Google is developing a custom AI chip codenamed 'Frozen v2' optimized for its Gemini model, targeting 2028 server deployment with expected 6-10x power efficiency improvements per token processed. Intel announced additional layoffs in its data center division affecting over 5,000 employees in California and Oregon as part of a broader efficiency strategy, while AMD launched its first rack-scale AI system 'Helios' with Microsoft as a customer for Azure data centers. TSMC expanded its Arizona investment by $100 billion to $265 billion citing sustained AI chip demand. These developments reflect intensifying competition and aggressive investment in AI infrastructure despite recent stock market corrections, with JP Morgan stating on May 20 that AI stock declines represent cyclical rotation rather than long-term downturn.
Google Develops Gemini-Optimized Custom AI Chip Targeting 2028 Deployment
Google is developing a new custom chip optimized for its Gemini AI model under the internal codename 'Frozen v2', according to multiple sources cited by technology publication The Information. The chip targets 2028 server deployment and incorporates an idea from Google Chief Scientist Jeff Dean to pre-embed Gemini's data processing methods and weights directly into the hardware. The name 'Frozen' derives from the concept that the chip will be permanently fixed in a form optimized for Gemini. Unlike existing TPUs designed for various AI models similar to NVIDIA graphics processing units, Frozen v2 will be specifically tailored to Gemini's architecture. Google expects the chip to process 6-10 times more tokens per power unit and deliver faster response speeds when running Gemini. The development reflects the severity of AI computational resource shortages facing the industry.
Intel Conducts Data Center Layoffs Affecting 5,000+ Employees
Intel is conducting additional layoffs in its data center division as part of what the company describes as a "broader strategy" to become more focused and efficient. Over 5,000 employees in the United States have been included in the layoffs to date, with most affected workers located in California and Oregon. An Intel spokesperson stated in a release that the company is "reorganizing the data center group's structure to align with the roles and capabilities needed for long-term business success as part of a broader strategy to become a more focused and efficient company." Intel has been pursuing efforts to transform into a "leaner, faster, and more efficient" enterprise to restore competitiveness in artificial intelligence and advanced semiconductor manufacturing. Intel is scheduled to announce Q2 2026 earnings after market close on July 23.
AMD Launches Helios Rack-Scale AI System with Microsoft as Customer
Semiconductor company AMD unveiled its first rack-scale artificial intelligence system 'Helios', competing directly with NVIDIA's Grace Blackwell and Vera Rubin systems. Microsoft announced it is joining Meta, OpenAI, and Oracle as a customer and will deploy Helios in Azure data centers. Helios represents AMD's first rack-unit AI system offering. AMD plans to begin product shipments to Microsoft and other major customers starting at the end of this year. The collaboration between AMD and Microsoft is not new—AMD chips have long been integrated into Surface PCs and Xbox consoles, and in 2023 Microsoft became the first company to adopt AMD's MI300X GPU as an alternative to NVIDIA AI chips.
TSMC Expands Arizona Investment to $265 Billion
Taiwan Semiconductor Manufacturing Company announced it will expand its Arizona investment by $100 billion to $265 billion (approximately 393 trillion won) following record quarterly results. According to Reuters and major foreign media on May 20, TSMC stated it is increasing investment scale by an additional $100 billion based on continued structural demand expected to persist for years. Chief Financial Officer Wendell Huang stated, "We continue to see strong demand from customers and multi-year structural demand. We are doing everything we can everywhere we can to support customer growth. We want to show that we have no intention of leaving food for anyone else." Regarding fundraising, CFO Huang responded to questions about new share issuance by saying, "If market conditions are favorable, we will not rule out issuing new bonds."
JP Morgan Analysis Characterizes AI Stock Corrections as Cyclical Rotation
JP Morgan stated on May 20 that recent sharp declines in AI-related stocks on global markets represent adjustments due to cyclical rotation rather than the beginning of long-term decline. The memory semiconductor-heavy KOSPI index fell 25% from its peak, while the Philadelphia Semiconductor Index (SOX) dropped 20%. Major AI and memory semiconductor stocks including Samsung Electronics and Micron Technology declined 20-50% from their peaks. However, JP Morgan noted that despite double-digit declines in major AI stocks, the MSCI Global Index remains only 1-2% below its all-time high. The firm highlighted that European semiconductor companies continue to see upward revisions in 12-month earnings forecasts despite underperforming stock prices relative to the market, indicating a widening gap between stock prices and earnings. JP Morgan stated "fundamentals remain solid" and advised "investors need to increase semiconductor sector allocation this summer." The firm maintained an optimistic outlook on memory market conditions, projecting that DRAM and NAND flash supply-demand imbalances will likely continue until 2028 due to expanded AI server investment and prioritization of high-bandwidth memory (HBM) production limiting general memory supply.
FAQ
What is Google's Frozen v2 chip and when will it be deployed?
Frozen v2 is a custom AI chip Google is developing specifically optimized for its Gemini model, with an internal target of 2028 for server deployment. The chip pre-embeds Gemini's data processing methods and weights directly into the hardware and is expected to process 6-10 times more tokens per power unit compared to existing solutions.
How much is TSMC investing in Arizona facilities?
TSMC expanded its Arizona investment by $100 billion to a total of $265 billion (approximately 393 trillion won). The expansion was announced following record quarterly results and is based on sustained multi-year structural demand for AI chips.
