Spending on AI infrastructure continues to rise, with major tech companies expected to invest trillions of dollars over the next few years to satisfy users' cravings for ChatGPT and Claude. This was reported by Bloomberg.
The massive expenditures are driven not only by investments from hyperscalers like Microsoft and Meta in data centers but also by rising prices for the components needed to build these enormous computing facilities.
Investment by American hyperscalers could exceed $1 trillion by 2027. Source: Bloomberg.
“Chipflation” and Investments
“Chipflation” is a problem not just for the AI sector; shortages are also affecting traditional semiconductors. This leads to increased prices for smartphones, gaming consoles, and other electronics.
The popularity of applications for programming and utilizing AI agents is driving demand for hardware components that support workflows: GPUs, memory, and CPUs. These components, which previously played a secondary role in the AI revolution, are now essential for large language model workloads.
Some of the most profitable tech companies and well-funded startups are competing to secure enough hardware, fearing they will fall behind in the race for superintelligence.
The largest chip manufacturer, TSMC, plans to invest a record $56 billion. Elon Musk is considering building his own chip factory, which could cost between $55 billion and $119 billion.
This situation is boosting the stock prices of AI equipment suppliers, which are outpacing most of their customers.
Chip manufacturers are outpacing AI giants. Source: Bloomberg.
“Chip companies are thriving at the expense of everyone above them in the supply chain,” said James Covello, global head of equity research at Goldman Sachs.
Recent reports from hyperscalers indicate that inflation is hitting them hard. Microsoft expects that rising component prices will increase annual capital expenditures by $25 billion, bringing the total to a record $190 billion. Meta has raised its projected spending range by $10 billion.
“Memory Tax”
Tech corporations are increasingly facing a “memory tax,” as advanced AI accelerators require significantly more high-bandwidth storage.
The three largest DRAM suppliers—SK Hynix, Samsung Electronics, and Micron Technology—have become favorites on the stock market, with a combined market capitalization exceeding $2.8 trillion.
Demand from data centers has boosted SK Hynix's profitability. Source: Bloomberg.
According to SemiAnalysis, total spending on various types of memory is expected to reach 30% by 2026, up from 8% in 2024.
Large investments in chips are prompting hyperscalers to seek ways to reduce computing costs. One approach is to use alternative AI processors like AMD. Some companies are developing their own solutions—Alphabet's tensor processors, Amazon's Trainium chips, and Microsoft's Maia 200.
Other innovations include Google's TurboQuant compression technology, which helps reduce memory costs.
The Inflation Culprit
The excitement surrounding artificial intelligence is causing smartphone, gaming console, and PC manufacturers to struggle with securing memory chip supplies, as their producers prioritize more profitable markets and long-term contracts.
Consumer electronics manufacturers are left with three options: pass the rising costs onto consumers, downgrade device specifications, or accept reduced profits.
Building semiconductor factories takes years, so there is little prospect for a quick response to demand.
Given the rising energy costs driven by energy-intensive data centers, artificial intelligence will continue to significantly impact inflation for some time, Bloomberg concluded.
It is worth noting that Samsung's semiconductor division reported record profits in the first quarter, exceeding expert expectations.