Tech • IA • Crypto
AI is quietly extorting you (and no one is talking about it)
8/10TL;DR
Rising demand for AI chips is driving a global memory and packaging bottleneck, quietly increasing consumer electronics prices through what some call an “AI tax.”
KEY POINTS
A hidden cost surge in everyday tech
Prices of standard components like RAM surged by up to 90% in early 2026, contributing to a nearly 20% increase in PC prices from manufacturers such as Dell, HP, and Lenovo. In smartphones, the effect is subtler: devices at the same price point now ship with reduced specifications, including less memory and downgraded features.
The emergence of an “AI tax”
Consumers are indirectly funding the AI boom without purchasing AI services themselves. The term “AI tax” has emerged in the United States to describe how supply constraints in key components ripple through the broader electronics market, raising costs across unrelated products.
The real bottleneck is not computing power
Contrary to popular belief, the limiting factor in AI hardware is not the GPU or compute unit. In 2025, major players including Nvidia, AMD, Google, and Amazon used only about 12% of global logic chip capacity, while consuming over 90% of advanced packaging and HBM memory supply.
Memory dominates chip costs
High-bandwidth memory (HBM) has become the most expensive component in AI chips, accounting for over 60% of manufacturing costs in 2025, up from roughly half the year before. For example, in Nvidia’s B200 GPU, memory and packaging together represent nearly two-thirds of production costs.
A supply chain controlled by few players
The global HBM market is dominated by just three companies: SK hynix, Samsung, and Micron. Their production capacity is already fully booked through 2026 and largely into 2027, with industry leaders warning shortages could persist until 2030.
Trade-offs reduce consumer memory supply
Producing HBM consumes significantly more resources than standard memory. Manufacturers shifting toward higher-margin AI memory effectively reduce output of conventional RAM, tightening supply for PCs, phones, and consoles.
Packaging: a critical and overlooked chokepoint
Advanced chip packaging, particularly CoWoS technology, is essential to connect processors and memory at high speed. This segment is dominated by TSMC, which controls most of the high-end capacity globally. Demand far exceeds supply despite rapid expansion.
TSMC’s constrained expansion
TSMC aims to scale advanced packaging capacity dramatically by 2026, yet Nvidia alone has reportedly secured around 60% of future output. This concentration reinforces supply bottlenecks across the industry.
A deeper dependency on a single supplier
Beneath chip production lies an even tighter constraint: lithography machines from ASML. These systems, costing up to €350 million per unit, are essential for manufacturing cutting-edge chips and are produced by a single company after decades of R&D.
The “memory wall” and physics limits
AI systems face a fundamental challenge known as the memory wall: compute units require massive, fast data flows to operate efficiently. Technologies like stacked HBM attempt to solve this, but manufacturing complexity and yield issues severely limit scalability.
Even tech giants cannot escape the bottleneck
Companies like Google, despite designing custom chips such as TPUs, remain dependent on the same constrained ecosystem: HBM suppliers, TSMC packaging, and ASML lithography tools.
CONCLUSION
The global AI race is increasingly constrained not by processing power but by scarce memory, packaging capacity, and specialized manufacturing tools, concentrating power among a handful of suppliers and quietly raising costs across the entire technology market.