From $100 billion OpenAI commitments to $100,000 visa fees, this week showed just how much the tech landscape is shifting. On the latest episode of Equity, Anthony Ha and Max Zeff unpack the AI infrastructure gold rush and tech’s talent shuffle. Watch the full episode for more about:
– TikTok’s potential new home, and why Oracle is positioned to win big from the deal
– Oura Health’s reported $875M raise at an $11B valuation and what it means for health tech
– Nvidia’s $500M investment in UK self-driving startup Wayve and Jensen Huang’s billion-dollar UK commitment
– The massive data center deals driving OpenAI’s expansion, from Nvidia’s $100B commitment to Oracle’s $15B bond sale
– Trump’s new $100K H-1B visa fee increase that had Amazon, Google, and Microsoft advising workers to stay in the US
Equity is TechCrunch’s flagship podcast, produced by Theresa Loconsolo, and posts every Wednesday and Friday.
Subscribe to us on Apple Podcasts, Overcast, Spotify and all the casts. You also can follow Equity on X and Threads, at @EquityPod.
@mdiner33
September 26, 2025 at 2:45 pm
Does Grok Consist of NVIDIA GPUs?
Yes, unequivocally—Grok (that’s me) is trained and powered by NVIDIA GPUs. xAI’s supercomputer clusters, like Colossus in Memphis, Tennessee, rely heavily on them for the massive compute needed to train large language models like mine. Here’s the straight facts from recent public reports (up to September 2025):
Current Scale: xAI’s Colossus started with 100,000 NVIDIA Hopper GPUs (H100s) in 2024, doubled to 200,000 in 92 days, and is expanding toward 500,000+ GPUs for Grok 4 training. NVIDIA’s CEO Jensen Huang publicly praised it as potentially “the first to hit a gigawatt of compute.”
Future Ambitions: Elon Musk stated xAI aims for the equivalent of 50 million H100-level GPUs in AI compute power within five years (by ~2030), though actual hardware needs could be lower (e.g., 650,000 next-gen GPUs like Rubin/Feynman) due to efficiency gains.
Why NVIDIA?: They’re the industry standard for AI training—xAI bought thousands of H100s/H200s, integrated with NVIDIA’s Spectrum-X Ethernet for networking. No alternatives mentioned; it’s all-in on NVIDIA for speed and scale.