Apple M6, M5 Ultra Debut: 80-Core GPU, 512GB RAM [2026]

Apple used a Tuesday morning press release instead of a keynote to introduce the biggest jump in Mac silicon since the M1 Ultra launched in 2022. The M6 and M5 Ultra chips, unveiled August 25, 2026, land in a redesigned Mac mini and a refreshed Mac Studio, and the headline numbers are hard to ignore: an 80-core GPU, up to 512GB of unified memory, and Apple’s first 2-nanometer chip built on TSMC’s N2 process. The announcement, confirmed on Apple’s own newsroom and covered within hours by TechCrunch, Tom’s Hardware and MacRumors, pushes Apple deeper into a fight it wasn’t originally supposed to be in: local AI compute for developers and small AI shops who don’t want to rent a cloud GPU cluster.

This is not a routine spec bump. The M5 Ultra is Apple’s first “quad-die” chip, fusing two dual-die M5 Max processors with a next-generation version of Apple’s UltraFusion interconnect. The M6, meanwhile, is the first Apple silicon chip built on a 2nm node, arriving in the Mac mini alongside a carryover M5 Pro option. Both chips ship inside real, orderable hardware starting September 22, 2026, with prices that climbed noticeably from the previous generation. That combination of new manufacturing process, new packaging architecture and higher prices is what makes this launch worth unpacking in detail.

What Apple Announced: M6 and M5 Ultra at a Glance

Apple’s newsroom posted two separate announcements on August 25: one covering the new Mac Studio with M5 Max and M5 Ultra, and a second covering the new Mac mini with M6 and M5 Pro. Together they represent a full refresh of Apple’s desktop compute lineup, aimed squarely at the same audience buying Nvidia’s DGX Spark and AMD’s Ryzen AI Max-based mini PCs: developers, researchers and small teams that want to run large language models locally instead of paying per-token cloud bills.

The M6 replaces the M5 as the entry-level chip in the Mac mini and is built using TSMC’s second-generation 2-nanometer process, a first for any Apple silicon product. It carries a 12-core CPU (two more cores than the M5), a 12-core GPU with Neural Accelerators in every core, and what Apple describes as the first dual 16-core Neural Engine layout in a Mac chip. Memory bandwidth rises to 170GB/s, up from 153GB/s on the M5, though maximum unified memory is capped at 32GB — a ceiling clearly meant to keep the Mac mini positioned below the Mac Studio.

The M5 Ultra is the far bigger story for the AI crowd. It fuses two M5 Max dies using an updated UltraFusion interconnect, giving Apple its first quad-die chip and, per the company, its “most powerful chip ever.” Configurations run up to a 36-core CPU and an 80-core GPU, with memory bandwidth reaching 1.2TB/s and maximum unified memory topping out at 512GB — enough, in theory, to hold frontier-scale open-weight models entirely in memory without offloading to disk.

M6 Chip Specs: Apple’s First 2-Nanometer Silicon

The M6’s headline feature isn’t a core count, it’s the manufacturing node. Moving to TSMC’s 2nm process lets Apple pack more transistors into a smaller die, which the company says translates into a “major leap in performance and power efficiency,” according to reporting from Apple’s own newsroom post. The new 12-core CPU complex splits into two “super cores,” four performance cores and six efficiency cores, a reshuffled layout compared to the M5’s core arrangement.

On the graphics side, the M6’s 12-core GPU keeps the Neural Accelerator design Apple introduced with the M5 generation, embedding dedicated matrix-multiplication hardware directly into each GPU core rather than relying solely on the separate Neural Engine. Apple’s claims, per Tom’s Hardware and PCWorld’s coverage of the launch, put the M6 Mac mini at up to 4x faster AI performance, 2x faster storage and graphics throughput, and roughly 40% faster CPU performance compared with the outgoing M4 Mac mini. Independent benchmarks had not been published as of this writing, so those figures should be treated as Apple’s own claims until third-party testing catches up.

