A Raspberry Pi 5 with 16GB of RAM cost $120 in November 2025. It costs $205 today. That single data point tells you more about the state of technology in 2026 than every AI keynote combined.
The memory shortage everyone is writing about is real, it is severe, and it will not end soon. But the press coverage treats it as a supply chain story. It is not. It is a resource allocation story, and the allocation has already been decided. The three biggest memory manufacturers have pivoted their limited cleanroom space toward higher-margin enterprise components. Every wafer allocated to an HBM stack for an Nvidia GPU is a wafer denied to the LPDDR5X module of a mid-range smartphone or the SSD of a consumer laptop. The interesting question is not "why are prices going up?" It is: what do builders do now?
Who Gets the Wafers
Samsung and SK Hynix confirmed that OpenAI's anticipated demand could grow to 900,000 DRAM wafers monthly, which may represent around 40% of total DRAM output. Let that arithmetic settle. One project. Nearly half the world's memory production. Micron is guiding for a 68% gross margin in Q2 2026. When your biggest customers are hyperscalers willing to pay whatever it takes, you do not allocate scarce fab capacity to a $70 phone or a $35 single-board computer. As Counterpoint's Tarun Pathak put it, memory companies are asking smartphone vendors to "stand in line behind the hyperscalers."
The numbers you have already seen in the shared facts tell the macro story: DRAM prices doubling quarter over quarter, smartphone shipments collapsing, PC prices up 20%. I will not rehash them. What I want to talk about is the people those numbers land on.
I build things on Raspberry Pis. I have a home automation cluster, a PiHole DNS server, a little project that scrapes weather data. Millions of tinkerers, educators, and indie hardware hackers do the same. The Raspberry Pi 5 16GB nearly doubled in price from $120 in November to $205, while Orange Pi 5B 16GB surged from $160 to $312. These increases hit directly at the profiles that have sustained the Pi ecosystem for years: education, home robotics, home automation, DIY labs, and small industrial deployments. A hobbyist who bought three boards to experiment now buys one. A school that ordered a classroom set now orders half.
Companies like Apple with lots of buying power have been able to limit the price increases, but smaller outfits like Valve and Raspberry Pi haven't been so lucky. This is the part that matters to me as a builder. The shortage is not evenly distributed. It is regressive. The companies with the deepest pockets and longest supply contracts absorb it. Everyone else gets squeezed.
Constraint Breeds Craft
Here is what it actually looks like in production: Framework, the repairable laptop company, now publishes monthly blog posts about memory pricing. Their DDR5 pricing ranges from $12 to $16 per GB, and they're selling modules as close to weighted average cost as they can. When retail options are cheaper, they literally tell you to buy elsewhere. That is a company responding to a crisis by being radically transparent. I shipped a project on a Framework Laptop 13 last year; their ethos is the reason I trust them with my money.
Raspberry Pi took a different approach. A new iteration of the Raspberry Pi 4 moved from a single memory module to a dual-module configuration. "As you can generally buy smaller RAM more easily than larger RAM, we can use a pair of back-to-back modules instead of a single larger one," CEO Eben Upton explained. Clever. Not flashy. Exactly the kind of engineering pragmatism that gets overlooked when the headlines are about trillion-dollar AI infrastructure deals.
Audrey will write about the global equity implications, and she will raise points worth thinking about, particularly around the sub-$100 smartphone disappearing from markets in Africa and South Asia. She is right that a structural reallocation of silicon toward AI infrastructure has real human costs. But I also see something she might not: constraint is one of the best things that can happen to engineering culture. When memory was cheap and abundant, software bloated to fill it. The median Electron app consumes 500MB of RAM to render what could be a 2MB webpage. We got lazy.
The Raspberry Pi community is already running OS with zram and zswap to extend available RAM on memory-limited models. Developers are rediscovering memory-efficient patterns. I am genuinely not sure whether the shortage will force a meaningful shift in how we write software, or whether it is just too brief to change habits that took a decade to form. But the opportunity is there. The best code I have ever written was for constrained environments.
The memory crisis is painful. It is also, for builders, clarifying. When every gigabyte costs real money again, you start caring about how you use it. Intel's CEO says there will likely be "no relief until 2028." That is not a quarter. That is a generation of hardware products. The builders who adapt to scarcity now, who write leaner software and design around constraint, will ship better things on the other side. The Raspberry Pi at $205 is not just a price tag. It is a design brief.
