The previous article discussed Intel’s preparation to embrace the ASIC design wave, and this article will focus on the key players in the ASIC field.
In the current rapid development of artificial intelligence, Application-Specific Integrated Circuits (ASICs) have become the core engine driving technological innovation. This highly customized chip design not only far exceeds general-purpose processors in performance and energy efficiency but also tailors solutions for scenarios such as AI training, edge computing, and high-speed networking. Unlike flexible but less efficient programmable chips, ASICs help companies minimize costs and maximize innovation through targeted optimization, particularly shining in data centers, 5G/6G infrastructure, and smart automotive sectors.
According to the latest market insights, the global ASIC market size is expected to exceed $25 billion by 2025, with a potential doubling in growth over the next five years. This explosive expansion is driven by the surging demand for AI applications. From cloud giant models to end devices, efficient dedicated chips are indispensable. Many tech giants have turned to self-developed ASIC paths, such as Apple’s first iPhone processor and Google’s Tensor Processing Unit (TPU), which quickly got off the ground with external design services. This ecological collaboration model not only accelerates product iteration but also fosters full-chain innovation from design to manufacturing.
The main players in the ASIC field can be divided into two categories: on one hand, there are integrated semiconductor companies like Broadcom and Marvell, which incorporate ASICs into their core product lines; on the other hand, there are specialized companies focused on service, such as Alchip and AION Silicon, which avoid direct competition with clients and provide pure custom support. Below is a brief analysis of several representative companies that are seizing market opportunities through AI-driven strategies.
As a giant in the semiconductor field, Broadcom’s ASIC business covers areas such as network storage, broadband access, and AI acceleration. In the second quarter of 2025, the company set a historic revenue record, with the custom chip division’s gross margin consistently exceeding 50%, thanks to deep ties with hyperscale customers like Google. Recently, Broadcom also partnered with OpenAI to advance a 10-gigawatt AI accelerator project, which not only strengthens its dominant position in data centers but also drives its stock price to a cumulative increase of over 65% for the year. The secret to Broadcom’s success lies in seamlessly integrating standard silicon wafers with personalized designs, facilitating the large-scale deployment of AI infrastructure.
Marvell, on the other hand, focuses on data infrastructure and has recently shifted significantly towards AI and 5G custom ASICs. Originally concentrating on storage controllers, it now emphasizes low-power, high-speed SoCs and interconnect chips. In the third quarter of fiscal year 2025, its revenue is expected to grow by 36% year-on-year, primarily driven by the surge in AI network demand. The company’s collaboration with NVIDIA is particularly noteworthy, providing scalable custom solutions through NVLink Fusion technology to help hyperscale users accelerate model training. Marvell’s agile transformation has allowed it to stand out in the cloud computing wave, with unlimited potential for the future.
Founded in 2003, Taiwan’s Alchip is renowned for its expertise in high-performance computing (HPC) and AI. As a leader in the fabless model, it excels in rapid prototype iteration and high first-silicon success rates, closely collaborating with top foundries like TSMC. In the second quarter of 2025, Alchip’s revenue reached $297.4 million, with nearly 80% coming from the North American market. The company recently joined the Arm Total Design ecosystem and partnered with Ayar Labs to launch AI data center co-packaged optics (CPO) technology, achieving ultra-high bandwidth and low-latency interconnects between racks. Additionally, it is involved in the development of the AWS Trainium 3 chip, expected to enter mass production in early 2026. These actions not only enhance its competitiveness in sustainable silicon design but also provide efficient, low-power end-to-end solutions for AI clusters.
In contrast, AION Silicon in the United States, though smaller in scale, is known for its innovation and flexibility, focusing on custom chips for AI and IoT edge applications. In 2025, the company joined the Intel Foundry Accelerator Value Chain Alliance, becoming a one-stop design and manufacturing partner, and secured a $12 million contract for RISC-V HPC and AI projects. At the TSMC OIP ecosystem forum, AION Silicon’s advanced SoC design leadership garnered significant attention. Its 2025 roadmap emphasizes low-power AI accelerators, providing highly customizable smart device solutions for cost-sensitive markets. Despite starting later, AION Silicon is quietly building barriers in niche areas through strategic alliances with foundries.
Other companies like Arm, Qualcomm, and MediaTek are also venturing into the custom silicon business, further heating up this sector. However, in this fiercely competitive environment, Intel’s latest moves are particularly noteworthy. After the appointment of new CEO Pat Gelsinger, the company quickly pushed for organizational restructuring, including the establishment of the Central Engineering Group. This engineering group integrates horizontal engineering functions, covering foundational IP development, test chip design, EDA tools, and design platforms, aiming to eliminate redundancy, shorten decision-making cycles, and enhance product consistency. At the same time, it will lead a new ASIC and design services business, providing a full spectrum of solutions from general computing to fixed-function for external customers.
Gelsinger emphasized that through NVIDIA NVLink interconnects, Intel will deeply integrate its x86 architecture leadership with NVIDIA’s AI acceleration expertise, enhancing customer experience and paving the way for Intel to establish a foothold in tomorrow’s AI platforms. The core of this strategy lies in recruiting top architectural talent, reshaping the core roadmap, and leveraging Intel Foundry’s manufacturing and packaging capabilities to create custom ASICs around Intel/NVIDIA IP. Unlike Broadcom or Marvell, which expand through acquisitions, Intel opts for an endogenous layout, avoiding direct confrontation with TSMC’s value chain alliance and instead focusing on ecological collaboration. This is not only a pragmatic move to fill the foundry but also a clever way to inject AI vitality into x86. In the ASIC battlefield, Intel’s actions may reshape the landscape, helping it shift from defense to offense.
In conclusion, the golden age of the ASIC market has arrived. In the face of the opportunities and challenges of the AI revolution, companies need to prioritize collaborative innovation, and Intel’s transformation path provides a model for this. In the future, whoever can more quickly convert custom silicon into commercial momentum will dominate the new era of semiconductors.