Personal AI computers compared: what each one actually runs

Choosing a machine? The AI computer picker checks every machine against the job you have in mind, chat, agents, images or video, and shows the memory to look for, where to buy, and what renting the same job costs. For a team, private AI for your team sizes one machine by how many people ask at once.

What counts as a personal AI computer

A personal AI computer is a machine that runs open models at home, all day, without sending anything to a cloud. Two numbers decide what one can do: how much memory its GPU can use, which sets the biggest model it holds, and how fast the GPU reads that memory, which sets how quickly answers come. Everything below is computed from the makers' published specs with the same formulas as the rest of this site. It is not a hands-on review.

It is not the same as an "AI PC". Laptops sold as AI PCs are built around an NPU, and Microsoft's Copilot+ PCs offer one of over 40 trillion operations per second (TOPS). TOPS measures arithmetic speed; it does not say which model fits, which memory decides, or how fast it answers, which bandwidth decides. Several machines here have an NPU as well. Every figure on this page is for the GPU.

Private here means the model runs on the machine, so your prompts and files stay on it. Whatever you connect it to, web search, a cloud model, an update server, still sends data out; the private AI server guide goes through who can see what, layer by layer. The name is also used for assistant software and cloud desktops; this page compares hardware.

The numbers that matter

"GPU can use" is the figure every fit on this site is measured against. On the DGX Spark it is what the system leaves free, less a 2 GB display reserve; on an RTX Spark it is the 102 GB NVIDIA's memory rule guarantees on a 128 GB machine, since NVIDIA does not print the size of the block it reserves for the GPU; on the Ryzen AI Max+ 395 it is the 96 GB AMD lets you set aside as graphics memory; on a Mac it is about 75% of RAM, the macOS limit observed on large-memory Macs. Each machine is shown at its largest memory option; the 512 GB Mac Studio M5 Ultra needs the 80-core GPU chip, and Apple's page, read on 3 November 2026, still said it comes in late October. Most of the machines that share memory hold more than a graphics card and read it more slowly; the Mac Studio M5 Ultra reads it faster than the Ghost Core's and the Olares One's GPUs.

NVIDIA's DGX Spark page lists the same GB10 chip in boxes from Dell (Pro Max with GB10), HP (ZGX Nano AI Station), ASUS (Ascent GX10), Lenovo (ThinkStation PGX), Acer (Veriton GN100), Gigabyte (AI TOP ATOM) and MSI. At 128 GB they run what the Spark runs; some partners also sell a version with half the memory, which holds less. The DGX Spark, Strix Halo and Mac Studio comparison goes deeper on those three.

What each one runs

Of the 40 open models tracked on this site, at Q4_K_M with 8k of context, and for an agent at the 65,536 tokens of context Ollama recommends for OpenClaw:

"The one to run" is the same pick as on each GPU page: the biggest class that fits with room for a conversation and generates at a usable speed, then the newest release. The agent column uses the rule from the OpenClaw guide: only models whose chat template passes tools through, with the context already counted.

Speed on the same models

Estimated tokens per second at Q4_K_M and 8k of context, single stream. These are ceilings worked out from memory bandwidth, not measurements; figures marked ≤ are where real runtimes fall furthest below them. "No" means the model does not fit.

On a model both can hold, the card with the faster memory wins: the RTX 5090 and the RTX PRO 6000 read memory at 1,792 GB/s, and the Ghost Core's card at 432 GB/s is still 1.6 times as fast as the DGX Spark. The machines with a large shared pool win as soon as the model stops fitting on a card. These figures describe generation; reading a long prompt first is a separate cost, and it is markedly slower on Apple silicon than on NVIDIA hardware.

Ghost Core, up close

Ghost's spec page lists an NVIDIA RTX PRO 4000 Blackwell SFF Edition with 24 GB of GDDR7 ECC memory and 432 GB/s of bandwidth, a 6-core Ryzen 5 9600X, 64 GB of DDR5 and a 1 TB SSD, and 770 AI TOPS. It said the first batch ships on 31 October 2026; on 3 November 2026 the page showed it sold out, with a form to register interest. The SFF Edition is the 70 W version of the card; the full-size RTX PRO 4000 reads its memory at 672 GB/s.

