The Sequence Opinion #900: Beyond the GPU: Is Google the Only Full-Stack Rival to NVIDIA?
Substack Jesus Rodriguez
Google is basically the only company that mirrors Nvidia's whole AI stack, chip to cloud to apps. But 'only' full-stack rival is a bit of a stretch, AWS and AMD have real claims too.
Based on reporting by Substack, Jesus Rodriguez — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Every time someone tries to rank AI hardware players, the conversation collapses into a spec sheet contest. FLOPS here, memory bandwidth there, tokens per second as the tiebreaker. It's the same instinct that makes people compare airlines by engine thrust alone, ignoring the airframe, the maintenance crews, the scheduling software, and the route network that actually gets a plane from Newark to Tokyo on time. Nvidia's real edge was never just the H100 or the B200. It's the whole industrial pipeline: chips, interconnects, servers, compilers, software frameworks, and cloud partnerships that let a model go from research paper to production API with almost no friction.
Seen through that lens, Google starts to look like the one company running a genuinely comparable playbook. Google isn't necessarily building the fastest chip on the market, and it may lose individual benchmark fights to specialized silicon from other vendors. That's not really the point. What sets Google apart is that it controls nearly every layer Nvidia does: its own TPU silicon, custom interconnects, data center hardware, a compiler and framework stack in JAX and XLA, cloud infrastructure at Google Cloud, frontier models through the Gemini line, and consumer-facing products used by billions of people every day. Nvidia built an empire by owning the full chain from transistor to deployed application. Google is one of the only other companies with the pieces in place to do the same thing, top to bottom.
But the word
My take — AI-written commentary, not fact-checked reporting
I'll take the contrarian angle here: framing this as a two-horse race between Nvidia and Google undersells how weird and fragmented AI infrastructure actually is right now. AWS owns the distribution rails that most enterprises already run on, and AMD's MI300 series is closing the raw performance gap faster than people give it credit for. The real story isn't who's the 'only' rival, it's that vertical integration is becoming table stakes, and every hyperscaler is racing to build the same six layers because nobody wants to be a rent-paying tenant in someone else's stack.
Read more about this at: Substack