Broadcom's backlog figure, whether Marvell's big agreements convert into booked revenue on schedule, and TSMC's monthly revenue reports as a real-time gauge of AI silicon production.
Conclusion
The move by Google, Amazon, Microsoft and other hyperscalers toward custom AI chips is established in the numbers: Broadcom holds an estimated $73 billion AI backlog and a 60% to 70% share of ASIC design services, Marvell counts potentially $120 billion in cumulative Google agreements, and TSMC posted 53% revenue growth in August 2026. The next concrete markers are whether Marvell's potential agreements convert into booked revenue on schedule, any change to Broadcom's backlog, and TSMC's monthly revenue reports, which offer a running measure of how much AI silicon is actually being produced.
Frequently Asked Questions
How large is Broadcom's AI backlog in October 2026?
Crypto Briefing estimates Broadcom's AI-related contract backlog at roughly $73 billion as of October 2026. The company also controls an estimated 60% to 70% of custom AI ASIC design services work, and its reported AI semiconductor revenue hit $16.7 billion in fiscal Q3 2026, up 221% from a year earlier.
What is Marvell's deal with Google worth?
Marvell's partnership with Google includes potentially $120 billion in cumulative agreements as of August 2026, according to Crypto Briefing. Marvell expects custom business revenue to exceed $12 billion by fiscal 2029 and projects it could reach about $30 billion by fiscal 2031, though a large share of that outlook depends on agreements that are not yet firm commitments.
How fast are custom AI chips growing compared with GPUs in 2026?
Custom AI ASIC shipments are projected to grow about 44.6% year over year in 2026, and TrendForce data cited by The Motley Fool shows custom processors growing an estimated 45% this year against 16% for GPUs. Custom ASICs are expected to capture roughly 27.8% of the AI server market by 2026.
Why did Nvidia invest $2 billion in Marvell?
The $2 billion stake gives Nvidia's customers access to Marvell's custom processors, according to The Motley Fool. The deal shows how intertwined the AI processor market has become, with custom chips working alongside traditional GPUs even as custom silicon could eventually erode Nvidia's market share.
How much of the world's advanced processors does TSMC manufacture?
About 73% of all processors globally roll off TSMC's production lines, and the foundry makes an estimated 90% of the most advanced chips, custom or otherwise, according to The Motley Fool. That position means TSMC benefits whether GPUs or custom AI chips gain ground, and its revenue grew 53% year over year in August 2026 on AI wafer demand.
Sources