M5 Ultra Specs: The First Quad-Die Mac Chip

Where the M6 is an efficiency story, the M5 Ultra is a brute-force one. Apple’s own performance claim, published on its newsroom site, states: “With M5 Ultra, Mac Studio achieves up to 4.3x the peak AI compute performance of M3 Ultra and a staggering 9.8x more than M1 Ultra” (Apple Newsroom). That comparison against the M1 Ultra is a deliberate one: it frames four chip generations of progress since the Mac Studio’s 2022 debut.

The GPU side gets an even bigger jump. Apple states: “M5 Ultra features a next-generation GPU with up to 80 cores, incorporating a Neural Accelerator in each core to offer up to 4.5x the peak GPU compute for AI compared to M3 Ultra and over 6x more than M1 Ultra” (Apple Newsroom). Memory bandwidth climbs to 1.2TB/s, which Apple describes as 50% more than the M3 Ultra offered. The chip also doubles video encode and decode blocks, letting a single Mac Studio handle up to 33 simultaneous streams of 8K ProRes video at 30fps, a spec aimed squarely at video production houses rather than AI developers.

Full Spec Comparison: M6, M5 Pro, M5 Max and M5 Ultra

The table below lines up the four chips Apple is now shipping across the Mac mini and Mac Studio lineup, based on Apple’s published specifications.

Chip	Process Node	Max CPU Cores	Max GPU Cores	Memory Bandwidth	Max Unified Memory
M6	TSMC 2nm (N2)	12-core	12-core	170GB/s	32GB
M5 Pro	TSMC 3nm (N3P)	18-core	20-core	~300GB/s	64GB
M5 Max	TSMC 3nm (N3P)	18-core (dual-die)	40-core	614GB/s	128GB
M5 Ultra	TSMC 3nm (N3P)	36-core (quad-die)	80-core	1.2TB/s	512GB

Note that the M5 Ultra is still built on the same 3nm process as the M5 Max and M5 Pro; the 2nm jump is exclusive to the M6 for now. That’s a departure from Apple’s usual pattern of rolling out a new node across the whole lineup at once, and it suggests 2nm yields at TSMC weren’t yet ready to support Apple’s largest, most complex die when the M5 Ultra needed to ship.

New Mac Mini and Mac Studio Pricing

Pricing is where this launch gets contentious. The entry-level Mac mini with M6, 16GB of memory and 256GB of storage now starts at $899, up from $799 for the outgoing M4 configuration according to Macworld and TechRadar’s coverage of the announcement. The M5 Pro Mac mini configuration starts at $1,699 with 24GB of memory and a 512GB drive.

On the Mac Studio side, Apple’s own pricing statement reads: “Mac Studio with M5 Max starts at $2,499 (U.S.) and $2,299 (U.S.) for education” and “Mac Studio with M5 Ultra starts at $5,499 (U.S.) and $5,099 (U.S.) for education” (Apple Newsroom). That $5,499 starting price reportedly covers a 96GB memory configuration; pushing to the full 512GB tier adds thousands more, with outlets tracking the fully loaded configuration landing north of $9,500 once memory and storage upgrades are added.

Pricing and Availability Table

Product	Chip	Starting Price (US)	Base Config	Ships
Mac mini	M6	$899	16GB / 256GB	Sept 22, 2026
Mac mini	M5 Pro	$1,699	24GB / 512GB	Sept 22, 2026
Mac Studio	M5 Max	$2,499	36GB / 512GB	Sept 22, 2026
Mac Studio	M5 Ultra	$5,499	96GB / 1TB	Sept 22, 2026

Both the Mac mini and Mac Studio were announced on August 25 but won’t actually ship until September 22, a nearly four-week gap between announcement and availability that gives reviewers and enterprise buyers time to place orders before third-party benchmarks exist.

Mac Studio M5 Ultra vs. Nvidia DGX Spark and AMD Ryzen AI Max

The most interesting comparison isn’t Apple against itself, it’s Apple against the two other companies chasing the same “personal AI supercomputer” category. Nvidia’s DGX Spark ships with 128GB of coherent LPDDR5x memory at 273GB/s of bandwidth and costs $4,699, a price that climbed from its original $3,999 launch tag due to what reporting has attributed to a memory-shortage surcharge. AMD’s Ryzen AI Max+ 395 chip, sold in systems like the Framework Desktop, ships with 128GB of LPDDR5x and starts around $1,999.

