Google (GOOGL) is taking direct aim at Nvidia’s (NVDA) dominance in the artificial intelligence (AI) chip market. At the company’s Cloud Next 2026 conference in Las Vegas this week, Google unveiled a new generation of its custom Tensor Processing Units (TPUs). Separately, Google is also creating two brand-new chips with semiconductor design firm Marvell Technology (MRVL), which will work with or alongside existing TPUs.
Claim 30% Off TipRanks
New trading tool for NVDA bearsGoogle Rolls Out New-Gen Chips to Challenge Nvidia
At its Cloud conference, which opened on Wednesday, Google announced its eighth-generation chips, TPU 8t and TPU 8i. The new chips are designed to handle both AI training and inference workloads, the phase where a model processes requests and generates results. This distinct setup places the chips in direct competition with the segment currently dominated by Nvidia’s Graphics Processing Units (GPUs).
Against this backdrop, Google chief scientist Jeff Dean confirmed the company is exploring chip specialization, stating that it has “now become sensible to specialize chips more for training or more for inference workloads.” Meanwhile, Nvidia CEO Jensen Huang has maintained that its chips support applications that TPUs cannot.
However, Google’s latest moves are already gaining traction among major AI firms. Anthropic recently secured access to up to 1 million Google TPUs to train and run its Claude models, while Meta Platforms (META) signed a multibillion-dollar deal to use the chips via Google Cloud.
Looking ahead, Google’s new-gen chips are expected to enter mass production in Q3 2026 using Taiwan Semiconductor Manufacturing Company’s (TSM) 3nm node, with shipments projected to grow more than 40% this year, the highest among cloud providers.
Google Partners with Marvell to Build Specialized AI Chips
Google is reportedly developing two new AI chips with Marvell Technology: a Memory Processing Unit (MPU) and a new inference-focused TPU. The MPU is built to address memory concerns and complement Google’s existing TPUs. Google and Marvell are targeting design finalization as early as next year, before moving into trial production of approximately 2 million units.
Meanwhile, the inference-specialized TPU is a custom chip that succeeds Google’s 7th-gen “Ironwood” TPUs. Unlike Ironwood, which evolved into a general-purpose chip that handles both training and inference, the new TPU is built exclusively for inference, at a lower cost and with greater power efficiency. The discussions have not yet produced a signed contract.
Is Google a Buy or Sell Right Now?
Google (GOOGL) currently carries a Strong Buy consensus rating from 31 Wall Street analysts tracked by TipRanks. 26 analysts recommend a Buy, 5 assign a Hold, and 0 suggest a Sell. The stock also has a 12-month price target of about $387.6, implying a roughly 15.2% upside from current levels. Additional information on its performance, ratings, and upgrades is available in the TipRanks Stock Comparison Center.
