measuredlocally.com

M5 Ultra Mac Studio wait times: 16 to 18 weeks for the full chip

On 11 September Apple's own store was quoting 16 to 18 weeks for any Mac Studio with the full M5 Ultra chip, which meant January. The 512GB option, the one Mac that could hold a four bit build of DeepSeek V4.1 Flash, could not be ordered at all.

The video: DeepSeek V4.1 Flash on 4 DGX Sparks, and the M5 Ultra WaitPlays here, from YouTube
This story starts 3 minutes 10 seconds into my video of 13 September 2026, DeepSeek V4.1 Flash on 4 DGX Sparks, and the M5 Ultra Wait

Apple announced the M5 Ultra Mac Studio on 25 August and said it would be available from 22 September. On the evening of Friday 11 September, Apple's own store was quoting 16 to 18 weeks for any Mac Studio with the full M5 Ultra chip, in the UK and the US. By my own maths that meant 1 to 15 January 2027 (11 September plus 16 to 18 weeks).

The full chip is the one with a 36-core CPU and an 80-core GPU, and the quote was the same with 96GB and with 256GB of memory. At the end of August, AppleInsider had reported a 10 to 12 week wait for a fully specified Mac Studio.

The smaller M5 Ultra

The version with the smaller M5 Ultra chip (a 30-core CPU and a 64-core GPU) and 256GB of memory was quoting 6 to 7 weeks in the US that day, and 28 October to 4 November in the UK. Apple lists the same memory bandwidth for both chips, 1.2TB a second, which is more than four times a DGX Spark (NVIDIA lists 273GB a second, and 1,200 divided by 273 is 4.4).

The 512GB option

The 512GB option could not be ordered at all. Apple said it was coming in late October and had not given a price.

That is the one Mac that could hold a four bit build of DeepSeek V4.1 Flash. One uploader's card for such a build lists 427.6GB on disk and 427GB in memory, which it calls tight on a 512GB Mac. Another uploader reports 14.6 tokens a second on the older M3 Ultra with that much memory. Their card gives that speed at 1,000 tokens and does not define what that means.

Two corrections

I also have two corrections to the video of the week before. The report that OpenAI bought tens of thousands of Mac minis and Mac Studios came from The Information, not MacRumors as my description said. TechRepublic credited The Information when it covered the report on 31 August, and the MacRumors piece I had linked does not mention OpenAI. Neither Apple nor OpenAI had commented on the report.

NVIDIA's README for its PAIR router says it sends each request to one machine as a whole and does not split one model across machines. So it will not combine their memory the way I suggested.

Read more

The numbers in this story

FigureWhat it isFromKind
16 to 18 weeks, 1 to 15 January 2027Apple's quote for the full M5 Ultra chip, in the UK and the US, and the dates that givesapple.comstore price
10 to 12 weeksThe wait AppleInsider reported for a fully specified Mac Studioappleinsider.comreported
6 to 7 weeksThe smaller M5 Ultra chip with 256GB, in the USapple.comApple's store
1.2TB/s against 273GB/sMemory bandwidth of the M5 Ultra against a DGX Sparkapple.comthe maker's claim
about 427GBA four bit build of DeepSeek V4.1 Flash in a Mac's memoryhuggingface.coan owner's report
14.6 tok/sDeepSeek V4.1 Flash on the older M3 Ultra with 512GBhuggingface.coan owner's report

If a number on this page is wrong, I would rather know. Tell me in the comments under the video and I will correct it here and pin the correction there. What has been corrected so far is on the corrections page.