Permanent copy of issue 55. Today's issue is always the front page.
Microsoft said it plans to invest more than $10 billion in capital and operating expenses across Kuwait, Qatar, Saudi Arabia, and the United Arab Emirates through 2030, including more cloud and AI infrastructure. Crusoe said Thinking Machines Lab will pay $65 million a year to serve open models on Nvidia B200 systems that Crusoe runs. Nvidia published a 100-million-parameter open model that labels who spoke in a recording, and a household can run that file on its own machine. Google put two new speech models on the Gemini API and did not state a price.
Date: 2026-09-24
Updated: 2026-09-24 06:07
Status: LIVE
Broadcast
A household in Kuwait, Qatar, Saudi Arabia, or the United Arab Emirates does not receive a share of Microsoft's $10 billion. If the halls are built, that household gets a login and a bill. The file a person can keep from Wednesday is Nvidia's speaker labeler, which can say who talked in a recording of up to eight people. It is not the Nvidia cluster Thinking Machines agreed to rent for $65 million a year.
Watch
Stories
Microsoft said it plans to invest more than $10 billion in capital and operating expenses across Kuwait, Qatar, Saudi Arabia, and the United Arab Emirates between now and 2030. The company said the plan includes more cloud and AI infrastructure, and set it out at a round table on the sidelines of the United Nations General Assembly in New York, with government representatives from those four countries.
-- The sum is a corporate plan through 2030, not a computer a household in those countries can take home.
2026-09-24 / Microsoft / energy, chip / central / #microsoft-plans-more-than-10-billion-in-the-midd
Crusoe said Wednesday that Thinking Machines Lab signed a $65 million a year agreement to run open models, including Inkling and GLM 5.2 and 5.3, on a dedicated set of Nvidia HGX B200 systems. Crusoe said that managed inference business is now past $100 million in contracted annual revenue, less than a year after launch.
-- The weights are open. The $65 million bill is still for Nvidia racks the lab does not own.
2026-09-23 / Crusoe / chip / mixed / #thinking-machines-signs-65-million-crusoe-infere
Nvidia published Nemotron 3 Diarization on Wednesday, an open-weight model of 100 million parameters that labels which of up to eight speakers talked, and when. Nvidia said the same file can run on a local machine, and that it is ready for commercial use.
-- A household can run the file on its own machine, but it only labels a meeting, and it does not replace a paid chat model.
2026-09-23 / Nvidia / chip, efficiency / local / #nvidia-releases-open-model-that-labels-speakers
Google said Wednesday that Gemini 3.8 Flash text-to-speech and a lighter Flash-Lite version are in the Gemini API and Google AI Studio, and in Gemini Notebook. Google said the larger model can make a new voice from a written description in more than 100 languages, and the post did not state a price per million tokens.
-- A household that wants the voice still pays Google, because the post did not hand over a file to keep.
2026-09-23 / Google / efficiency / central / #google-puts-new-speech-models-on-the-gemini-api
Axis
local [######################--------] central
Score: 74
clusters still winning
up 1
Microsoft said it plans to invest more than $10 billion across Kuwait, Qatar, Saudi Arabia, and the United Arab Emirates through 2030. Thinking Machines will pay Crusoe $65 million a year for Nvidia B200 racks. Nvidia published a small open model a household can run. Google put speech on its own API. The cluster still holds the contract and the meter.
Forces
| Force | Meter | Score | Pull | Note | ||
| CHIP / CAPEX |
|
89 | central | Thinking Machines rented Nvidia B200 racks from Crusoe for $65 million a year. Nvidia also published a 100-million-parameter open model a desk can run. | ||
| ENERGY |
|
95 | central | Microsoft plans more than $10 billion in capital and operating expenses across four Middle East countries through 2030. The post names dollars, not megawatts. | ||
| EFFICIENCY |
|
52 | mixed | Google put speech models on the Gemini API and stated no token price. Nvidia's new open model has no meter. |
Line charts (0=local 100=central 2026-09-09 to 2026-09-24)
Market overlay (2026-08-26 to 2026-09-24)
Latest: BTC $83,150 / hashrate ~920.47 EH/s / token markers avg $5.3033/M / axis 74. 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 $83,150 / hashrate ~920 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 | page 3 | archive | models | subscribe | data | board | mcp | llms.txt