Starting in late 2025, the world began experiencing a big shortage of memory chips used in the manufacture of smartphones, computers, and other consumer electronics. The shortage has been caused by chip makers across the industry deciding to manufacture more lucrative chips for AI data centers. As an example, during the last year, we saw Micron, Samsung, and SK Hynix stop making RAM for consumer devices in favor of AI chips.
Random access memory, or RAM, is a crucial component in devices like smartphones, computers, and game consoles. RAM chips are what allow a computer to perform functions like keeping multiple tabs open in a browser,
In the fourth quarter of last year, the demand for RAM chips exceeded supply by 10%, and the shortage is quickly growing. By the end of 2025, the price for RAM increased by 50%, and the supply chain delays to get chips suddenly slowed to a crawl. If an electronics factory wants chips sooner, they’re being forced to pay a premium price and pre-pay for a large supply. The shortage is expected to last at least into 2027. A few companies, like ChangXin Memory Technology and Yangtze Memory Technologies Corp. have stepped up to enter the consumer RAM market. There are predictions that RAM prices will increase at least 60% this year, with specialty chips possibly doubling or tripling in price.
This is bad news for the broadband industry since the price of computers and smartphones will climb, likely out of the reach of the budgets of many households. This is going to increase the cost of all of the network electronics used for fiber, cable HFC, and wireless networks.
This is bad news for the nonprofits that have been refurbishing used computers and smartphones. One important part of many upgrades is to increase RAM capacity for old computers to be able handle new web needs. If RAM prices double, these entities will not be able to help nearly as many people. The problem will be made worse since small buyers of RAM will probably be the ones seeing the biggest price increases.
Digitunity recently published an article that estimates that 32.9 million people can’t access broadband from due to the lack of a computer. That’s about 10% of households, a number that compares with other estimates of the homes with broadband.
More expensive computers will hurt broadband adoption, and that hurts the public and the economy. People are increasingly reliant on access to broadband. The federal government, and many state and local governments, are eliminating the ability to communicate with the government by anything other than web portals. Federal services of all sorts, like veterans benefits and Social Security, are moving online.
The IRS and many states expect taxpayers to file tax returns using online software. This software is difficult to navigate with a smartphone, as are many other government portals. The IRS and other federal agencies will also no longer issue paper checks, forcing people to have an electronic way to receive and access payments from the government.
FEMA announced last year that anybody affected by a disaster must make a claim online, which is a particularly ironic requirement for folks who might have just lost a home due to a flood, tornado, or hurricane. For anybody who has ever dealt with a disaster result, there is a mountain of communication needed to push a claim through to the finish line.
People in rural areas increasingly need to use telemedicine as rural hospitals and clinics continue to fail and close.
A computer at home is vital for working from home or taking college and other classes online. These are also tasks that can’t easily be done by smartphone.
I would like to add to this, as I don’t expect it to continue until 2027. A recent announcement from Google has them releasing a new AI model called TurboQuant that happens to reduce the ram needed by an exceptionally large 6 times the previous amount . This news instantly drop stock prices for the companies like Micron and the rest that have invested heavily in AI to drop dramatically. While ram prices were still rising, they have already seem to have peaked and some have already started the decline in price, though I expect this to still take more time to get back to normal, the process has already begun. I agree with everything else that was spoken about in this Article. Thank you.
I disagree. turboquant lets you run a higher quant model on a lower spec machine and get more context window. You’ll still want to fill up your available resources, just get a better experience. For instance, you can’t run qwen3.5:35b on a mac with 16GB of ram, you could do it in 24GB. It’s a modest improvement but it’s not changing the ram shortage in any way.
Only increased production capacity will dig us out of this.
I think what has plateaued pricing is that AI systems are needing faster and faster memory. So it’s a balance of keeping or increasing ‘consumer DDR’ production vs those same fabs building GDDR/HBM. If you ran a fab, which is the more lucrative market.
Take apple’s ddr5 and high memory bandwidth in it’s m5 lineup. The ‘TOPs’ of those chips are 1/10th of an nvidia 5070 ti for instance and they are a good balance of ram speed vs ai performance. If they trippled the AI capabilityies on the CPU they wouldn’t get much more out of a larger model because of ram speed. Move into nvidia’s space and DDR5 is the bottleneck. Also note that this is why it’s completely pointless to run a large model on a mixed memory system of nvidia 5070ti w/ 16GB and 128GB of main memory. The DDR5 in your system is just way too slow, not to mention PCIe etc.
We have utilized refurbished business machines for customers in the past, but the cost of DDR4 memory has skyrocketed which makes the refurbs a poor value.