In a world where the data center capital of the world slams on the brakes… one county dares to say "not yet."
Now showing: FROM LOUDOUN WITH LOVE.
⏸️ Loudoun County, VA voted 7-1 to stop acting on new data center applications until stricter rules land, expected July 2027. 33 applications are pending.
Why it matters: The densest data center market on Earth is slowing down, even though data centers are projected to fund 40% of its 2027 budget.
⚡ @pjminterconnect asked FERC to make 50 MW+ data centers ride through grid faults, after 3.9 GW of them switched to backup power at once in July.
Why it matters: The cost of keeping giant customers from rattling the grid moves onto the data centers themselves.
🧠 @TencentHunyuan released Hy4 Flash as open weights: ~220B parameters, 12B active, 1M-token context, $0.07 per million input tokens on its API.
Why it matters: Anyone can host it, and a GGUF build is already linked for llama.cpp-style tools.
🌲 Hillsboro, Oregon extended its data center moratorium by six months.
Why it matters: Another major hub is buying time to rewrite its rules as residents push back on tax breaks.
🍡 Marin published all 57 pre-training checkpoints of its 535B Mochi model.
Why it matters: Big labs never show the in-between steps. Now outside researchers can study how a model this size is built.
📍 Axis: 78 (from 79). The biggest data center market slowing approvals, plus two big open releases, nudge the needle one point toward local.
Should every data center county pause until it writes its rules, or does that just push the builds to the next county over?
Full issue: underclassic.com/issues/109.…
1
26
Hugo Eilenberg retweeted
Responding in my personal capacity, but wanted to share my perspective..
We are in a great-power competition with so much at stake in the years ahead. There is no one in America who has lived during a time when the United States wasn't the preeminent superpower and, despite our imperfections, a force for good in the world. A nation that fought and helped end wars on behalf of those who couldn't, pioneered powered flight, put humanity on the Moon, led revolutions in medicine and technology that improved billions of lives and distributed hundreds of billions in aid around the world.
I don't know how handicapping this nation by going after one of its largest employers, a company that has created wealth for millions, ushered in electric cars, rockets that don't need to be thrown away, satellites that connect the disconnected and have unquestionably saved lives, not to mention access to information that is pretty foundational to solving many of the world's problems, biotech to restore function to those with disabilities, and so much more, serves America's interests. Without the contributions of these companies, there is little question America would be behind in a competition where the outcome is far from certain and the consequence so great.
Now, maybe I am just a bit new to this, but from the billboards, TV and digital ads, spam texts, and fundraising calls, it would seem both parties have no issue raising money, including from billionaires, with the aim of influencing and buying votes. And if that means campaign reform, maybe include dialing back the spam calls and texts..But with all that is going on in the world, and so much at stake for our country, I question the priorities reflected in the post below.
Voters sided with limits in 46 of 47 local data center votes on Nov 3. Only exception: Harwood, N.D.
This week: Salt Lake City paused new ones for 180 days, Rochester, Minn. for a year, San Francisco extended its moratorium.
@ballotpedia @slcCouncil
underclassic.com/data-center…
1
28
92 people signed a statement saying "Anthropic is not a safe AI company," hours before its IPO prices.
Signers: @Thom_Wolf @ibab @So8res @beffjezos @tegmark. Organized by @CAIS. No reply yet from @AnthropicAI.
Who signed vs. the record: underclassic.com/misanthropi…
via @MTSlive
3
83
Wolf signed but said he wouldn't have used "reckless." Soares and Beff Jezos signed too, on opposite sides of how fast AI should go.
Is a lab's own safety policy enough, or does it need outside oversight?
1
Sources:
Statement + signers: misanthropicstatement.com/
CAIS release: safe.ai/news/anthropic-is-no…
Anthropic RSP v3.0: anthropic.com/news/responsib…
GovAI analysis: governance.ai/analysis/anthr…
IPO: cnbc.com/2026/11/18/anthropi…
50
Hugo Eilenberg retweeted
Replying to @mkraju
The American people deserve to know how their representatives vote.
There should be no secret voting by our employees in a democracy.
50K tokens in, 5K out: a 7x price spread.
