← Back to blog

No-Markup AI Hardware Sourcing, Sized to Your Workload

Invalid Date7 min readMaai Machines

Here is a question we get on almost every call: "Okay, I'm convinced local AI makes sense. But which computer do I actually buy, and where do I buy it so I don't get taken for a ride?"

As of today, that question has an official answer. Maai Machines now offers Hardware Recommendation and Sourcing as a standalone service: we size the machine to your actual workload, source it at the price you would pay Apple directly, and support the result anywhere in the country. No markup. No commission. No "recommended bundle" that mysteriously matches whatever a reseller has sitting in a warehouse.

This post explains why we built the service this way, how the sizing process works, and who it is for.

Why no markup matters in AI hardware advice

The AI hardware market has a trust problem. When the person recommending your computer also profits from selling it, every recommendation is suspect. Do you really need 128 GB of RAM, or does the bigger invoice need it?

We solved this by removing the incentive entirely. You pay a flat consultation fee for the sizing work, and the hardware itself costs exactly what Apple (or an authorized education or business channel) charges. If the right answer for your business is a $599 Mac Mini, we will tell you it is a $599 Mac Mini, because we make the same fee either way.

That independence cuts both ways, and we are honest about it. Sometimes the right answer is bigger than people hope. A solo bookkeeper drafting client emails does great on 16 GB of RAM. A law office feeding 50-page contracts to a 70B-parameter model does not, and we will say so with the numbers to back it up. No-markup sourcing means the recommendation is the product, so the recommendation has to be right.

How workload sizing works for a Mac Mini AI setup

Most hardware advice starts with spec sheets. We start with your Tuesday afternoon. The core of a Mac Mini AI setup recommendation is a short discovery process built around three questions:

  1. What documents does the AI need to read? Short emails and product descriptions are light work. Long contracts, patient notes, and multi-year financial records demand larger models and much more memory for context.
  2. How fast does it need to feel? Background summarization can run at 8 to 12 tokens per second. Interactive daily work needs 15 to 25 tokens per second, which is roughly as fast as you read.
  3. How many people touch it? One owner, a front-office team of four, and a multi-location group sharing a central machine are three different sizing answers.

From those answers we apply the same math we published in our Mac Mini vs Mac Studio sizing guide: a locally run model needs roughly 0.6 GB of RAM per billion parameters at standard 4-bit quantization, plus 6 to 10 GB of headroom for macOS and your everyday apps. Workload sizing turns "which Mac should I buy" from a guess into arithmetic.

Then we add one deliberate step: we size one tier above today's need, not three. Models keep getting more capable per gigabyte (our models page tracks the current picks), so buying enormous headroom "just in case" usually wastes money. Buying zero headroom wastes the machine in 18 months. One tier up is the boring, correct middle.

Apple Silicon AI: why the Mac is the box we recommend

We are a Mac shop for local AI, and the reason is architectural, not aesthetic. Apple Silicon AI performance comes from unified memory: the CPU and GPU share one pool of RAM, so an entire AI model loads into memory instead of being squeezed through a separate graphics card.

The practical consequences for a small business are big:

  • A $599 Mac Mini M4 with 16 GB runs 7B to 8B parameter models at 20 to 30 tokens per second, enough for drafting, summarizing, and Q&A.
  • A Mac Mini M4 Pro roughly doubles memory bandwidth (about 273 GB/s versus 120 GB/s), which roughly doubles response speed on the same model.
  • A Mac Studio M4 Max with 64 GB runs 70B-class models at about 18 to 22 tokens per second, which is genuine "having a conversation" speed with near-frontier reasoning quality.

The alternative path (a PC tower with a discrete GPU) can win on raw speed, but for a non-technical office it brings driver maintenance, higher power draw, and a machine that sounds like a hair dryer. The Mac draws 10 to 40 watts under typical AI load, sits silently on a shelf, and gets unboxed to useful in an afternoon. We covered the deeper technical story in Apple Silicon Changed the Local AI Equation.

