AI bubble forces PINE64 to pause open hardware production

Component lead times for the RK3588 jumped from 12 to 52 weeks after every hyperscaler started hoarding wafers for AI accelerators. That's the real story behind PINE64's announcement — not some philosophical stance on "bubbles," but raw economics. When NVIDIA and AMD pay 3x spot prices for 5nm/7nm allocation, foundries bump everyone else. A small-batch ARM board maker running 1,000-unit production runs simply gets priced out.

What actually stopped

PINE64's current lineup — Quartz64, PineNote, PinePhone Pro — all sit on Rockchip SoCs fabbed at TSMC 8nm or Samsung 8nm. Those same nodes now serve H100/A100 support chips, networking ASICs, and the flood of "AI PC" processors hitting the market. The PineBook Pro refresh (RK3588-based) was supposed to ship Q2. The Quartz64 Model B revision had a production slot booked for March. Both gone.

Lukasz Erecinski posted the notice on their forum without marketing fluff: "We cannot secure allocation at any reasonable price. We'll resume when the market corrects." No timeline. No pre-order bridge. Just a hard stop.

Why this hits different than 2021 shortages

The 2020-2022 crunch was broad — substrates, passives, legacy nodes. This one is targeted. 28nm and 40nm (where older Allwinner and A64 chips live) are relatively free. But the performance tier that makes a usable Linux laptop or tablet in 2024? 6nm-12nm. That's exactly where AI inference accelerators and NPU-heavy mobile SoCs fight for wafers.

Rockchip's RK3588 has an NPU rated at 6 TOPS. Marketing loves that number. So do procurement teams at Chinese EV makers and surveillance vendors who now order 500k units at a time. PINE64's 5k-unit orders don't register.

The open hardware trap

This exposes a structural problem nobody talks about: open-source hardware depends on commercial silicon availability. RISC-V boards like the Star64 (JH7110, 28nm) still ship because that node isn't "strategic." But anything with modern GPU/NPU/video encode — the bits that make a device daily-drivable — sits on contested nodes.

PINE64 could pivot to older SoCs. The A64 (40nm) still runs mainline Linux fine. But users expect 4K60 decode, Vulkan 1.2, and 8GB RAM in 2024. Dropping to 2016-era performance breaks the implicit contract with their community.

What happens next

Two scenarios. Most likely: Q4 2025, when Blackwell/Rubin ramp peaks and H100 demand plateaus, 7nm/8nm allocation loosens. PINE64 restarts with revised BOMs, maybe slightly higher prices. The community grumbles but returns.

Less likely but possible: the "AI PC" push (Meteor Lake, Ryzen AI, Snapdragon X Elite) consumes all leading-edge client SoC capacity through 2026. Then open hardware stays dead until someone funds a dedicated run — unlikely at PINE64's scale.

Either way, the Quartz64 Model B sitting in my drawer just became a collector's item. Not by design. Just physics and purchase orders.