Underclassic rss archive buyer's guide
Permanent copy of issue 84. Today's issue is always the front page.
Previous issue | This week | Next issue | archive
Britain has 315 hyperscale data centres waiting to connect to its national grid, and together they would need 73 gigawatts, far more than the 45 gigawatts the country draws on a cold winter day, the Guardian reported on Saturday. In Taiwan, Nvidia, Intel and AMD have reportedly paid about $5 billion in advance to secure the substrates their AI chips are mounted on. On the local side, Nvidia released an open model that runs a voice assistant in about 4 gigabytes of memory.
Date: 2026-10-25
Updated: 2026-10-25 06:30
Status: LIVE
Broadcast
The wait for central compute now runs through planning offices and substrate factories. In Britain the demand queued for the grid is larger than the country's own winter peak, and residents in Wales are fighting the gas turbines meant to bridge the wait. In Taiwan the largest chip buyers are paying in advance for factory capacity that is still being built. Nvidia's small model is the counterweight: a capable assistant that fits on a gaming card and needs no data center at all.
Watch
Stories
The Guardian reported on Saturday that 315 hyperscale data centres are waiting to connect to Britain's national grid and would together need 73 gigawatts, 28 gigawatts more than the typical winter peak of 45 gigawatts. More than 100 planned UK sites intend to run on gas turbines while they wait, including a proposed Cardiff data centre that the Welsh government has ordered the city council to put on hold.
-- When the queue for power is larger than the whole country's peak demand, the developers who can pay for their own turbines go first and everyone else waits. The fights in Newport and Cardiff show that the cost of central compute is now being argued street by street, in noise, fumes and lost bus services.
2026-10-24 / The Guardian / energy / central / #britain-s-queued-data-centres-seek-73-gigawatts-
Taiwan's Economic Daily News reported on Saturday that the three chipmakers have together paid more than NT$100 billion, about $5 billion, in deposits to Taiwanese makers of the high-end substrates that connect processors to circuit boards. Unimicron, Kinsus and Nan Ya PCB have announced a record NT$161.2 billion of new plant spending this year; the chipmakers have not confirmed the individual amounts.
-- Buyers paying years ahead for a part as plain as a substrate shows how scarce the components behind AI chips have become. Capacity reserved with cash goes first to the largest customers, and the same substrates also carry the processors in workstations and personal computers.
2026-10-24 / GMT Eight (citing Economic Daily News) / chip / central / #nvidia-intel-and-amd-prepay-about-5-billion-for-
Nvidia published Nemotron 3.5 Bolt E4B 4GB on Friday, an open small language model built on Google's Gemma 4 E4B and tuned for voice assistants, game characters and robots. Nvidia says it fits in about 4 gigabytes, scores roughly 40 percent higher than its previous small model on conversational agent tests, and generates about 133 tokens a second on an RTX 5070 graphics card.
-- A model that answers quickly on a consumer graphics card or an 8-gigabyte Jetson board keeps the conversation on the device instead of sending it to a data center. It also suits Nvidia, which sells the local hardware as well as the cluster chips.
2026-10-23 / NVIDIA Research / efficiency / local / #nvidia-releases-an-open-model-that-runs-a-voice-
Axis
local [#########################-----] central
Score: 82
clusters still winning
unchanged
Britain's 73-gigawatt queue and the reported $5 billion of substrate deposits both show capacity going to whoever can pay first and wait longest. Nvidia's four-gigabyte model shows how much a small device can now do, but it does not change who holds the capacity, so the score holds at 82.
Forces
| Force | Meter | Score | Pull | Note | ||
| CHIP / CAPEX |
|
96 | central | Nvidia, Intel and AMD have reportedly prepaid about $5 billion to Taiwanese substrate makers, and Unimicron, Kinsus and Nan Ya PCB have announced a record NT$161.2 billion of plant spending this year. | ||
| ENERGY |
|
98 | central | The Guardian reports that 315 hyperscale data centres waiting on Britain's grid would need 73 gigawatts, against a typical winter peak of 45 gigawatts, and more than 100 planned UK sites intend to use gas turbines while they wait. | ||
| EFFICIENCY |
|
72 | mixed | Nvidia's Nemotron 3.5 Bolt E4B runs in about 4 gigabytes at roughly 133 tokens a second on an RTX 5070, and Nvidia says it produces up to 80 percent fewer tokens than comparable models. |
Line charts (0=local 100=central 2026-10-08 to 2026-10-25)
Market overlay (2026-09-26 to 2026-10-25)
Latest: BTC $80,305 / hashrate ~1068.31 EH/s / token markers avg $5.2583/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 $80,305 / hashrate ~1068 EH/s. List-price gap (not a subsidy %): 123.1x list gap (blended) Mercury 3 -> Claude Fable 5.1 ($0.16/M to $20.00/M). (mempool.space + Coinbase/CoinGecko)
Marker prices (USD per 1M tokens; verified 2026-10-22; 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