Mac mini AI chip gets faster as memory crunch drives price to $899

The Mac mini's bargain era is over — and AI is the reason

For two years, the Mac mini was the single best deal in consumer AI hardware. A $599 box with Apple Silicon's unified memory architecture could run small and mid-size language models locally, sip power, and pay for itself against cloud API fees in months. That machine is gone. Apple raised the mini's effective starting price from $599 to $799 in May 2026 by discontinuing its entry-level configuration, and now the new M6-based mini starts at $899 — a $300 jump over the 2024 base model for the same 16GB of RAM and 256GB of storage151617.

The culprit is not Apple product strategy in isolation. It is the AI infrastructure buildout itself. The same memory and storage chips that feed hyperscale AI data centers are the components inside every Mac, and their prices have surged industry-wide11. Apple told The Wall Street Journal it had "never seen a component price increase this much, this quickly"11. The company that once positioned the mini as the cheapest on-ramp to the Mac ecosystem is now passing the cost of the AI gold rush directly to the customers who bought minis to run AI on their desks.

A supply crunch driven by AI demand

The timeline is instructive. In May 2026, Apple pulled the $599, 16GB/256GB configuration after it sold out at most retailers, moving the starting price to $799 while other configurations carried shipping delays of weeks or months12. CEO Tim Cook attributed the shortage to two forces: constrained availability of the advanced fabrication nodes that produce Apple's system-on-a-chip, and consumers "snapping up" Mac minis and Mac Studios specifically to run artificial intelligence — demand he admitted arrived faster than Apple predicted1214.

Then in June, Apple raised the M4 Pro mini's starting price by another $200, to $1,59911. By September, the entire M4 mini line was retired and replaced with the M6 model at $89916. In the UK, the jump was from £599 to £89916. Reviewers across the board treat these hikes as the story of the new machine: Ars Technica's headline calls the price increase "rough," Six Colors says the mini has "crossed an invisible threshold" away from impulse-buy territory, and MacRumors' review roundup concludes the new model is good but "no longer an insanely great deal"151718.

Where the coverage diverges is on the base config itself. Apple briefly discontinued the 16GB/256GB SKU in May, then reinstated it — but kept the $799 floor, which tells you the removal was a pricing move, not purely a supply one1110. Reviewers also disagree on how much of the increase is genuine component cost versus margin: one MacRumors commenter analysis circulating with the reviews claims Apple's per-unit profit on the base model rose from roughly $231 to $346, though that figure is a reader estimate rather than an audited number18. What the reporting agrees on is the cause — Apple itself says it is passing on skyrocketing RAM and storage prices tied to the AI boom1817.

The chip is genuinely good at AI

The irony is that the hardware keeps getting better at exactly the workload that made it expensive. The M6, debuting first in the humble mini, is described as the biggest leap for Apple's lower-end silicon since the M4 generation, with Apple quoting up to 4x faster AI performance, around 40% faster CPU, and the first dual Neural Engine in this tier1516. Reviewers uniformly describe the new mini as extremely capable for local AI work1817.

For context on why the mini became the default local-AI recommendation: the M4 base chip paired 16GB of unified memory with roughly 120 GB/s of bandwidth and a 65W power envelope, enough to run 7B–14B parameter models at conversational speed — around 30–50 tokens per second per most benchmarks286. Unified memory is the key differentiator. Unlike a PC with a discrete GPU capped at 24GB of VRAM, Apple Silicon lets models use the entire RAM pool, so a 64GB M4 Pro could run 70B-class models on hardware no consumer NVIDIA card could match43. Guides in the local-AI community consistently ranked the mini as the best-value inference machine precisely because of that memory ceiling — or lack of it310.

The 2026 config changes cut against that value proposition from the top as well. The 64GB M4 Pro build was discontinued, leaving 48GB as the maximum memory in the mini line — which narrows the gap between a mini and a modest GPU rig for large-model work10. Meanwhile the 16GB floor at $799 was widely judged inadequate for serious local inference, fine only for 3B–8B models10. In other words, the AI-workload sweet spot got more expensive and smaller at both ends.

The economics of local inference just shifted

The deeper significance for the AI inference hardware market is this: the mini's rise as a local-AI platform was built on a price thesis. At $599, the machine could plausibly replace $10–$30 per month of cloud API spending and recover its cost within months19. At $899, the payback period stretches by half, and the comparison against just paying OpenAI or Anthropic for inference gets tighter — especially since a $599 orchestration box running API-based assistants never needed local model capacity in the first place9.

There's also a second-order effect the reviews note almost in passing: resale value. Trade-in prices for old M4 minis have risen sharply — one user reported their $599 machine's trade-in value climbing from $290 to $480 — because scarcity has repriced used AI-capable hardware on the secondary market20. The Mac mini shortage made yesterday's bargain machine an appreciating asset, which is unheard of for consumer desktops.

My reading: this is less a story about Apple greed than about AI compute costs escaping the data center and landing on consumer price tags. Every GPU-accelerated server cluster being built this year is competing for the same DRAM and NAND supply as every laptop and mini PC, and Apple — which buys memory at enormous scale — is signaling that the inflation is structural, not cyclical1118. The company's statement that it must "begin raising prices" reads as a forecast, not an apology.

For local-AI builders, the practical advice from the guides converges: if you already own an M4 mini, keep it — it remains entirely adequate for 8B–14B model work620. If you need large-model local inference now, the M4 Pro 48GB at around $1,999 is the last big-memory mini available10. And if you're waiting for prices to fall, don't — the memory shortage has no reported end date, and every guide published since May treats further increases as the base case1416.

The Mac mini M6 is, by all accounts, a better AI computer than the M4 was. It is just no longer a cheap one. The machine that democratized local inference on a $599 budget is now a $899 product, and the industry dynamics that repriced it — AI data centers consuming the world's memory supply — are the same ones its buyers were trying to escape by running models locally. The bargain prices weren't taken away by Apple. They were outbid by the AI boom itself.