Andy Jassy's shareholder letter makes the case for Amazon's AI bet and $200B CapEx cycle

Amazon's AI Bet Requires Accepting a $200B CapEx Spike—And Jassy Is Making the Case for It

Andy Jassy's latest shareholder letter amounts to a defense of Amazon's willingness to sacrifice near-term free cash flow to dominate the AI infrastructure race. The letter resets the narrative around AWS's capital spending by drawing a parallel to the company's original cloud buildout—a bet that looked questionable internally until it proved transformational.

The core argument is structural: Amazon is in the middle of a "disproportionate inflection" in AI, and the company should invest as aggressively as possible, even if it means depressed free cash flow in the near term. Jassy frames this not as reckless but as consistent with Amazon's long-term playbook.

The Scale of the Bet

AWS added 3.9 gigawatts of new power capacity in 2025 and expects to double total power capacity by 2027. The company is monetizing that capacity as it installs it—Q4 2025 saw 24% year-over-year AWS growth and a $142 billion revenue run rate. Yet Jassy emphasizes that capacity remains the binding constraint: two large customers have already asked if they could buy all of AWS's Graviton instance capacity for 2026, a request the company declined out of fairness to other customers.

This is the demand story. But demand alone doesn't justify the spending. Jassy argues AWS's AI revenue run rate is over $15 billion in 2026, nearly 260 times larger than AWS was at the same three-year mark after its 2006 launch (when it was running at $58 million). That acceleration—driven partly by AI customers choosing AWS across model building, inference, agent building, and secure environments—sets up the financial justification.

The Cash Flow Lag Problem

The tension is timing. AWS must lay out cash upfront for land, power, buildings, chips, servers, and networking before it can bill customers. Depending on the component, that lag runs six to twenty-four months. Data centers have a thirty-plus-year useful life; chips, servers, and networking gear have five to six year lifecycles.

In periods of high growth where CapEx growth outpaces revenue growth—which is happening now—free cash flow gets pressured early on. Jassy acknowledges this directly. But he argues the return on invested capital compounds attractively once those assets are monetized, and Amazon has already lived through this cycle during the initial AWS buildout.

The precedent matters. Amazon's 1997 shareholder letter, which Jassy includes in his current missive, shows the company has always been willing to subordinate short-term profit to long-term position. The parallel is intentional: if AWS's trajectory validates that patience, the current AI CapEx cycle should too.

Custom Silicon as a Moat

Jassy also emphasizes Amazon's chip advantage. Graviton, launched in 2018, delivers 40% better price performance than Intel for CPU workloads. AWS also makes Trainium (for training) and Nitro (inference acceleration). The letter notes that virtually all workloads ran on Intel until Graviton arrived, and custom silicon is now becoming the competitive lever in AI infrastructure.

Two takeaways emerge from the letter. First, Jassy is signaling to investors that the capacity crunch is real and demand-driven, not a supply management error. Second, he's anchoring the free cash flow conversation to a precedent—AWS's first buildout—that eventually vindicated aggressive CapEx spending. The implicit argument is that shareholders who trusted the first wave should trust the second.

Whether that argument holds depends on whether AWS's AI revenue actually scales as fast as Jassy projects. The letter doesn't provide a detailed financial roadmap beyond the $15 billion run rate figure. But the shape of the case is clear: Amazon is asking for patience in exchange for optionality on an AI infrastructure market that may, like electricity a century ago, reorganize every industry on Earth.