AWS will be a $1T business, Jassy says — but first, it will spend big

Amazon CEO Andy Jassy is placing a massive bet on the future of computing, and he is being remarkably transparent about the price tag. During the latest earnings cycle, the message from Seattle was clear: Amazon Web Services is on a trajectory to become a trillion-dollar revenue engine, but getting there requires a capital expenditure program that would make most nations flinch.

The Math of the Trillion-Dollar Cloud

For years, we have seen AWS dominate the infrastructure layer of the internet. If you are building an app, a SaaS platform, or even a rival crypto exchange, there is a high probability your data lives in an Amazon data center. But Jassy is no longer just talking about being the world's hard drive and web server. He is positioning AWS as the fundamental factory for artificial intelligence.

The goal of becoming a trillion-dollar business is ambitious, even by Amazon standards. To put that in perspective, AWS currently generates around $100 billion in annualized revenue. Jassy is essentially predicting a 10x growth curve from an already massive base. This is not just incremental growth; it is a total transformation of how global compute is distributed and sold.

The Capex Wall

To reach this milestone, Amazon is spending aggressively. The company signaled that its capital expenditures for 2024 and 2025 will be significantly higher than in previous years, largely driven by the demand for generative AI. They are buying every GPU they can get their hands on and, perhaps more importantly, doubling down on their own custom silicon.

For those of us building in the space, this spending spree is a double-edged sword. On one hand, it ensures that the compute power we need will be available. On the other, it signals a massive consolidation of power. If Amazon is spending $75 billion or more a year on infrastructure, they are building a moat that virtually no one else can cross, save for maybe Microsoft or Google.

Why Builders Should Care About Custom Silicon

One of the most interesting parts of Jassy’s strategy is the shift toward Trainium and Inferentia chips. Amazon realizes that relying solely on Nvidia is a bottleneck for their margins and their customers' wallets. By building their own AI chips, they are trying to offer a lower-cost alternative to the standard industry hardware.

If you are a founder running high-inference workloads, this is where the real battle will be fought. If AWS can prove that their proprietary chips offer better price-to-performance ratios than the standard Nvidia H100s, they will lock in a new generation of AI startups who simply cannot afford to build elsewhere. This is the classic Amazon playbook: lower the barrier to entry, increase the scale, and become the unavoidable utility.

The Skeptic's View: Diminishing Returns?

As a founder, I have to ask: at what point does this infrastructure investment hit a wall? Jassy admits that the demand for AI currently outstrips their capacity, which justifies the spending. But we have seen tech cycles before where over-provisioning leads to a massive glut. If the actual utility of generative AI doesn't keep pace with the cost of the hardware, Amazon could find itself sitting on billions of dollars of depreciating silicon.

Furthermore, the move toward decentralized compute and edge AI could eventually chip away at the centralized cloud model. While AWS wants to be the trillion-dollar center of the universe, the cost of moving data in and out of their ecosystem—the dreaded egress fees—remains a major pain point for builders. A trillion-dollar AWS is a world where Amazon taxes nearly every digital interaction.

The Founder's Takeaway

For those of us in the trenches, the takeaway is simple. The era of cheap, general-purpose compute is transitioning into the era of expensive, specialized AI compute. Amazon is betting the farm that you will need their hardware to survive. Here is how I see the landscape shifting for builders:

• Diversify your stack: Relying on a single provider for AI compute is becoming increasingly risky as these giants lock you into their specific silicon ecosystems.
• Watch the margins: As AWS spends big, they will be looking for ways to recoup that investment. Don't expect the price of premium compute to drop anytime soon.
• Optimization is the new growth: If you can optimize your models to run on cheaper, custom hardware like Trainium, you have a massive competitive advantage over those just burning VC cash on standard instances.

Andy Jassy is playing a long game. He is building the physical foundation for the next twenty years of technology. Whether or not AWS hits that trillion-dollar mark depends less on their ability to build data centers and more on our ability to build AI applications that actually provide enough value to pay the bill.

The cloud isn't just a place to store files anymore; it's the raw material for intelligence. Amazon is trying to own the mine, the refinery, and the distribution network.

We are watching a massive consolidation of the infrastructure layer. For builders, this means we get better tools, but we also face a more powerful landlord. It is time to start thinking about how to build in a world where the cloud is no longer just a service, but a trillion-dollar sovereign entity.

Read the original at TechCrunch AI →