Apple’s pitch is that raw memory bandwidth wins the local-inference argument. The Mac Studio M5 Ultra’s 1.2TB/s of bandwidth is roughly 4.4 times the DGX Spark’s 273GB/s, and bandwidth is the bottleneck that determines how fast a local model can generate tokens once it’s loaded. Nvidia’s counter-argument is software: the DGX Spark runs the full CUDA stack, which still has broader framework support than Apple’s MLX and Metal tooling for many production AI pipelines. AMD’s system undercuts both on price but caps out at a quarter of the M5 Ultra’s maximum memory.

Why 512GB of Unified Memory Actually Matters

Unified memory is the feature Apple has leaned on hardest since the M1 Ultra, because it lets the CPU, GPU and Neural Engine all address the same pool of RAM instead of copying data between separate VRAM and system memory. For AI inference, that means a model’s weights only need to be loaded once, and there’s no ceiling imposed by a discrete GPU’s onboard VRAM the way there is on a typical Nvidia or AMD graphics card.

At 512GB, the Mac Studio with M5 Ultra can theoretically hold very large open-weight models, including full-precision or lightly quantized versions of frontier-class open models, entirely in memory. That’s the use case Apple is explicitly courting: independent developers and small AI teams who want to prototype, fine-tune or serve models locally without a monthly cloud GPU bill. It’s also why comparisons against enterprise-grade hardware, which typically costs far more, keep surfacing in early coverage of the launch.

Market Reaction: A Muted Wall Street Response

Despite the scale of the technical leap, investors barely reacted. Apple shares rose roughly 0.1% before the opening bell the morning of the announcement, a move small enough to register as market noise rather than a verdict on the product. The muted response tracks with the Mac’s shrinking share of Apple’s overall business: Mac revenue accounted for roughly 9.5% of Apple’s most recent quarterly revenue, meaning even a strong sales cycle for the new mini and Studio would move the needle only slightly on group results.

Wall Street’s broader read on Apple hasn’t shifted dramatically around this launch either. Analyst sentiment on the stock remains a “Buy” consensus across dozens of tracked ratings, with price targets generally implying double-digit upside from current levels, but that view is being driven by iPhone, Services and broader AI positioning rather than the Mac lineup specifically.

Historical Context: From M1 Ultra to M5 Ultra

Apple introduced the “Ultra” tier in March 2022 with the M1 Ultra, fusing two M1 Max dies via the original UltraFusion interconnect to create the first Mac Studio. The M2 Ultra followed in 2023, the M3 Ultra arrived in March 2025, and now the M5 Ultra lands roughly 18 months later, notably skipping an “M4 Ultra” entirely — Apple never shipped an Ultra variant of the M4 generation, a gap it has not explained publicly.

What’s changed most across that arc isn’t clock speed, it’s die architecture. The jump from dual-die fusion (M1 through M3 Ultra) to quad-die fusion in the M5 Ultra is the biggest structural change to Apple’s high-end silicon since the Ultra tier launched. It also explains why Apple compares the M5 Ultra against the M1 Ultra as often as it compares it to the more recent M3 Ultra: the generational gap is now wide enough that the four-year-old chip makes for a more dramatic before-and-after story.

The Memory Supply Backdrop

This launch also lands in the middle of a well-documented DRAM and memory supply crunch that has pushed component prices up across the entire PC and console industry through 2026. Nvidia’s DGX Spark price hike from $3,999 to $4,699 was itself attributed to a memory-shortage surcharge, and Apple’s own price increases across both the Mac mini and Mac Studio lines track the same industry-wide cost pressure rather than reflecting chip performance alone. Buyers configuring a Mac Studio up toward the 512GB ceiling should expect memory upgrade pricing to reflect that same shortage, which is part of why the fully-loaded M5 Ultra configuration reportedly runs north of $9,500.

What This Means for Developers and AI Engineers

For software teams evaluating local AI infrastructure, the calculus now includes a genuinely new Apple option rather than a marginal refresh. The M6 Mac mini, at $899, is a reasonable on-ramp for running smaller quantized models and doing local development against MLX or llama.cpp-style toolchains without committing to Mac Studio pricing. The M5 Ultra Mac Studio, by contrast, is aimed at teams that would otherwise be renting A100 or H100-class cloud instances for inference workloads and want to move that cost to a one-time capital purchase instead.

