How AI Is Taking Memory Away From Budget Smartphones: The DRAM Shortage and the End of the Cheap Era

Over forty years, the cost of computing power has fallen by roughly a trillion percent. In 1985, the best personal computer an average-income American could afford, the IBM PC AT, cost $19,400 in today's money. Today a basic Tecno Spark Go, sold at a street market in Nairobi, costs just $30 and contains a processor billions of times more powerful than that computer. No other product in human history has seen such a steep price drop alongside rising quality. That cheapness changed the world, giving billions of people education and access to information. But the age of affordability is ending, and one unexpected thing is to blame: memory.

When DRAM became gold

The memory in question is DRAM, the working memory every electronic device needs to function. Ten years ago, memory was a very cheap commodity, produced in huge quantities and seemingly as limitless as water or air. Manufacturers invested in new fabs, and prices fell every year. But over the past five years the picture has changed radically.

The reason is simple: data centers training AI models such as GPT-4 and Claude have begun absorbing memory in quantities nobody foresaw even five years ago. Training a single modern large AI model requires hundreds of petabytes of working memory running at maximum speed simultaneously. Training has to be fast and continuous, because every hour of downtime costs companies millions of dollars. That means data centers demand the fastest, most reliable and most expensive memory on the market.

The leading DRAM makers, Samsung, SK Hynix and Micron, have redirected almost all their production to this new task. They have signed long-term contracts worth colossal sums with OpenAI, Google DeepMind, Meta*, and Microsoft. Contract prices are above market, and supply is guaranteed. Cheap smartphones were left at the back of the queue, with no access to the product.

No memory even for flagships

Memory allocation in the market has become harsh and unsentimental:

• AI company data centers are the first and biggest buyers, taking 40-50% of all memory produced worldwide
• Flagship smartphones (Apple, Samsung Galaxy S25) fight noisily over the remaining 20-30%
• Mid-range smartphones (1000-3000 rubles) are already feeling the shortage and rising prices
• Budget phones ($20-100) are almost entirely shut out of competitive memory prices

DRAM prices rose 40-45% in the first half of 2026. For a smartphone with a 5-10% profit margin, that means a loss on every device sold. Budget phone makers are closing production lines and leaving emerging markets. MediaTek, the company that supplied processors for half of the world's cheap phones, has officially announced reduced investment in the budget segment.

"We can't compete with data centers for memory.

They pay three times more than we can offer," a manager at one of the largest memory makers told the Financial Times.

The end of a forty-year run

For forty years, electronics got cheaper and more accessible to every social stratum on Earth. In 2005, a smartphone cost $500 and was a toy for the rich. By 2015, decent smartphones for $100 appeared. By 2020, a perfectly respectable smartphone could be bought for $30-50. Every year, hundreds of millions of people in India, Africa and Southeast Asia touched a smartphone for the first time thanks to that price.

These devices did more than entertain. They opened access to schooling, telemedicine and financial services that had previously been out of reach. A cheap smartphone changed lives more than any premium flagship for the wealthy. It was a revolution in accessibility that rewrote human history.

Now the paradigm is running in reverse. The technological progress that brought affordability is starting to destroy it. Expensive computing for AI training is winning the fight for materials, labor, energy and chips.

What this means

The world risks splitting in two: those with access to advanced computing and AI technology, and everyone else, stuck on old hardware. The AI revolution demands sacrifices, and one of the first may be digital accessibility for billions of people in developing countries.

A strange paradox: AI, promised as a tool for everyone, is first monopolizing the most valuable production resources.

*Meta has been designated an extremist organization and banned in Russia.