Underclassic rss archive buyer's guide
Permanent copy of issue 70. Today's issue is always the front page.
Previous issue | This week | Next issue | archive
Lumentum said Friday that its optical parts for AI data centers are sold out until early 2029, and that it cannot meet about 70 percent of demand for some products next year. Broadcom is in early talks to arrange tens of billions of dollars of debt so OpenAI can buy the custom chips the two companies are designing. A Senate investigation found that none of seven large data center developers would agree to pay for grid capacity built only because of them. JetBrains released Mellum2.1, an open coding model built to run on a developer's own hardware.
Date: 2026-10-09
Updated: 2026-10-09 06:34
Status: LIVE
Broadcast
The scarce parts of an AI cluster are no longer only the GPUs: the lasers that connect them are spoken for until 2029, and the chips themselves are increasingly bought with borrowed money. Washington's question is closer to home, namely whose electric bill pays for the plant built to serve a server campus. On the desk, small open models keep getting better at the work programmers do every day.
Watch
Stories
Lumentum chief executive Michael Hurlston told Bloomberg in Tokyo on Friday that the company's optical components for AI data centers are completely sold out through early 2029. He said Lumentum cannot meet about 70 percent of demand for some products through next year and about 30 percent for others through 2028, six months after he expected capacity to sell out only through 2028.
-- The lasers that carry data between AI chips are now booked years ahead by the companies building the largest clusters, and Nvidia put $2 billion into Lumentum this year. A newcomer ordering optical links today waits behind clusters that ordered long ago.
2026-10-09 / The Japan Times (Bloomberg) / chip / central / #lumentum-says-its-ai-optical-parts-are-sold-out-
Broadcom has held early talks with lenders to raise debt that would help OpenAI buy the custom AI chips the two companies are developing, according to people familiar with the matter. Bloomberg put the early figure at about $30 billion and The Wall Street Journal at more than $50 billion, weeks after Broadcom launched a $60 billion debt package for Anthropic.
-- The chip seller is now arranging the loans its largest customers use to buy its chips. Labs that cannot yet pay for hardware out of revenue are borrowing against future compute, a cost only the biggest cluster builders can carry.
2026-10-08 / Bloomberg Law / chip / central / #broadcom-seeks-billions-in-debt-to-fund-openai-s
A yearlong investigation led by Senators Elizabeth Warren, Chris Van Hollen and Richard Blumenthal found that none of seven developers, among them Amazon, Google, Meta, Microsoft and CoreWeave, would agree to pay for power infrastructure that would not be needed without their data centers. The report, first shared with TIME, says some companies count roughly one permanent worker per megawatt and that sales-tax breaks on computer equipment are their most valuable incentive.
-- By the report's arithmetic, a 100-megawatt data center can use about as much electricity as 100,000 homes while employing about 100 people. The companies told investigators that new plants and lines benefit everyone, which leaves state regulators to decide whose bill pays for them.
2026-10-08 / TIME / energy / central / #senate-inquiry-finds-data-center-builders-refuse
JetBrains released Mellum2.1, a 12-billion-parameter mixture-of-experts model that uses 2.5 billion parameters per token, on Hugging Face under the Apache 2.0 license. The company says a summer of reinforcement learning taught it to explore a repository, edit files and check its own changes, and that builds for llama.cpp, Ollama and LM Studio are coming soon.
-- A coding assistant this small is built to run on the developer's own machine, so the code and the bill stay at home. JetBrains says it beats similar small open models on its tests, though the benchmarks are the company's own.
2026-10-08 / JetBrains / efficiency / local / #jetbrains-releases-mellum2-1-an-open-coding-mode
Axis
local [##########################----] central
Score: 85
clusters still winning
up 1
Lumentum's sold-out order book and Broadcom's debt talks for OpenAI show the cluster buying its parts years ahead, increasingly with borrowed money. The Senate report shows the same builders resisting the bill for the power plants they require. JetBrains' Mellum2.1 keeps a slice of programming work on local hardware. The money and the supply chain still run to the cluster.
Forces
| Force | Meter | Score | Pull | Note | ||
| CHIP / CAPEX |
|
95 | central | Lumentum says its optical components are sold out through early 2029 and that it cannot meet about 70 percent of demand for some products through 2027 and about 30 percent for others through 2028. Broadcom is in early talks to raise debt for OpenAI's custom chips: about $30 billion according to Bloomberg, more than $50 billion according to The Wall Street Journal. | ||
| ENERGY |
|
99 | central | A Senate staff report led by Senators Warren, Van Hollen and Blumenthal found none of seven developers would accept a standard requiring them to pay for power infrastructure needed only because of their data centers; some counted roughly one permanent worker per megawatt. | ||
| EFFICIENCY |
|
58 | mixed | JetBrains released Mellum2.1, a 12-billion-parameter mixture-of-experts model with 2.5 billion active parameters, under Apache 2.0; JetBrains says multi-token prediction makes single requests about 1.6 times faster. |
Line charts (0=local 100=central 2026-09-24 to 2026-10-09)
Market overlay (2026-09-10 to 2026-10-09)
Latest: BTC $82,505 / hashrate ~979.07 EH/s / token markers avg $5.2583/M / axis 85. 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 $82,505 / hashrate ~979 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 2026-10-08; 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 | buyer's guide | subscribe | data | board | mcp | llms.txt