And because the model runs on hardware you own, your documents never leave the building. Local processing is designed for privacy-sensitive workflows, which is exactly why the law offices and healthcare practices we work with went local in the first place.

Nationwide support, not a box on your doorstep

Hardware sourcing that ends at the shipping label is just shopping. Ours does not end there, because Maai Machines is a national AI education and implementation company, and this service plugs into everything else we do:

  • Sizing consult. A structured conversation about your workload, team, and growth plans, ending in a written recommendation with the exact configuration and where to order it.
  • Sourcing. We handle the ordering path, including business and education channel options where they apply, at zero markup.
  • Setup. If you want the machine arriving ready to work, our Local AI Setup on Mac service installs the models, configures the software, and hands you a working system.
  • Support that travels. Remote support works the same for a dental group in Phoenix, a restaurant operator in Austin, a boutique retailer in Portland, or a legal team in Manhattan. Distance does not change a screen-share.

That last point matters more than it sounds. Around 33 million small businesses operate in the US, and the overwhelming majority are nowhere near an "AI consultant" they could meet for coffee. The machine is local; the help does not have to be.

What it costs, and what it deliberately does not

The service is a one-time engagement, priced flat on our pricing page. There is no subscription attached, no monthly retainer required, and no percentage of the hardware bill. If you later want help beyond the initial setup, our ongoing local AI support is there when you ask for it, on your schedule rather than a contract's.

For context on the economics: we broke down in our cloud fee comparison how a stack of AI subscriptions for a small team routinely passes $2,400 per year, every year. A right-sized Mac is a one-time purchase in the $599 to $2,700 range for most businesses we talk to, and it is still a fully functional computer for everything else you do.

One honest boundary: hardware plus local models is the do-it-with-guidance path. If what you actually want is done-for-you AI marketing where someone else runs the whole thing, that is our sibling MOCO by Maai, and we will happily point you there instead of selling you a computer you will not use.

Who this is for (and a few who it is not)

The pattern we see across our use cases is consistent: the businesses that benefit most are the ones whose documents are sensitive, repetitive, or both.

  • Legal teams in New York triaging intake email and summarizing discovery documents on a machine that never sends a client file to a third party.
  • Dental and medical practices in Phoenix drafting treatment note summaries with local processing designed for privacy-sensitive work.
  • Restaurant groups in Austin running the automation workflows we outlined on one quiet box in the back office.
  • Retailers in Portland generating product descriptions and answering inventory questions on a $799 Mini.

Who should skip it? If your AI use is five prompts a week, a free chatbot tier is honestly fine, and we say that in consults regularly. The hardware conversation starts making sense when AI touches your business daily and the documents involved are ones you would not forward to a stranger.

Key Takeaways

  • Maai Machines now offers Hardware Recommendation and Sourcing nationwide: flat-fee sizing advice, hardware at the price Apple charges, zero markup.
  • Sizing is arithmetic, not guesswork: about 0.6 GB of RAM per billion model parameters, plus 6 to 10 GB of headroom, matched to your documents, speed needs, and team size.
  • Apple Silicon's unified memory is why a $599 to $2,700 Mac covers most small business AI workloads at 15 to 30 tokens per second.
  • The service is one-time, with optional setup and ongoing support behind it. No subscriptions, no commissions, no oversized "recommended bundles."
  • Support is remote-first and works the same in Phoenix, Austin, Portland, or New York.

Get sized before you get sold

The worst time to learn about RAM requirements is after the return window closes. If you are considering a local AI machine this year, start with a sizing consult: bring a description of your actual work, and leave with a written recommendation you can order from Apple yourself if you prefer.

Maai Machines (https://maaimachines.com) offers hardware recommendation and sourcing nationwide. See our pricing for the flat consult fee, or browse real setups to see what businesses like yours are running. The right machine is smaller than the sales guy says and bigger than the skeptic fears, and we will show you exactly which one it is.