Permanent copy of issue 68. Today's issue is always the front page.
Previous issue | This week | Next issue | archive
Google and Constellation Energy said Tuesday that Google will pay for 890 megawatts of new output from upgrades at 11 Constellation nuclear reactors over 20 years, and buy another 2,700 megawatts from the company over 15 years. AMD chief executive Lisa Su said demand for AMD's chips is still higher than its supply. Mistral AI opened a preview of a 1-trillion-parameter model and said the weights will follow by the end of October. Google released EmbeddingGemma 2, a 740-million-parameter open model built to run on phones and laptops.
Date: 2026-10-07
Updated: 2026-10-07 06:17
Status: LIVE
Broadcast
Google is paying to enlarge existing reactors on the PJM grid rather than wait years for new ones, and the grid it is enlarging is the one that lights homes from Chicago to Washington. AMD says it cannot make chips fast enough even after a year of expansion. The only file anyone could download on Tuesday was a 740-million-parameter search model small enough for a phone. The trillion-parameter model arrives at the end of the month, and it will need a server room.
Watch
Stories
Google and Constellation Energy said Tuesday they signed a 20-year power purchase agreement to add 890 megawatts through upgrades at 11 Constellation nuclear units in Illinois, Pennsylvania, and New Jersey, with the first upgrade due by 2028. The companies also signed a 15-year supply agreement for another 2,700 megawatts on the PJM grid, and Constellation said the package supports more than $4.3 billion of new investment.
-- A cloud company is now paying to enlarge reactors on the grid that serves 67 million people. Households and data centers in 13 states draw from the same system, and Google has booked the new output first.
2026-10-06 / Constellation Energy / energy / central / #google-funds-890-mw-of-new-nuclear-output-from-c
AMD chief executive Lisa Su said Tuesday in Taiwan that demand for the company's processors is higher than its supply, even after AMD added capacity through 2026, the Taipei Times reported. Su said AMD will substantially increase supply in 2027 and now plans three to five years ahead with partners including TSMC, because it needs more advanced wafer and packaging capacity.
-- The bottleneck is wafers and packaging, not buyers. In a shortage the largest contracts are filled first, and those belong to the clusters.
2026-10-06 / Taipei Times / chip / central / #amd-says-demand-for-its-chips-still-outruns-supp
Mistral AI on Tuesday opened a public preview of Mistral Large 4, a model with 1 trillion parameters and 49 billion active on each token, through its Mistral Studio API. The company said it trained the model from scratch on 3,800 Nvidia Grace Blackwell chips in its own European data centers and will release the weights by the end of October.
-- A second lab in two days has promised open weights later rather than now. A file of that size needs server racks to run, so even after release it is a local model for institutions, not households.
2026-10-06 / Mistral AI / chip, efficiency / mixed / #mistral-previews-a-trillion-parameter-model-weig
Google said Tuesday it has released EmbeddingGemma 2, an open model with 740 million parameters that turns text, code, images, video, and audio into one shared format for search, under an Apache 2.0 license with weights on Hugging Face and Kaggle. Google said the model is sized for on-device use and that the first version passed 20 million downloads.
-- This one can be downloaded today and run without a network connection. Search across a person's own photos and recordings can stay on the device.
2026-10-06 / Google / efficiency / local / #google-releases-embeddinggemma-2-for-phones-and-
Axis
local [#########################-----] central
Score: 83
clusters still winning
unchanged
Google contracted 3,590 megawatts from Constellation, including 890 megawatts of new nuclear output from reactor upgrades due from 2028. AMD said its chip supply still trails demand and will rise substantially in 2027. Mistral kept a 1-trillion-parameter model behind an API until its weights arrive at month end. Google shipped a small open model that runs on a phone today. The power contract and the chip queue still belong to the cluster.
Forces
| Force | Meter | Score | Pull | Note | ||
| CHIP / CAPEX |
|
93 | central | AMD chief executive Lisa Su said demand is higher than supply after a year of added capacity, and that AMD needs more advanced wafer and packaging capacity from TSMC and other partners. Mistral said it trained Large 4 on 3,800 Nvidia Grace Blackwell chips. | ||
| ENERGY |
|
99 | central | Google signed a 20-year agreement for 890 megawatts of new nuclear output from upgrades at 11 Constellation reactors in Illinois, Pennsylvania, and New Jersey, the first due by 2028, plus a 15-year deal for 2,700 megawatts on the PJM grid. | ||
| EFFICIENCY |
|
57 | mixed | Google released EmbeddingGemma 2 with 740 million parameters and weights on Hugging Face and Kaggle. Mistral Large 4 uses 49 billion of its 1 trillion parameters on each token, but its weights are not due until the end of October. |
Line charts (0=local 100=central 2026-09-22 to 2026-10-07)
Market overlay (2026-09-08 to 2026-10-07)
Latest: BTC $83,794 / hashrate ~988.13 EH/s / token markers avg $5.3033/M / axis 83. 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 $83,794 / hashrate ~988 EH/s. List-price gap (not a subsidy %): 100.0x list gap (blended) GPT-6 Luna -> Claude Fable 5.1 ($0.20/M to $20.00/M). (mempool.space + Coinbase/CoinGecko)
Marker prices (USD per 1M tokens; verified 2026-10-07; 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 V4.1 Flash | DeepSeek | CN | open ~284B total / ~13B active (MoE) |
commodity | $0.30 | $1.20 | $0.53 |
If useful, support the broadcast.
bc1qfs3yw5qlq8sxs50crzh3g84ug27gvww9npu85n
Subscribe
$ echo you@email.com | underclassic --subscribe
One email per issue. No tracking pixels, no filler.
home | rss | page 2 | page 3 | archive | models | subscribe | data | board | mcp | llms.txt