Amazon employees admit to using AI unnecessarily to pump up internal usage scores

Recent reports from the Financial Times reveal that Amazon employees have been manipulating AI tool usage to meet internal targets, echoing similar behaviors previously documented at Meta and Microsof... Read original →

Amazon’s 'tokenmaxxing' reveals a critical flaw in AI infrastructure investment logic. Artificial inference loads distort real demand signals for GPUs and HBM, risking NVIDIA-led overcapacity by 2027. Under the EU AI Act, such internal metric gaming could trigger compliance audits, forcing cloud providers to overhaul cost allocation models. Competitors like Microsoft and Google may seize this moment to champion 'effective AI utilization' as a new enterprise benchmark, gaining client trust. Over the next 12–24 months, the industry will pivot from volume obsession to efficiency—redirecting capex toward sparse inference, model distillation, and custom ASICs rather than generic GPU clusters. This performance-driven bubble is now accelerating a necessary recalibration across the AI supply chain.