Permanent copy of issue 63. Today's issue is always the front page.
Previous issue | This week | Next issue | archive
Micron chief executive Sanjay Mehrotra told investors on Wednesday that 75 percent of the company's 2027 memory output is already accounted for, and that he expects demand to exceed supply. Nvidia said Thursday that OpenAI's GPT-6 Astra Ultrafast, running on Blackwell chips, generates tokens up to eight times faster than the standard mode, and that the mode is available in OpenAI's API. California Attorney General Rob Bonta said Thursday that his office served OpenAI with an investigative subpoena over cybersecurity incidents involving the company and its models. Google said Thursday that a prototype satellite is in orbit and that it will test how its chips handle space. None of those announcements puts a frontier model on a computer a person owns.
Date: 2026-10-02
Updated: 2026-10-02 06:07
Status: LIVE
Broadcast
The memory Micron has already promised for 2027 is high-bandwidth memory for AI servers, not the module in a laptop. The faster Astra mode is a setting on a rental. A subpoena asks questions of the company that holds the weights. It does not hand those weights to a household.
Watch
Stories
Micron chief executive Sanjay Mehrotra told investors on Wednesday that he expects demand for the company's memory to exceed supply, and that the shortage will continue for at least the next couple of years. He said 75 percent of 2027 output is already accounted for, and that prices for high-bandwidth memory sold for 2027 are much higher than 2026 prices.
-- The memory left for a home computer is what remains after AI servers take the wafers.
2026-10-01 / Ars Technica / chip / central / #micron-says-memory-shortage-runs-to-2028
Nvidia said Thursday that OpenAI's GPT-6 Astra Ultrafast is available in the OpenAI API and to eligible ChatGPT Work and Codex users, running on Nvidia Blackwell chips. Nvidia said the mode generates tokens up to eight times faster than Astra's standard mode.
-- The faster replies stay on Nvidia's chips inside OpenAI's service. A household does not get a file.
2026-10-01 / Nvidia / efficiency / central / #nvidia-says-astra-ultrafast-generates-tokens-eig
California Attorney General Rob Bonta said Thursday that his office served an investigative subpoena on OpenAI on Wednesday, as part of an inquiry into cybersecurity incidents and risks involving the company and its models. The office said the subpoena follows its investigation of the Hugging Face incident.
-- The subpoena asks the company that holds the model for answers. It does not move the model onto a home computer.
2026-10-01 / California Attorney General / efficiency / central / #california-attorney-general-subpoenas-openai
Google said Thursday that a prototype satellite for Project Suncatcher, built with Planet, reached orbit on SpaceX's Transporter-18 mission, and that the team has confirmed contact. The company said it will collect data on how its tensor processing units handle the stress, radiation, and heat of space.
-- The test puts Google's chips above the ground. It does not put a model on a kitchen table.
2026-10-01 / Google / chip / central / #google-says-its-test-satellite-is-in-orbit
Axis
local [#########################-----] central
Score: 82
clusters still winning
up 1
Micron has already accounted for 75 percent of its 2027 memory, with high-bandwidth memory for AI servers taking the supply. Nvidia's faster Astra mode runs on Blackwell chips inside OpenAI's API. Google is testing its own chips in orbit. The cluster still holds the memory and the fast model.
Forces
| Force | Meter | Score | Pull | Note | ||
| CHIP / CAPEX |
|
93 | central | Micron said 75 percent of 2027 memory output is already accounted for, and that high-bandwidth memory for 2027 is priced above 2026. Nvidia's faster mode runs on Blackwell chips in a data center, not in a PC a person buys. | ||
| ENERGY |
|
98 | central | No new order today on who pays for data-center power lines. Google's satellite test does not change a home electric bill. | ||
| EFFICIENCY |
|
57 | central | Nvidia's Ultrafast mode is an API setting on Blackwell, up to eight times faster than Astra Standard. The weights are not for sale. California's subpoena asks about incidents. It does not release a file. |
Line charts (0=local 100=central 2026-09-17 to 2026-10-02)
Market overlay (2026-09-03 to 2026-10-02)
Latest: BTC $86,410 / hashrate ~986.17 EH/s / token markers avg $5.3033/M / axis 82. 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 $86,410 / hashrate ~986 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
One email per issue. No tracking pixels, no filler.
home | rss | page 2 | page 3 | archive | models | subscribe | data | board | mcp | llms.txt