Issue 23: Apple Puts 512GB on the Desk
Permanent copy of issue 23. 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: 2026-08-25
Updated: 2026-08-25 09:46
Status: LIVE
Broadcast
Apple.com today is a desk-silicon ad: Mac mini with M6, Mac Studio with M5 Ultra. Apple says the Ultra holds 512GB of unified memory and can run models with hundreds of billions of parameters locally, with no token meter. That is the local pole in hardware. The other pole did not move. Cloud operators still spend more than they take in on cluster capex, and grid queues still throttle new halls. Clusters still win on volume. The new fact is that a quiet box on a desk now claims the memory class that used to mean a rack.
Watch
Stories
Apple Ships M6 and M5 Ultra for On-Device Models
Apple announced Mac mini with the 2 nm M6 chip and Mac Studio with M5 Ultra. Apple says M5 Ultra offers up to 512GB of unified memory and 1.2TB/s of bandwidth, enough to run models with hundreds of billions of parameters on device. Pre-order today, ships September 22.
A desk Mac with 512GB of unified memory is a real alternative to rented cluster inference when the weights fit, which is the local pole of this site.
2026-08-25 / Apple / chip, efficiency / local
AI's Absurd Spending Boom? Hyperscalers Are Spending 102% of Cloud Revenue on Capex
Cloud revenue growth no longer covers AI hardware outlays. The ratio signals sustained central concentration of spend.
Capex at 102% of revenue locks capital into large clusters and delays any shift of workloads to smaller local sites.
2026-08-22 / Yahoo Finance / chip / central
Hyperscalers' Off-Grid Power Push Comes With Risks
Hyperscalers are building behind-the-meter generation to bypass grid queues. The move raises fuel, emissions, and permitting questions.
Central operators are spending heavily to secure power, widening the gap between grid-tied local clusters and hyperscale builds.
2026-08-20 / WSJ / energy, chip / central
Poolside Launches Laguna S 2.1 Open-Weight Coding Model
A new 200B-class open-weight model targets code completion and agent tasks. Weights are released under a permissive license.
Open weights allow organizations to run code models locally or on private clusters, reducing reliance on central token APIs.
2026-08-24 / HPCwire / efficiency / local
Axis
local [##################------------] central
Score: 59
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
| Force | Meter | Score | Pull | Note | ||
| CHIP / CAPEX |
|
74 | central | Hyperscaler AI capex still exceeds cloud revenue. Apple's new silicon is desk-side volume, not cluster halls. | ||
| ENERGY |
|
82 | central | Texas still halts new data-center grid connections; power queues limit central buildout speed. | ||
| EFFICIENCY |
|
38 | local | Apple now sells a desk Mac with 512GB unified memory for local models; list token prices keep falling. |
Line charts (0=local 100=central 2026-08-10 to 2026-08-25)
Market overlay (2026-07-27 to 2026-08-25)
Latest: BTC $80,681 / hashrate ~866.35 EH/s / token markers avg $5.3033/M / axis 59. 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 $80,681 / hashrate ~866 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)
| # | Model | Org | CC | Weights | Tier | In | Out | Blend |
| 1 | Claude Sonnet 5 | Anthropic | US | closed undisclosed |
workhorse | $2.00 | $10.00 | $4.00 |
| 2 | GPT-5.5 | OpenAI | US | closed undisclosed |
flagship | $5.00 | $30.00 | $11.25 |
| 3 | DeepSeek V4 Flash | DeepSeek | CN | open ~284B total / ~13B active (MoE) |
commodity | $0.44 | $1.32 | $0.66 |
If useful, support the broadcast.
bc1qfs3yw5qlq8sxs50crzh3g84ug27gvww9npu85n
home | rss | archive | models | data | board | mcp | llms.txt