Skip to main content

Amazon Doubles Down on NVIDIA GPUs, AI Arms Race Heats Up

Amazon is going all in on AI. On August 26, the tech giant announced it would deploy an additional 2 million NVIDIA GPUs across its data centers over the next two years. That's double the commitment it made just five months ago, when it pledged to deploy over 1 million GPUs. The chips—spanning Blackwell Ultra, Rubin, and Rubin Ultra—are slated to power large-scale AI model training and inference, with integration into AWS data centers planned for 2027 and 2028.

The news came on the same day NVIDIA revealed its quarterly earnings, which blew past expectations. Revenue hit $96.2 billion, with the data center segment alone bringing in $89 billion—a 117% year-over-year jump. NVIDIA also projected $108 billion in third-quarter revenue and said its next-gen Rubin chip is already in mass production. To keep up with demand, the company is investing $279 billion to secure supply and manufacturing capacity.

But here's the twist: Amazon isn't just buying chips—it's also building its own. The company has been accelerating development of its custom Trainium and Graviton processors, which now boast an annualized revenue run rate exceeding $25 billion. That's a clear signal that Amazon wants to reduce its dependence on NVIDIA, even as it places this massive order.

So why double down on NVIDIA if you're trying to wean yourself off? The answer lies in the sheer scale of demand. AI workloads are exploding, and NVIDIA's GPUs remain the gold standard for training and inference. Amazon needs the compute power now, and its own chips aren't ready to take over the heavy lifting just yet.

The partnership extends beyond hardware. NVIDIA's networking, open models, CPUs, data software, and robotics platforms will be more deeply integrated into AWS. Vera CPUs, for instance, are being paired with Rubin chips, and the integration has been progressing since the third quarter. Both companies are also eyeing warehouse automation, with Amazon adopting NVIDIA's "physical AI" stack—Omniverse, Cosmos, and Isaac—to drive its fleet of robots.

This isn't just a business deal; it's a strategic alignment. Amazon gets the compute power it needs to stay competitive in AI, while NVIDIA secures a massive customer and a showcase for its latest technology. The stakes are high, and the investment is enormous. But the real question on everyone's mind is whether these billions in infrastructure spending will translate into profits.

NVIDIA CEO Jensen Huang is optimistic, saying AI has already generated profitable "tokens." But the market remains skeptical. Can the tens of billions poured into data centers and chips yield returns that justify the cost? That's the billion-dollar question—literally.

For now, the arms race shows no signs of slowing. Amazon's move is a clear signal that the demand for AI computing power is far from peaking. And with NVIDIA's earnings confirming the trend, it looks like the AI boom is here to stay. But as with any gold rush, the real winners will be those who can turn the hype into sustainable profits.

Key Points

  • Amazon doubles NVIDIA GPU order to 2 million units, up from 1 million five months ago.
  • NVIDIA's Q2 revenue hits $96.2 billion, with data center segment up 117% year-over-year.
  • Amazon is developing its own Trainium and Graviton chips, with annualized revenue run rate exceeding $25 billion.
  • Partnership extends to networking, robotics, and warehouse automation.
  • NVIDIA invests $279 billion to secure supply and capacity.
  • Market remains cautious about whether massive infrastructure investments will yield profits.