Why Are Epic Games and Other Customers Leaving Amazon for Google?

For years, Amazon Web Services (AWS) has been the go-to utility for internet-scale computing—the place where businesses rent as much power as they need, on demand. But in the summer of 2025, that seemingly infinite supply hit hard limits.

Internal documents obtained by Business Insider reveal that AWS suffered "critical capacity constraints" for its flagship AI service, Bedrock. The shortage was so severe that it forced major customers like Epic Games and Thomson Reuters to defect to rival Google Cloud, costing Amazon tens of millions of dollars in lost revenue.

This is a sign that the AI boom is physically outstripping the ability of the world's largest cloud provider to build infrastructure. As Amazon races to spend $125 billion on capital expenditures in 2025, the question is no longer "is there demand for AI?" but rather "is there enough computer to run it?"

How much money did Amazon leave on the table?

The leaked memos highlight a cascade of missed opportunities. Bedrock, which lets companies access AI models like Anthropic's Claude, buckled under surging demand, with "quota" limits (caps on AI compute usage) proving too restrictive.

• Epic Games: The maker of Fortnite shifted a $10 million project to Google Cloud after AWS failed to provide enough "quota" (the limit on how much AI compute a customer can use).
• Vitol: The massive oil trader weighed moving projects away, risking a $3.5 million revenue hit due to "prolonged quota approvals."
• Atlassian & GovTech Singapore: These major clients were left waiting for quota increases, delaying at least $52.6 million in projected sales.

As the document bluntly stated, these constraints were "threatening customer adoption and potentially causing substantial revenue loss across multiple industries."

Why is Google winning?

The capacity crunch exposed a vulnerability that Google was all too happy to exploit. While AWS struggled to find GPUs, Google Cloud was offering "five to six times larger quota limits" and better performance.

Thomson Reuters, a key enterprise client, moved its "CoCounsel" AI product to Google after finding that AWS was 15% to 30% slower and lacked key government compliance certifications. Financial startup TainAI moved 40% of its workload to Google’s Gemini Flash model, saving $85,000 a day.

The internal AWS report admitted that Google’s models often outperformed Bedrock "across multiple benchmarks," and warned that AWS was "still missing an inspiring long-term vision and a holistic strategy" for AI inference.

Is the "AI Bubble" actually an "AI Shortage"?

This crisis flips the current narrative on its head. While Wall Street worries that Big Tech is overspending on an "AI bubble," Amazon's reality suggests the opposite: they aren't spending fast enough.

CEO Andy Jassy has been aggressively ramping up infrastructure, adding 3.8 gigawatts of power capacity in the last year alone—more than any rival. Yet, it wasn't enough to stop the bleeding. The fact that customers are fleeing because they can't buy the product suggests that demand is real, urgent, and currently uncapped by supply.

For Amazon, the lesson is expensive but clear: In the AI era, being the biggest cloud provider doesn't matter if you're the one telling customers "sold out.”

The Reference Shelf

• Amazon AI Capacity Crunch Pushed Customers to Rivals Like Google (Business Insider)