Apple is testing an app called Harbor that turns a Mac into a private AI inference server and pools the unified memory of two or more Macs connected over Thunderbolt 5, AppleInsider reported on 10 November, citing a description first posted on X by Aaron Perris. Harbor is in a private preview under Apple’s Claris brand, the reports say.
Two Mac Studios with the M5 Ultra chip at the 512GB maximum memory listed on Apple’s specification page come to 1,024GB, or about 1TB, of unified memory, by our arithmetic. Apple has not announced Harbor; we found no Apple or Claris page about it.
Business Pill · The Memory Wall
An evergreen Business Engineer explainer on why memory, not raw chip speed, limits AI performance, the constraint Harbor’s memory pooling is described as addressing. It teaches the general idea only and does not discuss Harbor.
The key insight: As we read it, the reported product moves AI work onto hardware the customer already buys: Harbor would pool the memory of Macs a user owns, so two top-memory Mac Studios give about 1TB of unified memory, by our arithmetic, with no cloud service in the loop.
The description Perris posted says Harbor runs open-weight models locally on Apple silicon and that “your data never leaves your hardware, and nothing reaches the cloud”.
It says users can “Connect two or more Thunderbolt 5-capable Mac computers and Harbor pools their unified memory into a single inference resource”, to run models that exceed one computer’s capacity or serve more requests at once.
Setup is pitched as “Set up in minutes, not hours”, with “no manual configuration, no command line, no infrastructure expertise required”. Users browse, download and load open-weight models inside the app, the description says.
Unified memory by Mac Studio configuration, from Apple’s specification page, and the 1,024GB two top-memory M5 Ultra Mac Studios would pool if Harbor works as the reported description says. The 1,024GB is our arithmetic; Apple has not announced Harbor.
What Was Posted on X
Apple is working on a new app called Harbor
Harbor turns your Mac into a private AI inference server. Run open-weight models locally on Apple siliconโyour data never leaves your hardware, and nothing reaches the cloud. Connect multiple Mac computers over Thunderbolt 5 to pool unified memory, enabling large models and higher concurrency that a single computer can handle alone.
Set up in minutes, not hours Harbor automatically detects your devices over Thunderbolt 5 to form a cluster. Plug in, start, and youโre runningโno manual configuration, no command line, no infrastructure expertise required.
Download and run models directly Browse, download, and load open-weight models from within the app. Harbor handles modelmanagement so you can focus on the work, not the setup.
Cluster for scale Connect two or more Thunderbolt 5-capable Mac computers and Harbor pools their unified memory into a single inference resource. Run models that exceed any single computerโs capacity, or serve more concurrent requests across your cluster.
Harbor is currently in a private preview under the Claris brand.
Aaron Perris posted at 18:56 UTC on 10 November a description of Harbor and four screenshots, saying Apple is working on the app and that it is in a private preview under the Claris brand. This piece quotes that description; Apple has not announced Harbor.
I told yโall
@johnternus is gonna bet the $AAPL farm on local models
The investor Jason Calacanis posted at 03:01 UTC on 11 November, quoting a post that linked AppleInsider’s report, that Apple’s John Ternus would bet the company on local models. This piece reports it as his view.
The Hardware Harbor Would Pool
Apple lists the Mac Studio with the M5 Max chip at 36GB of unified memory, configurable to 128GB, and with the M5 Ultra at 96GB, configurable to 256GB or 512GB. The M5 Ultra is listed with 1.2TB/s of memory bandwidth.
The Mac Studio has four Thunderbolt 5 ports on the back, each listed at up to 120Gb/s, per the same page. Harbor detects Macs over Thunderbolt 5 and forms a cluster automatically, the description says.
AppleInsider says Harbor was formerly known as Fabric, and that Apple had already added Thunderbolt clustering to its MLX machine-learning framework and later Thunderbolt 5 support in macOS and MLX. Harbor is about making setup easier rather than making queries faster, it wrote.
Local Models and Apple’s AI Spending
The investor Jason Calacanis read the report as a strategy signal. He posted on 11 November: “I told y’all @johnternus is gonna bet the $AAPL farm on local models”. That is his view; Apple has said nothing about it.
As we read the description, the pitch is about ease of setup more than speed, which AppleInsider also says: clustering Macs over Thunderbolt was already possible with Apple’s MLX framework, and Harbor is described as removing the command line from it.
As we read the hardware page, memory is the constraint the app targets: the description says pooling lets users run models that exceed any single computer’s capacity, and the largest Mac Studio lists 512GB.
Harbor app description, as posted on X by Aaron Perris
“Run models that exceed any single computer’s capacity, or serve more concurrent requests across your cluster.”
Three Implications
FOR BUYERS OF AI The description says data never leaves the user’s hardware, which is the case it makes against sending prompts to a cloud service.
FOR APPLE’S HARDWARE Pooling is tied to Thunderbolt 5-capable Macs, and the reported cluster is built from Macs the user buys.
FOR OPEN MODELS Harbor is described as running open-weight models, which users browse, download and load inside the app.
The framework argues: “Close the SCARCE layer (keep proprietary) + Open the ABUNDANT layer (commoditize)”.
As we read it, Harbor as described fits that pattern: the Macs and their unified memory stay Apple’s, while the models it runs are open-weight. The framework predates the reports and does not discuss Harbor.
What Is Not Established
The reports we read give no release date, no price, no list of supported models and no figure for how fast a cluster runs. Whether Harbor will ship under the Apple or Claris name, or at all, is not stated.
The 1TB figure assumes two top-memory Mac Studios and that Harbor pools all of their memory; how much a model can actually use is not in the description. We did not contact Apple.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Apple’s Claris is privately previewing Harbor, an app that pools the memory of Thunderbolt 5-linked Macs to run open-weight AI models locally, according to the reports; two 512GB Mac Studios would give about 1TB, by our arithmetic.
A note on sourcing. Apple has not announced Harbor and we found no Apple or Claris page about it. The app’s description is from the X post by Aaron Perris of 10 November 2026, read in full, and from AppleInsider’s report of the same day, read in full. Memory and port figures are from Apple’s Mac Studio specification page. Jason Calacanis’s post was read in full. We did not contact Apple.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
