News
Google attempts to attract TPU customers during Nvidia GTC event in San Jose
Tuesday, March 17, 2026 at 11:16 PM
Google is reportedly engaging in business development activities in San Jose during Nvidia's GTC event to attract customers to its Tensor Processing Unit (TPU) hardware. This indicates a direct competitive push in the AI accelerator market as cloud providers seek alternatives to Nvidia's GPUs.
Context
At the NVIDIA GTC 2026 conference in San Jose, Google is reportedly making a direct play for NVIDIA's core customer base by promoting its TPU (Tensor Processing Unit) infrastructure. While NVIDIA continues to dominate the training market with an estimated 80-90% share, Google is highlighting the economic advantages of its custom ASICs for inference. Recent industry data suggests that Google TPU v7 can offer 30-50% lower costs than NVIDIA GPUs, with some enterprises like Midjourney reportedly achieving a 65% reduction in inference spend after migrating to Google Cloud pods.
During the event, Google's presence includes technical sessions on its AI hypercomputer architecture and a joint Federal Start-Up Accelerator workshop with NVIDIA and Carahsoft. The competitive tension centers on the shift toward inference, which is projected to represent 75% of future AI compute. Google aims to leverage its 4.7x better cost-per-dollar efficiency to capture a larger portion of the market, which is expected to reach a 5-6% share by the end of 2026.
Sources (6)
Academic Grant Program for Researchers | NVIDIA Google - GovEventsNvidia to Google TPU Migration 2025: The $6.32B Inference Cost CrisisEquity Insights: Google TPU v7 vs NVIDIA GPUs: The AI Hardware Battle Explained | Investment InsightsTPU vs GPU: Which AI Hardware Should You Choose? | HorizonIQGoogle TPU vs NVIDIA GPU: A Comprehensive Technical Comparison - LoveChip
Related Companies
Nvidia
NVDA
Google
GOOGL