Geerling argues that Samsung, SK Hynix, and Micron are prioritizing fab capacity for high-margin HBM3E and LPDDR5X, leaving commodity LPDDR4/LPDDR4X at the bottom of the priority list. He documents a 40-60% price increase since early 2024, which flows directly to SBC sticker prices since DRAM is a major BOM component.
The editorial highlights that when an 8GB SBC costs $80-100 while a used Lenovo ThinkCentre M720q with 16GB DDR4 and NVMe goes for $90 on eBay, the value proposition inverts completely. The x86 mini PC delivers more RAM, faster storage, better single-thread performance, and broader compatibility, undermining the SBC's traditional price-driven appeal.
Geerling's analysis implicitly argues that affordable SBCs created a generation of developers comfortable with ARM, Linux, and bare-metal deployment. The Raspberry Pi's genius was never the hardware but the $35 accessibility for students, hobbyists, and homelab builders — a threshold now being eroded by forces outside the SBC ecosystem's control.
The editorial emphasizes that the SBC market has always been price-sensitive and that the original Raspberry Pi's $35 price point made physical computing accessible to students, hobbyists, and developers. The Pi 5's jump to $60-80 already strained this, and further DRAM-driven increases threaten the ecosystem's foundation.
The editorial notes that Pine64, Orange Pi, and Radxa face identical DRAM supply constraints but with even less purchasing power than the Raspberry Pi Foundation. This suggests the hobbyist SBC market's pain is not limited to one vendor but is a structural problem affecting the entire ecosystem disproportionately compared to larger electronics buyers.
Jeff Geerling — the closest thing the single-board computer world has to a consumer advocate — published an analysis arguing that DRAM pricing has effectively broken the hobbyist SBC market. The post hit 463 points on Hacker News, resonating with a community that has watched their favorite $35 boards slowly become $80 boards over the past three years.
The core argument is straightforward: DRAM manufacturers — Samsung, SK Hynix, and Micron — are allocating fab capacity toward high-margin products like HBM3E (used in NVIDIA's H100/B200 GPUs) and LPDDR5X (used in flagship phones and AI-capable laptops). The DRAM that SBCs need — commodity LPDDR4 and LPDDR4X in small packages — sits at the bottom of every fab's priority list, and prices have risen 40-60% since early 2024.
For a Raspberry Pi or similar board where the SoC and RAM account for the majority of the BOM cost, that price increase flows directly to the sticker price. The Raspberry Pi 5, which launched at $60 for the 4GB model and $80 for 8GB, already represented a significant step up from the Pi 4's original $35/$45 pricing. Other SBC makers — Pine64, Orange Pi, Radxa — face identical supply dynamics with even less purchasing leverage than the Raspberry Pi Foundation.
The SBC market has always been a price-sensitive ecosystem. The original Raspberry Pi's genius wasn't the hardware — it was the $35 price point that made physical computing accessible to students, hobbyists, and developers who wanted a dedicated Linux box for homelab projects, IoT prototypes, or Kubernetes learning clusters. That price point created an entire generation of developers comfortable with ARM, Linux, and bare-metal deployment.
When an 8GB SBC costs $80-100 and a used Lenovo ThinkCentre M720q with 16GB DDR4 and an NVMe slot goes for $90 on eBay, the value proposition inverts completely. The x86 mini PC gives you more RAM, faster storage, better single-thread performance, and compatibility with every Docker image ever built. The SBC's advantages — GPIO pins, low power draw, compact form factor — only matter for a subset of use cases.
The Hacker News discussion surfaced a recurring theme: hobbyists are already making this switch. Multiple commenters described replacing Pi clusters with used mini PCs or Intel NUCs, getting better performance at comparable or lower cost. The homelab subreddit has seen a similar shift, with "used mini PC" posts increasingly displacing Raspberry Pi build guides.
This isn't just a hobbyist problem. The industrial SBC market — think factory automation, digital signage, edge computing — is also feeling the squeeze. Companies that designed products around $15-20 compute modules are now facing $25-35 module costs, enough to blow margins on cost-sensitive deployments. Some are redesigning around RISC-V SoCs with integrated SRAM for the simplest workloads, but that's a multi-year re-architecture.
To understand why this isn't a temporary blip, follow the fab economics. A single HBM3E stack (used in AI accelerators) consumes roughly 3x the DRAM die area of a conventional DDR5 module, while commanding 5-10x the price per bit. When NVIDIA, AMD, and hyperscalers are placing orders for tens of millions of HBM units, the rational move for Samsung and SK Hynix is to convert every available wafer to HBM production.
TrendForce data from late 2025 showed HBM already consuming over 10% of total DRAM bit supply, projected to reach 15% by the end of 2026. That might sound small, but DRAM supply grows slowly — total bit output increases maybe 15-20% per year. When HBM absorbs most of that growth (and then some), everything else gets squeezed: DDR5 for PCs, LPDDR5X for phones, and especially the low-density LPDDR4 modules that SBCs use.
The situation has an uncomfortable parallel to the GPU crisis of 2021-2022, when crypto mining demand priced gamers out of graphics cards. The difference: GPU supply eventually recovered when crypto mining became unprofitable. AI demand for HBM shows no sign of a similar demand shock. If anything, the trajectory is steepening as more companies build training and inference clusters.
If you're running homelab infrastructure on Raspberry Pis or similar SBCs, the economics now favor a different approach:
For new projects: Used enterprise mini PCs (Dell OptiPlex Micro, Lenovo ThinkCentre Tiny, HP ProDesk Mini) offer dramatically better price-per-performance. They're x86-native, take standard DDR4 SO-DIMMs (still cheap on the secondary market), and have NVMe slots. The power delta versus a Pi — maybe 15-20W idle versus 5W — costs roughly $10-15/year in electricity.
For existing Pi clusters: Don't panic-replace working hardware. But when boards fail or you need to expand, price out mini PCs before defaulting to another Pi. A three-node K3s cluster on used ThinkCentres with 16GB each will outperform a five-node Pi cluster with 8GB each — at roughly the same total cost.
For IoT and embedded: This is where SBCs still win. If you need GPIO, low power, or a specific form factor, there's no x86 substitute. But consider whether you actually need Linux and 4GB of RAM, or whether an ESP32 or RP2040 microcontroller handles the workload at 1/10th the cost.
For edge AI inference: Ironically, the same DRAM dynamics pushing SBC prices up are also pushing the industry toward more efficient on-device inference. Models like TinyLlama and Phi-3-mini that run in 2-4GB of RAM exist precisely because not every edge device can afford 16GB. Constraint breeds optimization.
The hobbyist SBC golden age — when $35 bought a capable Linux computer — is probably over, at least until DRAM supply catches up with AI demand or a new memory technology disrupts the economics. The Raspberry Pi Foundation has enough brand loyalty and educational mission to survive at higher price points, but the broader SBC ecosystem of clones and alternatives will thin out. For developers, the practical takeaway is simple: stop thinking of SBCs as the default for homelab and learning projects. They're now a specialized tool for specialized problems — which, honestly, is what they probably should have been all along.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.