Enough with supply chain disruptions: Big Tech companies are taking chip design into their own hands.
As the saying goes, if you want a job done right, do it yourself. And Big Tech companies with the skilled labor and resources to invest in in-house semiconductor chip design are doing just that.
For some, in-house chip design offers a welcome solution to the widespread chip shortage spurred by a surge in demand for personal electronics, IoT products, emerging electric vehicle technology, and AI/ML advances with high computing power requirements.
How Are Companies Scaling Up In-House Semiconductor Chip Design?
Prices skyrocket when limited availability meets high demand. NVIDA’s industry-leading H100 GPUs are being sold for over $40,000 — so it’s not surprising to hear rumors of in-house semiconductor chip design at companies like Microsoft specifically to sidestep these high price tags. “The trend could eventually threaten Nvidia and other chipmakers, who have exploded in the AI boom,” says Richard Nieva, technology reporter and senior writer at Forbes.
The initial capital investment required to begin in-house semiconductor chip design can be cost-prohibitive to smaller companies, and there are inevitable delays when starting a new venture. Yet, for those who manage to overcome these hurdles, there are inherent advantages to getting exactly what they need out of their chips. As shortages restabilize, companies that have already adjusted their systems and made the investments will likely continue to produce their own.
“Building custom silicon solutions, especially for the first time, is a significant undertaking. From this initial program, we have learned invaluable lessons that we are incorporating into our roadmap, including architectural insights and software stack enhancements that will lead to improved performance and scale of future systems,” said Meta on the release of its first generation of the Meta Training and Inference Accelerator (MTIA) ASIC in May 2023.
In-House Chip Design in Action
Meta and Microsoft are hardly the only companies shifting to in-house semiconductor chip design. Apple’s M1 was introduced to consumers in November 2020, followed by its M2 release in 2022, and the company is set to release the R1 in early 2024 specifically for real-time sensor processing in its upcoming Vision Pro virtual/augmented reality headset.
With the ability to tailor designs for specific use cases and budgets, tech companies are rapidly shifting to acquire existing chip manufacturers or fabless models. In-house design also means companies benefit from in-house innovations — like how Google’s DeepMind is using artificial intelligence to automate optimized AI chip designs, which will be used to give their new chips a boost.
In December 2023, Google announced its fifth generation of Tensor Processing Units (TPUs) — application-specific integrated circuits (ASICs) specifically designed to accelerate machine learning algorithms. The new TPU trains large language models (LLMs) 2.8 times faster than the previous model, with 2.3 times price performance improvements. In a similar vein, Amazon is on the second generation of its Trainium machine learning training accelerator and Inferentia deep learning AI accelerator chips.
While semiconductor design companies aren’t going away any time soon, the winds are certainly shifting. With Meta, Microsoft, Apple, Google, Amazon, and other major tech companies transitioning to in-house semiconductor chip design as a way of minimizing risk in their supply chains, capitalizing on in-house innovation, and keeping costs low, it’s likely that both the semiconductor and information technology industries will continue to evolve in tandem.
Read more: