Amazon’s 2025 looked wrong on the surface. The broader market climbed strongly, the company kept growing, and yet the stock mostly went nowhere for long stretches. For a business that has trained investors to expect motion, it felt like a stall.
That disconnect is why Amazon’s next move matters more than a typical spending cycle. After all, this is a company trying to win the shape of the next decade of computing.
Amazon has committed to an AI infrastructure buildout so large it reframes every other debate around the stock. Supporters see a platform moment, critics see a capital bonfire, and both sides are arguing about the same thing: whether demand will arrive quickly enough to justify what Amazon is building.
The risk is that it picked the right direction and still overbuilt, too early, at a scale that punishes near-term cash flow and leaves little room for execution mistakes.
The $200 billion question
Amazon’s plan expects roughly $200 billion in capital expenditures in 2026, heavily tied to AI, data centers, chips, and the physical infrastructure required to run modern models at scale. That number is narrative-changing. It tells the market Amazon is betting that AI demand is durable, that it will be cloud-heavy, and that the winners will be the companies that can deliver compute without delay.
There is a second, less glamorous implication. When a company builds at this pace, the financial pain comes first. Depreciation and operating costs show up before utilization ramps, which can make earnings growth look modest even if the underlying businesses are healthy. In plain terms, Amazon can be right about AI and still look “wrong” for a while, because the bill arrives before the payoff.
That is why this story is less about hype and more about timing. If the capacity fills, Amazon looks prescient. If it does not, Amazon looks like it spent years building expensive empty rooms.
The importance of AWS
Most investors understand AWS is important, but many still underestimate how central it is to Amazon’s financial identity. AWS has been responsible for a disproportionate share of operating profit for years, and it is the division most directly positioned to turn AI demand into high-margin recurring revenue.
Amazon’s argument is straightforward: AI workloads are compute-hungry, customers do not want to wait, and cloud providers that cannot deliver capacity will lose deals they never get to price. That is why Amazon has emphasized how aggressively it has expanded power and data center capacity, because in AI, the bottleneck is often not ideas, it is electrons.
For readers who want a refresher on how the AWS ecosystem fits together, this explainer on what AWS is and how it works is a useful baseline. And if you are trying to understand the competitive framing, AWS vs Azure vs Google Cloud is the comparison investors implicitly make every time they read a cloud growth headline.
The quiet stabilizer
While AWS is the strategic battlefield, advertising is the quiet machine that can help fund the war. Amazon’s advertising business has grown into a major revenue stream, and it behaves differently than retail. It benefits from Amazon’s shopping intent data and on-site traffic, and it can produce strong margins even when consumers pull back on discretionary spending.
This matters for a simple reason. If the AI buildout compresses free cash flow in the near term, the market looks for other engines that can keep operating income moving in the right direction. Advertising is one of the best candidates, because it is already scaled and still growing.
Custom chips as a control lever
The most misunderstood part of Amazon’s AI strategy is silicon. Many investors treat custom chips as a side plot, a defensive move against Nvidia pricing, or a branding exercise. In reality, it is a leverage point. If Amazon can shift meaningful workloads onto its own chips, it controls more of the cost structure and can shape pricing in ways competitors cannot easily match.
That does not mean Amazon replaces Nvidia. It means Amazon reduces its dependence on any single supplier, improves its negotiating position, and can offer customers an alternative stack when GPU supply is tight or when price-performance matters more than ecosystem comfort.
For learners, this is also a clue about where technical value concentrates. The future is systems work. Knowing how to deploy, profile, and optimize workloads across different accelerators is increasingly part of real-world AI engineering.
The risks are real (and mostly operational)
The bull case is easy to tell, because it is clean. AI demand grows, AWS fills capacity, chip economics improve, advertising keeps compounding, and Amazon emerges as a primary utility for intelligence at scale.
The bear case is that the curve is slower, messier, and more competitive than the capex assumes. If enterprise adoption takes longer than expected, utilization lags. If rivals price aggressively, margins compress. If regulators, power constraints, or supply chain limits slow buildouts in key regions, Amazon can spend heavily while still failing to meet demand where it matters most.
There is also a subtle risk that the market does not reward the journey. A buildout of this magnitude asks investors to tolerate years where cash flow looks worse in exchange for a future that looks better. That trade only works if confidence stays intact.
What we need to see
Investors should treat 2026 as a proof-of-concept year. Not because everything must pay off immediately, but because Amazon needs to show that the machine it is building is starting to load.
- AWS growth re-accelerates and stays strong, not just for one quarter, but across multiple quarters.
- AI-related services translate into measurable customer adoption, not just “interest,” pilots, or press releases.
- Advertising continues to grow at a pace that meaningfully offsets capex-driven cash flow pressure.
- Margins hold up despite pricing pressure, especially if competitors try to buy market share in AI workloads.
- Capex discipline shows up in execution, with fewer delays, fewer costly pivots, and clearer signals about utilization.
If those pieces align, the spending starts to look less like a gamble and more like an early land grab in a market that rewards scale. If they do not, the market will eventually treat the buildout as a mispriced asset, and those are painful to unwind.
What this means for people learning AI
For people building careers in AI (and those just learning AI), Amazon’s strategy is a signal that most durable opportunities will cluster around deployment, infrastructure, and the unglamorous work of making models reliable, fast, and cost-effective in production.
If you are early in the journey, start broad, then specialize. Learn the cloud basics, then learn how teams actually run AI systems. A practical sequence looks like this:
- Understand cloud fundamentals and the AWS service landscape, then build comfort with identity, networking, and storage.
- Add DevOps skills, because most AI systems fail in the handoff between notebooks and production.
- Learn containers and orchestration, because scalable inference is an operations problem as much as a model problem.
- Build portfolio projects that prove you can ship, not just train.
Hackr has solid starting points for each step, including an AWS certification roadmap, a list of AWS projects that translate well into a portfolio, an overview of DevOps fundamentals, and a clear breakdown of Kubernetes vs Docker.
If you want AI-specific project ideas that force you to work through data pipelines and deployment constraints, machine learning projects is a good map, and this guide to data engineering helps explain the part of AI work that rarely goes viral but often decides whether a system works.
Takeaways
Amazon is betting that AI becomes a persistent layer of the economy, that cloud delivery remains the default for most organizations, and that the providers who can supply compute without friction will own the pricing power. The plan is coherent.
The open question is execution. Amazon does not need AI to be real, it already is. Amazon needs AI demand to be big enough, soon enough, and sticky enough to justify the scale of what it is building. In 2026, the market will not be grading the vision. It will be grading whether the rooms are filling.
