By Bo Yang, Tencent Technology
Anthropic is building its own chip and, like Musk, has chosen Samsung over TSMC.
The Information, citing three knowledgeable sources, reported that Anthropic has initiated early work on its proprietary AI chip and is in discussions with Samsung to manufacture the chip using Samsung’s most advanced 2-nanometer process and packaging technology.
The aforementioned insider said that Anthropic is still determining the chip’s functionality, performance metrics, and its integration方案 within servers or clusters. The company has already begun discussions with several design service providers, but it is far from the detailed design, testing, and mass production stages. The insider also cautioned that Anthropic could still abandon the project midway.
Public records show that Anthropic’s chip team has already begun forming. For example, the company recruited Clive Chan, an early member of OpenAI’s custom chip team, on June 9.
Clive Chen, an early member of OpenAI’s custom chip team, has joined Anthropic.
Anthropic has been very cautious in its public statements about its self-developed chips, emphasizing only that future compute expansion will still primarily rely on AWS’s Trainium, Google’s TPU, and NVIDIA’s GPUs, and has declined to comment on its own chip roadmap.
In fact, rumors about Anthropic developing its own chips had been circulating for several months. In April, Reuters reported that Claude’s computing demands were outpacing available capacity, prompting Anthropic to consider building its own chips—but at the time, it was still just an initial idea. Now, it appears the project has taken a step forward.
It should be noted that the news of Anthropic entering the chip design market emerged at a time when global semiconductor stocks had just experienced two consecutive trading days of sharp declines. The Philadelphia Semiconductor Index fell nearly 12% over the two days, marking its largest drop since June 5. Memory chip stocks were hit especially hard, with Samsung Electronics dropping 9.1% overnight and SanDisk falling 14.1%.
01 A financing round that set the stage, and the battle to break through in Samsung’s contract manufacturing business
It is not surprising that Samsung has become a partner of Anthropic.
In May of this year, Anthropic completed a $65 billion Series H funding round, with Samsung Electronics participating as a strategic investor, alongside memory chip giants SK Hynix and Micron Technology.
At the time, when memory chip supplies were tight, consumer electronics manufacturers like Apple were forced to raise prices; one direct benefit of this investment was tying Anthropic to the chip suppliers it relies on for growth.
But Samsung's ambitions extend beyond financial returns.
The company is the leader in memory chips but has consistently lagged behind TSMC in contract manufacturing. TSMC’s advanced production lines remain the industry standard for cutting-edge AI chips, but current demand for AI chips has strained TSMC’s capacity, creating an opportunity for Samsung to win customers with its 2-nanometer technology.
According to The Information, Google is considering having Samsung produce some of its future TPUs. Securing Anthropic as well would be a major victory for Samsung’s foundry business.
Two sources revealed that Anthropic is evaluating Samsung’s 2-nanometer manufacturing process and advanced packaging facilities. The “2-nanometer” node is among the most advanced chip manufacturing technologies in the industry, enabling more compact and energy-efficient processors. Advanced packaging places the main processor closer to high-speed memory, allowing data to move faster within the chip.
Samsung is a manufacturing partner of NVIDIA for HBM memory and LPU chips, and in turn, Samsung relies on NVIDIA’s software to produce chips. The two companies are jointly building an AI chip factory in South Korea. If Anthropic joins the customer list, Samsung will gain another advantage in catching up to TSMC in advanced processes.
Samsung's bold move to pursue Anthropic stems from South Korea's national commitment to the semiconductor industry.
The South Korean government recently announced a decade-long investment plan led by Samsung Group (parent company of Samsung Electronics) and SK Group (parent company of SK Hynix), with the two companies collectively investing $518 billion to build four memory chip factories in South Korea.
Just Samsung announced on July 2 an investment plan for the Chungcheong region exceeding 140 trillion Korean won, approximately $90 billion. Samsung Electronics will build a next-generation HBM production facility in Chungcheong and Wonyang, adding five new production lines. Since the AI boom, HBM has been in constant short supply; after expansion, Samsung is expected to generate billions of dollars in additional revenue over the coming years. Samsung Electro-Mechanics will expand its AI server package substrate capacity in Sejong City and simultaneously invest in advanced packaging technology development.
Samsung participated in Anthropic's funding round, betting on becoming its long-term hardware partner.
02 AI inference chips: A breakthrough in a costly closed loop
Anthropic's first 2nm chip is expected to focus on inference, similar to OpenAI.
Custom inference chips can make Anthropic's models run faster, consume less power, reduce cloud bills, and provide hardware-specific optimizations for Claude.
Previously, OpenAI selected Broadcom to design a推理 chip named "Jalapeño," aimed at running large language models more efficiently. OpenAI stated that the chip offers higher energy efficiency and better performance per watt than its competitors.
