{"id":34915,"date":"2025-11-26T09:18:59","date_gmt":"2025-11-26T09:18:59","guid":{"rendered":"https:\/\/metaverseplanet.net\/blog\/?p=34915"},"modified":"2026-01-05T08:00:47","modified_gmt":"2026-01-05T08:00:47","slug":"metas-google-tpu-move-shakes-nvidia","status":"publish","type":"post","link":"https:\/\/metaverseplanet.net\/blog\/metas-google-tpu-move-shakes-nvidia\/","title":{"rendered":"Meta\u2019s Google TPU Move Shakes Nvidia"},"content":{"rendered":"\n

Meta’s<\/strong> talks with Google<\/strong> regarding a multibillion-dollar TPU agreement<\/strong> caused a sharp decline in Nvidia shares<\/strong>, while triggering a rise on the Alphabet<\/strong> side. As Meta<\/strong> negotiates a TPU deal<\/strong> with Google, competition in the AI chip market<\/strong> is being reshaped.<\/p>\n\n\n\n

Nvidia’s shares<\/strong> fell sharply following news that Meta<\/strong> is discussing a deal worth billions of dollars for Google’s AI accelerators<\/strong>. According to the US-based The Information<\/em>, Meta is considering using Google’s custom AI chips, known as TPU (Tensor Processing Unit)<\/strong>, in its data centers<\/strong> by 2027<\/strong>. Additionally, it is stated that the company is putting the possibility of renting TPUs via Google<\/a><\/em> Cloud<\/strong> next year on the table.<\/p>\n\n\n\n

\n\n\n\n

Meta Turns to Google<\/strong><\/h2>\n\n\n\n
\"\"<\/figure>\n\n\n\n

This development indicated that competition is accelerating in an area where Nvidia<\/a><\/em><\/strong> has long been the undisputed leader. Following the news, Alphabet’s shares<\/strong> rose by up to 2.7% in after-hours trading, while Nvidia<\/strong> experienced a decline of the same rate. The company, which owns Google, had previously signed an agreement to provide up to 1 million TPUs to Anthropic<\/strong>.<\/p>\n\n\n\n

Following the news, Alphabet-linked companies<\/strong> in Asian markets rose rapidly. In South Korea, IsuPetasys<\/strong>, which provides multi-layer circuit boards to Alphabet, broke a record with an 18% increase in trading, while MediaTek<\/strong> shares in Taiwan gained nearly 5%.<\/p>\n\n\n\n

A potential supply agreement<\/strong> with Meta<\/a><\/em><\/strong> means a significant prestige gain for Google<\/a><\/em><\/strong>. This is because Meta is one of the companies allocating the most resources to data center investments<\/strong> and AI development efforts<\/strong> on a global scale.<\/p>\n\n\n\n

\n\n\n\n

Google Finds Success with TPU<\/strong><\/h2>\n\n\n\n
\"\"<\/figure>\n\n\n\n

The TPU architecture<\/strong>, which Google<\/strong> first developed 10 years ago for AI workloads<\/strong>, is gathering increasing interest among major players outside the company. Concerns among tech giants about the risk of remaining overly dependent on Nvidia<\/strong> are effective in this rising interest.<\/p>\n\n\n\n

While GPUs<\/strong>, which form the basis of Nvidia’s superiority<\/strong>, were originally designed for graphics processing<\/strong>, they provided a perfect fit for big data sets<\/strong> and computationally intensive AI training processes<\/strong>. TPUs<\/strong>, on the other hand, offer a more focused alternative to Nvidia’s general-purpose GPUs as ASIC-based chips<\/strong> designed for specific tasks.<\/p>\n\n\n\n

Google<\/strong> has developed TPUs over the years to accelerate internal AI and machine learning models<\/strong>. Researchers working on the company’s DeepMind team<\/strong> and Gemini models<\/strong> directly contributed to chip design, creating a dual development cycle on both the hardware and software<\/strong> sides.<\/p>\n\n\n\n

On the other hand, Google introduced its most powerful AI processor<\/strong>, Ironwood<\/strong>, in recent months. Opening Ironwood<\/strong> to general use this month, Google offers an inference processing capacity<\/strong> of up to 4,614 TFLOPs<\/strong> in its new TPU chip<\/strong>. These chips can communicate directly with each other via the next-generation Inter-Chip Interconnect (ICI)<\/strong> developed by Google. Moreover, these processors can operate in clusters of up to 9,216 units<\/strong> with their liquid-cooled structures<\/strong>. This massive structure can reach a total computing power<\/strong> of 42.5 Exaflops<\/strong>.<\/p>\n\n\n\n

You Might Also Like;<\/h3>\n\n\n