Ghost lists four models. Three are on this site's list:

The fourth, "Qwen 3.8-Next", matches no model on Hugging Face by that name. The nearest is Qwen3.8-Flash-Next: 125B parameters with 6B active, plus a 51B n-gram embedding that Qwen says suits offloading. On a 24 GB card most of that would have to live in the 64 GB of system memory. This site computes GPU memory only, so it gives no figures for that model; ask Ghost which version and quantisation it runs.

On this card those three generate at about 17 to 19 tokens per second at best, below the 25 this site calls comfortable for chat. That is why the everyday pick in the table above is the smaller Gemma 4 12B.

For an agent, the card holds what it needs. Muse Glimmer 30B, Meta's model built for local agents, needs 19.4 GB at 64k of context, against the 22.8 GB this site plans for on a 24 GB card. Qwen3.8 27B needs 21.7 GB at the same setting, at about 19 tokens per second. The card cannot hold a 70B model: Llama 3.3 70B needs about 45.8 GB at Q4 with 8k of context.

Which to choose

• Ghost Core if you want a quiet box that runs a 27B to 30B model all day and gives your agents a local endpoint. It holds the least memory here, with the Olares One, and its card is the slowest of the graphics cards on this page.
• Olares One if you want a small, quiet box that answers faster on the same 24 GB: its laptop RTX 5090 reads memory at 896 GB/s. Its 96 GB of system memory does not add to what the GPU holds.
• DGX Spark, or another maker's GB10 box, if you need CUDA and room for 70B and 120B-class models. It reads memory at 273 GB/s, so big models generate slowly.
• An RTX Spark laptop if you want CUDA and room for 70B and 120B-class models in a laptop, on sale from 16 October 2026. It reads memory at 300 GB/s, so big models generate slowly, and only the 128 GB configurations hold them.
• A Ryzen AI Max+ 395 mini PC if you want one x86 machine that runs Windows and holds large mixture-of-experts models. Check the power limit of the specific machine.
• A Mac mini M5 Pro if you want a small, quiet Mac. At Q4 it runs the same models as a 32 GB card, more slowly: Qwen3.8 27B generates at about 13 tokens per second. Its extra memory goes to 8-bit files: at Q8 it holds 34 of 40 models, against 17 on an RTX 5090. Choose 64 GB when you order, since Mac memory cannot be added later.
• A Mac Studio M5 Ultra if you want the most memory and the fastest generation of the shared-memory machines, and do not need CUDA.
• A PC with an RTX 5090 if speed matters most and your models fit in 32 GB.
• An RTX PRO 6000 if you want 96 GB on one card at the 5090's speed, in a workstation.

Before you buy

• Memory the GPU can use, not the memory on the box. On a machine with a graphics card, system memory does not hold the model: the Olares One has 96 GB of it and 24 GB for models. A Mac's GPU can use about 75% of its memory.
• Bandwidth. It sets how fast answers come. On a model both can hold, a machine that reads memory twice as fast generates about twice as fast.
• Software. CUDA runs nearly every local AI tool, and image and video tools mostly expect NVIDIA. AMD runs llama.cpp and Ollama well. Macs run MLX and llama.cpp. RTX Spark laptops run Windows on Arm, so check that your apps do.
• Memory is fixed. On every shared-memory machine here, the memory is chosen when you order and cannot be added later.
• Power and heat. An RTX 5090 needs a large power supply and case. Each Ryzen AI Max+ 395 maker sets its own power limit, so the same chip runs at different speeds.

Where to buy

• Ghost Core: Ghost
• Olares One: Olares
• NVIDIA DGX Spark: NVIDIA, Dell, Lenovo, Acer
• RTX Spark laptops: NVIDIA Marketplace
• Ryzen AI Max+ 395 mini PCs: GMKtec, GMKtec EU, Framework, Minisforum, Beelink, Corsair
• Mac mini M5 Pro: Apple
• Mac Studio M5 Ultra: Apple
• A PC with an RTX 5090: NVIDIA Marketplace
• A workstation with an RTX PRO 6000: NVIDIA Marketplace

Each link goes to the maker's or the seller's own page, through /go/, which adds one to the day's count of clicks and keeps nothing that identifies you (privacy notice). There are no prices here: hardware prices have moved too much this year for one written on this page to stay true, so check today's on the page the link opens.

