Sometime in the first week of March, Apple's 512GB Mac Studio disappeared from its online store. No press release. No product page update. The $9,499 configuration — the ceiling of what Apple offered for unified-memory compute — was simply gone, replaced by a $400 price hike on the 256GB model, now listed at $2,000.

The most likely explanation is one Apple hasn't offered: conventional DRAM has become expensive and hard to get. Samsung, SK Hynix, and Micron have been shifting production capacity toward high-bandwidth memory — the specialized chip stacks that power AI data center accelerators like Nvidia's H200 — for years, and the resulting squeeze has been working its way down the supply chain. Tim Cook acknowledged the pressure on the company's most recent earnings call, warning that memory costs could begin compressing margins later in 2026. For a company that has historically used its purchasing scale as a near-total buffer against component volatility, that's a concession worth noting.

The loss lands hardest in a specific corner of the hardware market. Apple's unified memory architecture exposes all system RAM as GPU-accessible memory, which made the 512GB Mac Studio something of an anomaly: a consumer-purchasable machine capable of running frontier-scale language models locally, without cloud infrastructure, at costs that beat most comparable setups. The ceiling on discrete GPU VRAM — 24GB for a standard RTX 4090, around 80GB for the most expensive professional options — means there's no obvious PC-based replacement. Apple's answer in macOS Tahoe 26.2 is a memory-pooling feature that allows multiple Thunderbolt 5 Macs to share compute resources, but reaching 512GB that way means buying two Mac Studios instead of one.

Exactly when the 512GB configuration disappeared is worth noting, because Apple hasn't said. Third-party tracking puts the removal window somewhere between March 4 and 6, but that's an approximation — no official statement has followed. The absence of explanation, for one of the most capable personal compute configurations available anywhere, is its own kind of signal.