Amazon’s Andy Jassy reiterates the need to spend billions on building out AI infrastructure – News

Amazon.com Inc. Chief Executive Andy Jassy doubled down on the need for his company to invest heavily in artificial intelligence infrastructure, saying it must spend billions of dollars now to be able to participate in a “once-in-a-lifetime reinvention of everything we know.”

The comments from Jassy (pictured) came in his annual letter to shareholders, where he argued that spending this money now will ensure it sees strong returns in the future.

Amazon is already investing heavily in generative AI, having released a variety of chatbots serving sellers, businesses and consumers. Last month, after multiple delays and billions of dollars of investment, it unveiled an AI-infused revamp of its Alexa voice assistant.

Amazon has also invested about $8 billion in the AI startup Anthropic, and has incorporated its Claude AI models into what it is calling Alexa+.

During its fourth-quarter earnings call in February, Amazon revealed plans to increase its capital expenditure in fiscal 2025 to more than $100 billion, with the “vast majority” of that money to be spent on building out its data center capabilities for AI.

Explaining the need for this investment, Jassy reiterated his belief that generative AI is going to “reinvent virtually every customer experience” and enable newer experiences that were previously only fantasized about.

“The early AI workloads being deployed focus on productivity and cost avoidance (e.g. customer service, business process orchestration, workflow, translation, etc.),” Jassy wrote. “This is saving companies a lot of money. Increasingly, you’ll see AI change the norms in coding, search, shopping, personal assistants, primary care, cancer and drug research, biology, robotics, space, financial services, neighborhood networks — everything.”

Though AI in some of these areas is making rapid progress, it’s still in its infancy in many of the others, Jassy said. But he warned that if companies aren’t already planning to leverage AI and its future “agentic capabilities” in the near-term future, they will not be able to keep up with the competition.

“If your mission is to make customer’s lives better and easier every day, and you believe customer experience will be reinvented by AI, you’re going to invest deeply and broadly in AI,” Jassy explained.

That’s why Amazon Web Services Inc., the company’s cloud computing business, is working so hard to develop the key primitives, or building blocks for AI, Jassy said. An example of this is AWS’ custom AI training chips, known as Amazon Trainium, which he said provide better price-performance on training and inference workloads. Other examples include Amazon Bedrock and Amazon SageMaker, and the company’s Amazon Nova frontier models.

“The faster demand grows, the more data centers, chips, and hardware we need to procure,’ he wrote. “We spend this capital upfront, even though these assets are useful for many years.”

Besides the significant investment, Amazon must also strive to reduce the costs of AI for customers too, Jassy said.

He explained that the latest generation of Amazon’s AI training chips, Trainium2, already provides 30% to 40% better price-performance than existing graphics processing unit-based computing instances. But in the future, the most important goal is to reduce the cost of AI inference, which refers to the outputs or predictions of AI models.

“We feel a strong urgency to make inference less expensive for customers,” Jassy said.

Importantly, he believes this can be achieved through a combination of more efficient chips and improvements in areas such as model distillation, model architectures and prompt caching.

“It’s like what happened with AWS,” Jassy continued. “Revolutionizing the cost of compute and storage happily led to lower cost per unit, and more invention, better customer experiences, and more absolute infrastructure spend.”

Photo: News