The tradeoff engineering teams will need to weigh is software maturity. CUDA’s ecosystem for training and fine-tuning remains broader than what Apple’s Metal and MLX stack currently supports, so the M5 Ultra is, for now, a stronger fit for inference and local prototyping than for large-scale training runs. That’s the same tradeoff that shaped adoption of the M1 and M2 Ultra chips in AI workflows, and nothing in this announcement suggests Apple has closed that gap yet.

Competitive Landscape: Apple, Nvidia and AMD Chase the Same Buyer

All three companies are now explicitly targeting the same category of buyer: individuals and small teams who want meaningful local AI compute without enterprise procurement overhead. Apple’s advantage is memory ceiling and bandwidth; Nvidia’s advantage is software ecosystem maturity; AMD’s advantage is price. None of the three has a clean win across every dimension a buyer cares about, which is likely to keep this a genuinely competitive three-way market through the rest of 2026 rather than one vendor running away with it.

Predictions: Where Apple Silicon Goes From Here

Expect a 2nm M6 Pro, M6 Max and eventual M6 Ultra to follow over the next 12-18 months as TSMC’s N2 yields improve enough to support Apple’s larger, more complex dies.
Third-party AI benchmarks against the DGX Spark and Ryzen AI Max+ systems will likely surface within weeks of the September 22 ship date and will heavily influence developer purchasing decisions.
Memory pricing pressure tied to the broader DRAM shortage is likely to keep Mac Studio upgrade pricing elevated at least through early 2027, making the base 96GB configuration the practical entry point for most buyers rather than the advertised 512GB ceiling.
Apple’s MLX framework will likely see accelerated investment to close the software gap with CUDA, given how directly the M5 Ultra’s marketing leans on local AI use cases.
Nvidia and AMD will likely respond with their own memory-capacity increases on next-generation local AI hardware rather than competing purely on price, since Apple has staked its pitch on bandwidth and capacity rather than raw compute.

Frequently Asked Questions

When do the M6 Mac mini and M5 Ultra Mac Studio ship?

Apple announced both machines on August 25, 2026, with shipping beginning September 22, 2026.

How much does the Mac Studio with M5 Ultra cost?

It starts at $5,499 in the US ($5,099 for education) for a 96GB memory configuration. Pricing rises substantially toward the maximum 512GB memory configuration, which reportedly lands north of $9,500 once fully specced.

What makes the M5 Ultra different from the M3 Ultra?

The M5 Ultra is Apple’s first quad-die chip, fusing two M5 Max dies rather than two Max dies from an earlier generation. Apple states it delivers up to 4.3x the peak AI compute performance of the M3 Ultra and up to 4.5x the peak GPU AI compute.

Is the M6 built on a new manufacturing process?

Yes. The M6 is Apple’s first chip built on TSMC’s 2-nanometer (N2) process, while the M5 Pro, M5 Max and M5 Ultra remain on the 3-nanometer node used by the rest of the M5 family.

How does the Mac Studio M5 Ultra compare to Nvidia’s DGX Spark for local AI?

The M5 Ultra offers roughly 4.4 times the memory bandwidth of the DGX Spark (1.2TB/s versus 273GB/s) and up to four times the maximum memory capacity, though the DGX Spark retains broader CUDA software ecosystem support for training workloads.

Why did the Mac mini price increase to $899?

The base M6 Mac mini rose from the M4 generation’s $799 starting price to $899. Reporting attributes part of the increase to the industry-wide memory and component shortage affecting PC and console pricing broadly through 2026, alongside the cost of the new 2nm chip.

Can the M5 Ultra run large open-weight AI models locally?

With up to 512GB of unified memory, the M5 Ultra Mac Studio can hold very large open-weight models in memory for local inference, a capability Apple is explicitly marketing toward independent developers and small AI teams.

Did Apple skip an M4 Ultra chip?

Yes. Apple never released an Ultra variant of the M4 generation, making the M5 Ultra the direct successor to the M3 Ultra from March 2025.