Product1 distinct publisher3 min readPublished
The 2025 configurations are gone from the store rather than fading out, which turns a refresh most shops would have taken at renewal into a purchase they have to justify on this quarter's numbers.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
Somewhere there is an IT standards page that reads "Mac Studio, M4 Max, 14-core CPU, up to 64GB," priced at $2,499 [8], and this week it describes a machine Apple will not sell [1]. Old configurations do not linger on the price list while a purchasing cycle finishes. The person who owns that page now re-specs, and the new base build has no CPU line to fill in: 18 cores standard, four more than the old floor and two above the most you could previously order [5][5].
The Ultra tier shows the money more clearly. The M3 Ultra finished its run at $5,299 with a 96GB memory limit, and MacRumors puts the M5 Ultra $200 above that [9], so $5,499 [1]. March 2025 launch buyers paid $3,999 for the Ultra and $1,999 for the Max [14]. Replacing the Ultra now is a 37.5% increase [2]; replacing the Max at $2,499 is 25% [3].
Teams justify a workstation refresh with the work they actually push through it. Apple's own comparison sheet for the M5 Ultra measures something adjacent: up to 4.3x text-to-image, 4x LLM prompt processing in LM Studio, 3.3x CopyCat training in Foundry Nuke, and exactly one figure covering the whole CPU, 1.3x multithreaded [12]. Put that next to the 37.5% price move and the cost of a unit of Apple's claimed multithreaded throughput rises about 6% (1.375 divided by 1.30) [4]. The comparison is rough, since it sets a 2025 launch price against a 2026 price after Apple's June increase [16]. It is still the arithmetic a compile queue or a colour bay has to sign for. The core split changed too, from performance-and-efficiency to 6 super cores plus 12 performance cores [6], so per-core intuitions carried over from the 14-core part will not survive contact with the machine.
The buyer this refresh actually serves is visible in the same list. Matrix multiplication runs three to four times faster on the Neural Accelerators now sitting in each GPU core [13], and Apple says the M5 Max processes prompts up to 10.7x faster than the M1 Max while the M5 Ultra trains models up to 15.4x faster than the M1 Ultra, on machines that also lack Thunderbolt 5 and AV1 decoding [15]. That is a 2021 or 2022 owner, not a 17-month-old fleet.
For Monday, sort each seat by what its work waits on. If it waits on matrix multiplication or memory bandwidth, the case is already made in Apple's numbers: 1.2TB/s is 50% up, which implies roughly 800GB/s before [10][6], and the cores you used to pay extra for are now included [5]. If it waits on top-end core count, disk capacity or the media engine, the machine on the desk stays, because those ceilings did not move [7], and the budget line becomes refurbished 2025 stock, checked against the fact that lightly discounted older units often sit too close to new after the June rise [16].
Ranked by verification strength, evidence, and original report placement.
Apple announced a new Mac Studio with the M5 Max and M5 Ultra chips, discontinuing the M4 Max and M3 Ultra models that had been on sale since March 2025.
The M4 Max started with a 14-core CPU and could be configured up to 16 cores.
The M5 Max has an 18-core CPU as standard, with no CPU upgrade available to buy.
Apple changed the core architecture, dropping the performance and efficiency split in favour of 6 super cores and 12 performance cores.
The M3 Ultra ended at $5,299 with a 96GB memory limit; for $200 more the M5 Ultra brings back 256GB and 512GB options and adds 50% more memory bandwidth.
The M5 Ultra's memory bandwidth is 1.2TB/s, which is 50% higher than before.
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 27, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
product
Apple's M5 Ultra stops being two Maxes glued together, and that changes how you size a Mac Studio1 distinct publisher
product
Apple's M6 is a 2nm inference upgrade with a CPU refresh attached1 distinct publisher
build
Mac Studio M5 Ultra vs DGX Spark: capacity says what fits, bandwidth says what you wait for1 distinct publisher
product
The Mac mini is sold by the gigabyte, which makes its refresh a DRAM story4 distinct publishers
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Specs and prices concrete, performance claims vendor-only
Configuration, pricing and connectivity details are stated precisely and are internally consistent, and the pricing arithmetic checks out. But every performance number originates with Apple as an 'up to' figure with no disclosed methodology, and the cluster contains a single publisher restating them, so nothing about real-world speedups is independently verified.
Announced and orderable, no usage evidence
Observable adoption is limited to commercial availability signals: pre-orders open, a September 22 arrival for most configurations, a later 512GB Ultra, and 10-12 week Ultra lead times in the US, which is the only demand-adjacent datapoint and is ambiguous between demand and supply constraint. There is no shipment, install-base or user-reported deployment data.
Vendor multipliers outrun what is shown
The launch narrative leans on 3.9x to 15.4x AI multipliers that no one in the cluster has tested, while the durable, checkable facts are narrower: unchanged chassis, unchanged 80-core GPU and 32-core Neural Engine ceilings, unchanged 16TB storage and 512GB memory maxima, and an up-to-1.3x multithreaded CPU gain on the Ultra alongside a $200 higher entry price. The same source partially deflates the hype by telling 2025 owners not to upgrade, which is why the gap is moderate rather than large.
Vendor marketing relayed in a purchase-guide format
All performance figures come from the manufacturer at launch, which has a direct interest in framing a same-chassis refresh as transformative, and the article is a buyer's guide whose function is to steer purchase decisions and which discusses used and refurbished stock. Offsetting this, the publisher gives explicit don't-upgrade advice and flags the price increases, which cuts against a pure promotional read.
Single publisher, checkable specs, untested performance
Confidence is capped by having one publisher and one article with no corroboration and no independent testing. It is supported by the fact that the load-bearing spec and price statements are concrete, mutually consistent and arithmetically verifiable, so the procurement conclusions are firmer than the performance conclusions.