Claude Fable 5.2: $0.60
GPT-6.2 Astra: $0.60
Gemini 4 Carbon: $0.15*
Grok 4.7: $0.13
Qwen4-Max: $0.098
DeepSeek V4 Pro: $0.086
*Intro price, doubles Dec 1
@AnthropicAI @OpenAI @GoogleDeepMind @grok
underclassic.com/models.html
2
136
Fable 5.2 is #1 on @ArtificialAnlys' Intelligence Index at 65, ahead of Gemini 4 Carbon at 60. OpenAI lists GPT-6.2 Astra at $8/$40; the rumored $4/$20 is its Batch price. Gemini 4 Neon isn't on Google's price page yet; a Cloud SKU listing shows $3/$15 (unconfirmed).
247
The US is weighing a "name and shame" label for firms using Chinese open-weight AI, per @Reuters. Not a rule yet.
• Moolenaar: risk remains even on your own hardware
• Same job: $0.60 closed US API, $0.09 open DeepSeek
Our take: a push to centralize AI.
underclassic.com/us-companie…
1
49
The government's stated worry is security, and @RepMoolenaar says risk remains even when firms run these models on their own infrastructure.
Open weights can run offline, stay on your hardware and be tested by anyone.
Should running open weights on your own machine ever need a label?
20
Same 128GB of memory, from $3,449 to $6,950:
Framework Desktop (Ryzen AI Max+ 395): $3,449, 256 GB/s
Mac Studio M5 Max: $5,399, 614 GB/s
DGX Spark: $6,950, 273 GB/s
Mac: 2.2x the Spark's bandwidth for $1,551 less.
@FrameworkPuter @Apple @nvidia
underclassic.com/home-ai-com…
1
91
Bandwidth isn't the whole story. @seapy tested DeepSeek V4.1 Flash on a Mac Studio 512GB vs four ASUS GX10 (128GB each), Mac first: 1 user, 53 vs 73 tok/s; 16 users, 137 vs 330; 128K prefill, 3,325 vs 7,522. Which would you buy? @CerebralCoding_ @niklaslenz_ai
53
In a world where five towns hit pause in two days… one industry learns the meaning of the word "no."
Now showing: Megabyte vs. the Zoning Board.
🏔️ Salt Lake City votes unanimously to pause new data centers for up to six months.
Why it matters: It's one of five local governments, from Utah to Missouri, that voted for new or longer pauses on Monday and Tuesday.
🌉 San Francisco extends its data center moratorium to September 2027.
Why it matters: The city's own planners say the grid is near capacity even before any data centers are added. That leaves housing and clusters competing for the same power lines.
💰 Nvidia posts $115.8B in quarterly revenue (up 103%) and raises next year's growth forecast to about 75%.
Why it matters: Spending on central AI clusters keeps growing even as towns decide where those clusters can't go.
🛂 Commerce now requires a license before overseas data centers rent advanced AI chips to Chinese firms.
Why it matters: Cloud providers pick up a new compliance cost. A model running on your own machine doesn't need any check.
💻 SenseTime releases SenseNova Loop-9B, an open model whose 4-bit version runs in under 10 GB.
Why it matters: It fits on a gaming GPU or a recent laptop (benchmarks are self-reported).
Axis: 79. Clusters are still winning, but local pushback and cheaper home models are holding the line.
Would your town vote to hit pause?
Full issue in the reply 👇
1
69
Today's issue: underclassic.com/issues/108.…
14
Mesh LLMs explained!
Builders: @Mesh_LLM @exolabs @Gradient_HQ @darkbloomai @chutes_ai
Works today on your own machines. Unsolved: proving a stranger ran the real model.
underclassic.com/mesh-llms.h…
1
41
Sources: Mesh LLM's own roadmap lists proof of inference as open work: github.com/Mesh-LLM/mesh-llm…
Darkbloom's docs say its coordinator and providers see plaintext: docs.darkbloom.dev/security/…
exo: github.com/exo-explore/exo
1
23
Peers have already trained models over the internet: Covenant-72B (@covenant_ai) arxiv.org/abs/2603.08163, Pluralis Agora (@Pluralis) arxiv.org/abs/2607.13332, Hermes 4.3 on Psyche (@NousResearch). Serving is the harder part. Full table: underclassic.com/mesh-llms.h…
11











