A year ago, a 32-gigabyte kit of DDR4 memory, the kind that ships in millions of ordinary desktops and budget gaming PCs, cost about $55 to $70. Today the same kit costs $250 to $350, a 400 to 500 percent increase. DDR5 kits show the same pattern: $80 to $120 a year ago, $300 to $500 now. Prices were already three to four times their mid-2025 level by January 2026, and the second quarter of the year added another 89 percent on top of that, with analysts forecasting 40 to 50 percent more in the third quarter and 30 to 40 percent more in the fourth.
Nobody buying a laptop caused this. The AI industry did.
This is the sixth chapter in an argument this publication has been making about AI generally: that it is not a technological revolution but a regime of extraction, taking language, art, labor, and legitimacy from the many and converting it into power for the few. The first chapter traced that extraction across the whole economy. The second traced it through YouTube, the third through software engineers, the fourth through the attention economy, and the fifth through the pollution and water strain data centers leave behind in the communities that host them. This chapter is about a more diffuse extraction: the AI buildout’s raw material costs, quietly loaded onto the price tag of every laptop, every gaming PC, and every electric bill, for people who never opted into any of it.
Start with memory. Data centers now consume an estimated 70 percent of all memory chips produced worldwide, and IDC frames the resulting shortage as structural rather than cyclical: 2026 supply growth for both DRAM and NAND flash is running at 16 to 17 percent year over year, well below the 20 to 30 percent that would normally be expected. Samsung, SK Hynix, and Micron control more than 95 percent of global DRAM production, and all three have been reallocating factory capacity toward high bandwidth memory, the type AI accelerators need, because it earns an estimated three to five times the revenue per wafer of the standard memory that goes into consumer devices. Contract prices, the wholesale rates manufacturers negotiate with each other in bulk, rose 90 to 95 percent quarter over quarter in the first quarter of 2026 alone, with NAND flash contract prices rising 70 to 75 percent in the second quarter. Consumer SSD prices have roughly doubled since late 2025, with a 1-terabyte drive going from about $45 to about $90.
No company has said this more plainly than Micron. In December 2025, the company announced it was exiting the consumer memory and storage market entirely to prioritize AI data center customers instead. Micron CEO Sanjay Mehrotra was direct about why: “we see demand for memory in AI data centers remaining robust for years to come, and supply will struggle to keep pace.” That is a company choosing, in public, to walk away from ordinary buyers because AI customers pay more. Nobody had to strong-arm Micron into deprioritizing consumers. The margins did that on their own.
The effects are already showing up at checkout. Dell and Lenovo have said they will raise PC prices by as much as 15 percent, and IDC projects average PC prices climbing 4 to 8 percent across the industry in 2026. Where manufacturers are reluctant to raise sticker prices further, TrendForce senior vice president Avril Wu predicts a quieter version of the same squeeze: a $600 laptop in 2026 may look identical to last year’s model while shipping with 8 gigabytes of memory instead of 16. Some vendors have gone further still, selling prebuilt desktops with no memory installed at all, leaving the buyer to go source their own RAM in the same shortage that made the computer’s own manufacturer unable to.
The same extraction is happening to electricity, just with the bill split differently. In Richland Parish, Louisiana, Meta’s Hyperion data center campus needed so much new power that Entergy asked state regulators to fast-track seven to ten new gas-fired plants, a roughly $21 billion buildout, skipping the independent review that normally protects ratepayers from footing bills like this. Meta and Entergy say Meta will cover the construction costs, not ratepayers. The Union of Concerned Scientists and other consumer advocates warned that skipping the review is exactly what makes that promise unenforceable if Meta’s plans change.
Across the thirteen-state PJM grid, which includes Virginia’s “Data Center Alley,” data center demand alone drove 63 percent of the price increase in the region’s most recent capacity auction, which cleared at $329.17 per megawatt-day for the 2026 to 2027 delivery year, a 22 percent jump from the year before. That is not an abstract future cost. Starting in June 2026, ratepayers across the region began paying a collective $1.4 billion more in capacity costs alone, which works out to roughly $18 more a month for the average household in western Maryland and about $16 more a month in Ohio. In Ohio, a second charge stacks on top of that: AEP Ohio added a $7.90 monthly transmission charge for a typical household in April 2026, tied directly to its data center large-load tariff. The Natural Resources Defense Council separately projects the average household bill in the PJM footprint could run about $70 a month higher by 2028 than it would have been without the AI buildout, with cumulative ratepayer costs estimated between $100 billion and $163 billion by 2033.
The clearest version of who is actually paying for this comes from Maryland. The state’s Office of People’s Counsel found that Maryland homeowners will pay $1.6 billion over the next decade for transmission line upgrades that mostly serve data centers in Northern Virginia, not Maryland’s own buildout. Maryland has since filed a complaint with the Federal Energy Regulatory Commission, arguing that PJM is forcing its ratepayers to shoulder the cost of grid expansion built to serve someone else’s state. Virginia’s own regulators pushed back on their own rate case in July 2026, approving a smaller increase than Dominion Energy requested specifically to shift more of the cost onto data centers instead of residents, and Georgia’s Public Service Commission froze Georgia Power’s base rates through 2028 for the same stated reason. That two states pushing back counts as notable news says something about what the default has become everywhere else.
Sometimes the bill gets skipped entirely, at least until someone notices. In Fayette County, Georgia, a data center campus operated by Quality Technology Services, developed under the code name Project Excalibur, drained more than 29 million gallons of water through a connection that was never billed, one hookup installed without the county utility’s knowledge and a second that existed but was never linked to QTS’s billing account. The company had exceeded the peak-usage limit it agreed to during planning long before anyone at the utility noticed. It was residents complaining about low water pressure, not an audit, that caught it.
So yes, say this plainly too. It is economic colonialism when three companies that control the overwhelming majority of the world’s memory supply can decide, in public and without apology, that AI customers are worth more to them than the people who buy laptops, and walk away from the consumer market entirely because the margins told them to. It is the same colonialism when a data center’s power needs get fast-tracked past the review meant to protect the ratepayers next door, or when a regulator pushing back against that pattern is rare enough to be news. And it is the same colonialism, in its most literal form, when a company simply uses a public resource without paying for it until someone else’s water pressure drops low enough to ask why. None of the people paying more for a laptop, or a power bill, or a water bill asked for any of this. They are simply the ones left holding the cost of somebody else’s compute.