Permanent copy of issue 41. Today's issue is always the front page.
Apple introduced iPhone 18 Pro, iPhone 18 Pro Max and the foldable iPhone Duo, all powered by a 2-nanometer A20 Pro chip that Apple said doubles on-device AI processing versus A19 Pro. Google said it will invest 13 billion euros in Finnish digital infrastructure, clean energy and local partnerships over the next two years, its largest single outlay in Europe. DeepSeek released V4.1-Flash, a 552 billion parameter mixture-of-experts model, and posted MIT weights. OpenAI said it is working with Samsung on next-generation chips.
Date: 2026-09-10
Updated: 2026-09-10 06:03
Status: LIVE
Broadcast
John Ternus, in his first Apple event as CEO, called iPhone Duo the most transformational change to iPhone since the original. Apple said Siri AI rolls out in beta with iOS 27 on 14 September, in English first, and will not be initially available in the EU on iOS. Google signed a life-extension power purchase with Fortum for the Loviisa nuclear plant, contracted 629 MW of Finnish onshore wind, and said it is enabling a 94 MW battery near Kajaani. DeepSeek said V4.1-Flash needs a quarter of the prior HBM for its KV cache. The Register quoted Kim on Samsung; the foundry declined to confirm.
Watch
Stories
Apple unveiled iPhone 18 Pro, iPhone 18 Pro Max and the foldable iPhone Duo, all powered by A20 Pro built on a 2-nanometer process. Apple said a Dual 16-core Neural Engine doubles on-device AI processing versus A19 Pro, and that Apple Intelligence uses on-device processing plus Private Cloud Compute.
-- A doubled Neural Engine on a phone is the local pole. Apple still sends some jobs to its own cloud.
2026-09-09 / Apple / chip, efficiency / local / #apple-ships-a20-pro-for-on-device-ai
Google said it will spend 13 billion euros over the next two years on digital infrastructure, clean energy and local partnerships in Finland, its largest single investment in Europe. The company named new data-center work in Hamina, Kajaani, Muhos and Vaala to serve Search, Maps and Gemini.
-- A 13 billion euro campus is cluster power. It is not a desk.
2026-09-09 / Google / energy, chip / central / #google-commits-13-billion-euros-in-finland
DeepSeek launched V4.1-Flash, a 552 billion parameter mixture-of-experts model that activates 8 billion parameters on input and 16 billion on output. The company said the model is live on its API as deepseek-flash, posted MIT weights on Hugging Face, and will route V4-Pro requests to Flash from 14 September.
-- MIT weights and a smaller KV cache cut the rent. They do not move training off the cluster.
2026-09-10 / DeepSeek / efficiency, chip / mixed / #deepseek-releases-v4-1-flash-weights
Harrison Kim, OpenAI's Korea general manager, said in Seoul that the lab is working with Samsung Electronics on joint production and research for next-generation chips. Samsung said it could not confirm customer information. OpenAI's first custom inference chip, Jalapeno, was designed with Broadcom and made at TSMC.
-- A second foundry for OpenAI custom silicon keeps inference in the lab's rack.
2026-09-09 / The Register / chip / central / #openai-says-samsung-will-make-next-chips
Axis
local [####################----------] central
Score: 66
clusters still winning
down 1
Apple put a 2-nanometer A20 Pro with a doubled Neural Engine into new iPhones. Google pledged 13 billion euros of Finnish infrastructure and a nuclear life-extension deal. DeepSeek shipped MIT weights. OpenAI is courting Samsung for custom silicon. The cluster still holds the training and the power.
Forces
| Force | Meter | Score | Pull | Note | ||
| CHIP / CAPEX |
|
83 | central | Apple's A20 Pro is pocket silicon. OpenAI's next chips and Google's Finnish campuses are still cluster capex. | ||
| ENERGY |
|
88 | central | Google signed a Fortum life-extension deal for Loviisa nuclear and a 94 MW battery near Kajaani. That power is for data centers. | ||
| EFFICIENCY |
|
46 | local | Apple says on-device processing plus Private Cloud Compute. DeepSeek cut KV cache to a quarter of prior HBM. Both still leave training in the rack. |
Line charts (0=local 100=central 2026-08-25 to 2026-09-10)
Market overlay (2026-08-12 to 2026-09-10)
Latest: BTC $77,999 / hashrate ~936.23 EH/s / token markers avg $5.3033/M / axis 66. 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 $77,999 / hashrate ~936 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
Subscribe
$ echo you@email.com | underclassic --subscribe
Previous issue | This week | Next issue | archive
One email per issue. No tracking pixels, no filler.
home | rss | page 2 | archive | models | subscribe | data | board | mcp | llms.txt