If you are building on AWS, you are likely paying for the infrastructure boom whether you realize it or not. At an internal staff meeting in September, AWS CFO John Felton told Prasad Kalyanaraman, the executive overseeing Amazon's AI infrastructure, "You've taken us free cash flow negative as a company, so congratulations." The remark, captured in a recording obtained by Business Insider, highlights a stark reversal for Amazon. Two years ago, Wall Street debated what to do with Amazon's cash hoard. Now, the company is burning through capital, spending $220 billion this year alone to build the physical backbone for AI.

The Capacity Crunch Extends to 2027

For developers, the most critical takeaway is not the financials, but the timeline. Kalyanaraman recalled telling CEO Andy Jassy in 2023 that AWS would "surely" be out of its capacity crunch by 2027. Instead, during the September meeting, he admitted that Amazon still expects demand to exceed its available capacity in 2027. This means that if you are planning large-scale AI deployments or expecting immediate access to new GPU clusters, you may face continued wait times. The company’s ability to meet developer demand is being throttled by physical infrastructure limits, not just software scaling.

Early Rack Receipt Delays Deployment

The logistical reality for developers is messy. Internal emails from August show Amazon preparing a new "Early Rack Receipt" process for server racks carrying Nvidia's GB300 chips. While this sounds like progress, it reveals a bottleneck: many AWS data center sites lacked the power capacity to turn these racks on. Consequently, Amazon is staging equipment in intermediary warehouses. Under this process, Amazon takes ownership of hardware earlier in the supply chain, tying up cash and exposing the company to delays if power permits slip. For a developer, this means the hardware you need is physically present but logically unavailable until the grid catches up.

Power Constraints Create Friction

The coordination challenges are severe and directly impact service availability. Kalyanaraman noted that the internal outlook for returns is volatile, stating, "The forecast is changing on a daily basis." At Amazon's scale, maintaining an eight-week supply buffer can mean holding billions of dollars in inventory. Furthermore, the company is facing new constraints like labor shortages in West Texas, where AWS is paying to construct housing for pipefitters and electricians. These are not abstract corporate issues; they are friction points that slow down capacity availability for users. Even trivial items like shipping crates have become material expenses, illustrating how the sheer volume of the AI buildout creates bottlenecks that delay the provisioning of new compute resources.

Key Takeaways

  • AWS capacity crunch is now expected to last until at least 2027, delaying access to new AI infrastructure.
  • "Early Rack Receipt" processes show Nvidia GB300 racks are being staged in warehouses due to power constraints at data centers.
  • Power permits and grid bottlenecks are the primary blockers preventing deployed hardware from becoming available to developers.
  • Amazon is building housing for skilled labor in West Texas to mitigate construction delays that impact cloud capacity.

The Bottom Line

Developers should stop treating AWS capacity as an infinite, instant resource and start planning for physical-world delays. The era of frictionless cloud scaling is over; now, grid power and labor supply are the true rate-limiting factors for AI deployment.