The competitive landscape of the AI chip market is facing structural reshaping. Anthropic is exploring the possibility of self-developing chips, while Amazon is considering selling its internal chip business externally—two movements pointing to the same trend: Tech companies are accelerating their search for alternative paths to bypass NVIDIA.According to media reports citing three people familiar with the matter, AI startup Anthropic is evaluating the feasibility of designing its own chips to address the compute power shortages required to support more advanced AI systems. Meanwhile, Amazon CEO Andy Jassy is reportedly weighing the possibility of selling self-developed chips to external customers, with the internal chip business's annual revenue potential exceeding $20 billion. If operated independently, the annualized revenue could reach $50 billion.The simultaneous emergence of these two reports puts pressure on NVIDIA's dominance in the AI chip market. As cloud service providers and AI companies increase investment in custom chips, market research firm TrendForce predicts that the share of ASIC-based AI server shipments will climb from 27.8% in 2026 to nearly 40% by 2030, and NVIDIA faces long-term erosion.It is worth noting that Anthropic's self-developed chip plan is still in its early stages and may ultimately lead to nothing; Amazon's chips are currently still provided through AWS under a rental model. However, these two movements have sent a clear signal: tech giants and AI companies are actively reducing their dependence on NVIDIA.<h2>Anthropic's Self-Developed Chips: Early Stages, Cost Threshold Around $500 Million</h2>Anthropic's exploration of self-developed chips comes against the backdrop of rapidly expanding demand. The company disclosed this week that the annualized revenue of its AI model Claude has surpassed $30 billion, while this figure was only around $9 billion at the end of 2025.Media reports citing sources familiar with the matter revealed that the plan is in its early stages, with the company having neither determined a specific design nor assembled a dedicated team, and it may still choose direct procurement over self-development. Industry insiders cited by the media pointed out that the cost of designing an advanced AI chip is approximately $500 million, covering engineer salaries and quality control investments in the manufacturing process.Currently, Anthropic uses a diverse range of chips for training and running Claude, including AWS's Trainium, Google's TPU, and NVIDIA's GPU. The exploration of self-developed chips reflects the company's intention to seek higher autonomy in supply security.Meanwhile, Anthropic just signed long-term agreements with Google and Broadcom this week. Broadcom will provide Anthropic with approximately 3.5 GW of AI computing resources using Google's AI processors, with supply planned to begin in 2027. Broadcom has also reached a long-term cooperation with Google to develop and supply custom chips and related components for Google's next-generation AI racks, with the partnership extending to 2031. Exploring self-developed chips while locking in external computing resources reflects Anthropic's dual-track strategy for dealing with supply risks.<h2>Amazon's Chip Sales: From Cloud Infrastructure to Semiconductor Supplier</h2>Amazon is studying the feasibility of selling its self-developed chips to external customers, breaking the existing model of only leasing them within AWS. Andy Jassy stated that the internal chip business's annual revenue could potentially exceed $20 billion; if this business is spun off as an independent entity to supply semiconductors to AWS customers and other third parties, the annualized revenue scale could reach approximately $50 billion.The core driver of this shift is the tight supply and demand for AI training processors. As demand for computing power surges, companies are actively seeking alternatives to NVIDIA products, creating a market window for Amazon to expand its chip sales business.<h2>ASIC Acceleration and Pressure on NVIDIA's Market Share</h2>The strategic layouts of Anthropic and Amazon are not isolated cases. According to reports, Meta and OpenAI are also advancing their own chip development plans, making the collective acceleration of large tech companies in the custom AI chip field a clear trend.TrendForce data shows that as cloud service providers such as Google and Amazon continue to increase investment in internal chip research and development, the share of ASIC-based AI server shipments in total AI server shipments will gradually increase from 27.8% in 2026 to nearly 40% by 2030. NVIDIA's share in the AI computing market will face continuous and long-term structural pressure.

NVDA

AMZN

NVDU

Anthropic is exploring in-house chip development to address computing power shortages, while Amazon is considering the external sale of its internal chip business, which has a potential annual revenue scale of $50 billion. Both moves aim to reduce reliance on NVIDIA. As giants like Meta and OpenAI accelerate custom ASIC development, the proportion of related servers is expected to rise to 40% by 2030, posing a severe challenge to NVIDIA's long-term dominance

- The AI chip market is experiencing structural changes as Anthropic explores the potential for self-designed chips and Amazon considers selling its internal chip business.  
- These moves signal a trend among tech companies seeking alternatives to Nvidia, with Amazon's chip revenue projected to exceed $20 billion annually.  
- Industry forecasts suggest that the share of ASIC-based AI servers will rise from 27.8% in 2026 to nearly 40% by 2030, indicating ongoing pressure on Nvidia's market dominance.

NVIDIA "Under Siege": Anthropic to Build AI Chips, Amazon to Sell Chips