Google is on track to roll out custom-designed Arm-based data center server chips starting in 2025, according to a report by The Information. The initiative marks a major push by the search giant to reduce hardware costs, improve energy efficiency, and narrow the performance gap with cloud infrastructure leader Amazon Web Services.
Inside Google’s Custom Server Chip Strategy
Google’s server chip research and development team is currently developing two distinct 5-nanometer, Arm-based chips, according to reporting from Tom’s Hardware. The primary design, known internally as ‘Cypress,’ is an in-house project managed by Google’s engineering team in Israel. The secondary backup design, codenamed ‘Maple,’ utilizes the foundations of a Marvell Technology system-on-chip and is currently in trial production at TSMC.
Both silicon initiatives are overseen by Uri Frank, a 25-year Intel CPU design veteran who joined Google as vice president of engineering for server chip design in March 2021. According to sources cited by Tom’s Hardware, Frank views the in-house Cypress design as Plan A, with the Maple architecture waiting in reserve as Plan B. Mass production for the new processors could begin as early as 2024, paving the way for data center deployment by 2025.
Prior Silicon Development and Competitive Pressures
Google is no stranger to custom silicon. The company introduced its Tensor Processing Unit (TPU) in 2015, an application-specific integrated circuit designed to accelerate artificial intelligence and neural network workloads within its TensorFlow framework. Over subsequent years, Google expanded custom chip usage into mobile devices with its Tensor system-on-chip line, utilized in Pixel smartphones and manufactured by Samsung on a 5nm process.
Despite these successes, Google has continued to rely heavily on third-party CPUs and GPUs for standard cloud computing tasks. Developing custom Arm-based server processors gives the company greater architectural control and offers a path toward significant total cost of ownership reductions. The hardware pivot directly challenges established x86 server hardware manufacturers like Intel and AMD while responding to Amazon’s early market dominance with its AWS Graviton processor series, which launched its third generation of custom Arm-based server silicon.
Project Tembo Data Center Expansion
Alongside its hardware developments, Google continues to scale its physical data center footprint. The company is a co-developer behind Project Tembo, a massive data center project located in Laramie County, Wyoming, according to reporting from Cap City News. Formerly known as Project Jade, the development sits off Terry Ranch Road near the Colorado-Wyoming border.
Unlike standard facilities that plug directly into municipal power grids, Project Tembo will operate within unincorporated Laramie County and generate its own power via a dedicated natural gas energy generation hub. Initial infrastructure costs for the site represent a $7 billion investment, with total costs expected to scale up to $50 billion as the data center reaches full operating capacity. Project representatives noted that the facility will draw water from the Lance Formation Aquifer while utilizing closed-loop cooling systems to maintain low water consumption.