OpenAI partners with Broadcom to design a推理 chip named "Jalapeño"
Google has TPUs, Amazon has Trainium, Microsoft has Maia, and Meta is also developing its own. Anthropic is nearly the last major AI lab still relying entirely on others’ chips. The temptation for autonomous control in this field is too great.
In February this year, Anthropic CEO Dario Amodei discussed the logic behind compute decisions during an interview with podcast host Dwarkesh Patel.
Amody said: "If I assume revenue continues to grow 10x annually, it will reach $100 billion by the end of 2026 and $1 trillion by the end of 2027. I could then order $1 trillion in computing power. But what if I'm wrong? What if the growth factor is 5x instead of 10x? What if revenue in 2027 is only $800 billion instead of $1 trillion? Then I'd be bankrupt. No hedge could save me."
So, Amodei must find a balance for Anthropic: buy enough computing power to handle high growth, but avoid overexpanding to the point where a slowdown in growth could overwhelm the company with prior investments.
Amodi revealed that this year, global computing power capacity built is around 10 to 15 gigawatts, roughly tripling each year, and by 2028 or 2029, the industry’s annual investment in computing power could reach trillions of dollars. He believes this figure reflects the current trend the industry is following.
Developing our own chips is equivalent to increasing our stake in this gamble.
On one hand, even a slight improvement in inference efficiency from in-house chips can save a significant amount of cost in clusters scaling to gigawatts. On the other hand, when everyone is competing for processors, data centers, and power, having your own chips gives you more leverage in negotiations.
03 Supply Chain “Matryoshka Dolls”: Who Relies on Whom?
Anthropic wants chip independence, but this web is becoming tighter and tighter.
Programmer Benny Lam pointed out on social media that each U.S. AI lab is partnering with Asian manufacturers to customize chips, resulting not in reduced dependence, but in tighter supply chain integration.
Anthropic needs Samsung’s wafer fabs to reduce its reliance on NVIDIA, while NVIDIA needs Samsung’s wafer fabs to meet its growing production demands. Meanwhile, Samsung needs to retain both of these major clients to keep its foundry business competitive against TSMC.
The three parties are bound together; none can exist without the others.
Lem commented that the trend toward custom chips was supposed to give AI labs more independent capabilities, but each new collaboration between U.S. labs and Asian foundries is making this network tighter and harder to unravel.
Tech influencer Vaibhav Sisinty also noticed Samsung’s unique role in this game.
He wrote that Samsung’s investment in Anthropic’s $65 billion funding round was not solely for financial returns. As a chip manufacturer, memory supplier, and advanced packaging facility, Samsung is betting on becoming Anthropic’s hardware partner. With millions of queries daily, Anthropic must pay for computational resources with each invocation—resulting in an astronomical bill. Whoever controls its own inference chips controls the pace and cost of scaling.
04 NVIDIA's moat remains strong
Designing a competitive AI chip is extremely complex—every step, from architecture and software to manufacturing, testing, and large-scale deployment, requires significant time. Anthropic cannot quickly outpace its existing hardware partners in the short term.
From Anthropic’s current compute strategy, its approach is already very clear. The company has consistently adhered to a multi-cloud model and has not exclusively tied itself to NVIDIA hardware, unlike OpenAI or xAI. It utilizes chips from Amazon, Google, and NVIDIA, and is currently in discussions with Microsoft and the UK startup Fractile regarding chip partnerships.
Anthropic has particularly deep ties with Amazon and Google.
In April 2026, Anthropic expanded its partnership with AWS, committing over $100 billion in investment over the next decade to secure 5 gigawatts of new compute capacity, including custom chips such as Trainium2, Trainium3, and Trainium4. Amazon has now invested billions of dollars in total into Anthropic.
Meanwhile, Anthropic has also entered into long-term agreements with Google and Broadcom to secure approximately 3.5 gigawatts of next-generation TPU compute power starting in 2027. Previously, in 2025, the two parties had already signed agreements worth tens of billions of dollars, covering millions of TPUs and over 1 gigawatt of capacity.
Claude is one of the few cutting-edge models available on AWS, Google Cloud, and Azure.
This isn't over. In May 2026, Anthropic leased the entire capacity of the SpaceX Colossus 1 compute cluster from xAI.
This cluster, located in the Memphis data center, houses over 220,000 NVIDIA GPUs, consumes more than 300 megawatts of power, and has a monthly rental cost of approximately $1.25 billion. The total contract value amounts to tens of billions of dollars and extends until around 2029. The parties have even discussed the possibility of deploying multiple gigawatts of computing power in orbit.
Broad布局 and parallel multi-vendor strategies are how Anthropic mitigates risk.
But for now, the dealer at the table is still NVIDIA. According to The Information, despite ongoing growth in funding and design activity in the inference chip market, NVIDIA’s market share has risen to 74% in recent years. CEO Jensen Huang continues to assert that NVIDIA’s chips are more efficient than any alternative for inference tasks.