Underclassic   rss   archive

Issue 24: Apple Ships 512 Gb Desk Box

Permanent copy of issue 24. Today's issue is always the front page.

Apple just put 512GB of unified memory in a desk Mac and said it can run hundred-billion-parameter models without a token meter. Hyperscalers still spend more than they take in on cluster chips, so the money remains central.

Date: 2026-08-26
Updated: 2026-08-26 02:00
Status: LIVE

Broadcast

Apple released new desktop chips with 512 GB unified memory and listed them for local AI workloads. At the same time, hyperscalers reported capex above cloud revenue and U.S. data centers approached 40 % of global electricity use. Grid queues remain long. The result is two parallel tracks: desk-side boxes for models that fit in one chassis, and larger clusters that still require new substations and multi-year build times.

Watch

• Track Apple M5/M6 volume shipments and reported local-model usage.
• Watch state-level grid rulings on new data-center loads.
• Compare list token prices for open-weight models against the cost of a loaded Mac Studio.

Stories

Apple introduces M6 and M5 Ultra for a big leap in performance and AI compute
Apple's latest silicon doubles unified memory on desk machines. A single Mac Studio can now hold models previously reserved for racks.
Unified memory in a desk box lowers the bar for running large models without a metered token plan.
2026-08-26 / Apple / chip, efficiency / local

Apple Refreshes the Mac Mini and Mac Studio, Both With New AI-Ready Chips
New M5 and M6 chips ship in Mac Mini and Mac Studio lines. Apple lists the machines as on-device AI workstations.
Volume silicon in consumer chassis widens the set of users who can run frontier models locally.
2026-08-25 / WIRED / chip, efficiency / local

AI's Absurd Spending Boom? Hyperscalers Are Spending 102% of Cloud Revenue on Capex
Public filings show hyperscalers spent more on AI hardware than they earned from cloud services in the latest quarter. The gap is funded by other lines of business.
Capex above revenue keeps the largest models and the cheapest tokens inside central clusters.
2026-08-22 / Yahoo Finance / chip / central

The U.S. Now Uses Nearly 40% Of The World's Data Center Electricity
Data centers account for a rising share of U.S. power demand. Grid operators report queues and delayed connections for new halls.
Power constraints slow new central capacity and raise the relative value of local inference on existing sockets.
2026-08-23 / Forbes / energy / central

Axis
local [#################-------------] central
Score: 57   clusters still winning   -2 toward local this week
Most AI still runs in big central data centers. The new fact is a desk Mac with 512GB of unified memory, which is a local-inference box, not a campus.

Forces

Line charts (0=local 100=central 2026-08-11 to 2026-08-26)

Market overlay (2026-07-28 to 2026-08-26)

Latest: BTC $79,071 / hashrate ~850.36 EH/s / token markers avg $5.3033/M / axis 57. These lines are different units. We compare timing, not dollars-to-hashrate.

Not shown: GPU cloud $/hour, hub electricity, miner hashprice, or lab cost per token -- none of those are a clean public feed.

Power and list prices
BTC $79,071 / hashrate ~850 EH/s. List-price gap (not a subsidy %): 44.4x list gap (blended) GPT-5.6 Luna -> Claude Fable 5 ($0.45/M to $20.00/M). (mempool.space + Coinbase/CoinGecko)

Marker prices (USD per 1M tokens; verified 2026-08-19; blended = (3*input + 1*output) / 4; -- = open/self-host or no single public list)

Full 20-model board

If useful, support the broadcast.
bc1qfs3yw5qlq8sxs50crzh3g84ug27gvww9npu85n

home  |  rss  |  archive  |  models  |  data  |  board  |  mcp  |  llms.txt