Or no hardware at all

None of these is the cheap way to run a big model now and then. If the large models, a long video render or a fine-tune are an occasional need, renting a larger GPU for those hours costs less than a machine sized for them. The build vs rent calculator finds the break-even month with your own prices and hours.

To rent one, the cloud GPU price guide lists the cheapest hour for each card, and ComfyUI in the cloud compares the services that bill per generation instead.

Questions

What is the best personal AI computer?

It depends on the biggest model you want to run. For the largest models, a Mac Studio M5 Ultra: its GPU can use up to 384 GB. For CUDA with a lot of memory, the DGX Spark or another GB10 box (126 GB), or an RTX Spark laptop for the same idea on the move (102 GB). For speed on models up to 32 GB, a PC with an RTX 5090, which reads memory at 1,792 GB/s. For a small, quiet machine, a Mac mini M5 Pro (48 GB), or for 27B to 30B models all day the Ghost Core or the Olares One, whose GPU reads memory 2.1 times as fast on the same 24 GB.

Is a personal AI computer the same as an AI PC?

No. AI PCs are sold on their NPU, and Microsoft's Copilot+ PCs offer one of over 40 trillion operations per second (TOPS). Which open model a machine runs is decided by the memory its GPU can use, and how fast it answers by how fast the GPU reads that memory. An NPU adds neither.

What are the alternatives to the DGX Spark?

The same GB10 chip and 128 GB come in boxes from Dell, HP, ASUS, Lenovo, Acer, Gigabyte and MSI, and run what it runs. Off that chip, a Ryzen AI Max+ 395 mini PC gives its GPU 96 GB on x86, with Windows, and a Mac Studio M5 Ultra up to 384 GB, read 4.4 times as fast, without CUDA.

Can the Ghost Core run OpenClaw?

Yes. At the 64k of context Ollama recommends for OpenClaw, Qwen3.8 27B fits, but tightly: it needs 21.7 GB of the 22.8 GB this site plans for on the card, at about 19 tokens per second. Muse Glimmer 30B, which Ghost also lists, fits with room at 19.4 GB.

Ghost Core or DGX Spark?

The Ghost Core reads memory 1.6 times as fast (432 GB/s against 273 GB/s), so it generates faster on any model both can hold. The DGX Spark gives its GPU 126 GB against 24, about 5 times as much, which is what it takes to run 70B and 120B-class models. Both use NVIDIA GPUs.

Can a personal AI computer run a 70B model?

At Q4_K_M with 8k of context, Llama 3.3 70B needs about 45.8 GB. That fits on the DGX Spark, an RTX Spark laptop, a Ryzen AI Max+ 395 mini PC, a Mac Studio M5 Ultra and an RTX PRO 6000 workstation. It does not fit on a 24 or 32 GB card.

Do I need a personal AI computer, or can I rent one?

Buying makes sense when the machine is busy most days. If the big models are an occasional need, renting a larger GPU for those hours costs less than a machine sized for them; the build vs rent calculator finds the break-even with your own numbers.

• Ghost Core: Ghost's product page, read on 6 October 2026, 28 October 2026 and 3 November 2026. Card figures from NVIDIA's RTX PRO 4000 Blackwell SFF and RTX PRO 4000 Blackwell pages.
• RTX Spark: NVIDIA's product page for the memory, and its developer porting guide and unified-memory guide for the bandwidth and the GPU's share, read on 9 October 2026.
• Olares One: Olares's store page (an RTX 5090 Mobile with 24 GB of GDDR7, 96 GB of DDR5) and product page, read on 28 October 2026 and 3 November 2026. The GPU's memory and bandwidth from NVIDIA's RTX 50 laptop page.
• DGX Spark and the other GB10 boxes: NVIDIA's DGX Spark page, read on 3 November 2026. Mac Studio and Mac mini: Apple's Mac Studio and Mac mini specifications, read on 3 November 2026.
• AI PCs: Microsoft's Copilot+ PC page for the TOPS figure, read on 3 November 2026.
• Memory and bandwidth for every other machine: each maker's specification, linked on each machine's GPU page.
• Models: each model's config.json; Qwen3.8-Flash-Next's parameter counts from its model card.
• Fits and speeds: the same formulas as the VRAM calculator and the speed estimator.