Have a lead? Contact us!

Underclassic   rss   archive   buyer's guide

Issue 108 · 2026-11-18

Salt Lake City Votes to Pause New Data Centers for Six Months

Movie-poster art for Issue 108: Megabyte vs. the Zoning Board

Permanent copy of issue 108. Today's issue is always the front page.

Previous issue  |  This week  |  archive

The Salt Lake City Council voted unanimously on Tuesday to pause all new data centers in the city for up to six months, one of five local governments from Utah to Missouri that halted or extended pauses on data centers on Monday and Tuesday. The same evening Nvidia reported $115.8 billion in quarterly revenue and raised its forecast for next year, so the chips for large clusters keep selling even as more towns decide where those clusters may not go.

Date: 2026-11-18
Updated: 2026-11-18 06:30
Status: LIVE


Broadcast

Five local governments voted in two days to keep data centers out while they write rules: Salt Lake City for six months, San Francisco until next September, Rochester, Minn., for a year, San Joaquin County, Calif., for nearly another year, and Springfield, Mo., until February. None of them is reacting to a flood of applications; most have none pending. They are acting on the scale of what is being built elsewhere, and Nvidia showed that scale on Tuesday with $115.8 billion in quarterly sales and a forecast of 75% growth next year. Washington, meanwhile, now wants a license before an overseas cloud rents those chips to Chinese firms. At the other end, SenseTime's new open model needs less than 10 GB of memory, small enough for a household machine.

Watch


Stories

Salt Lake City votes to pause new data centers for six months

The Salt Lake City Council voted unanimously on Tuesday for an ordinance that halts new data centers in the city through mid-May while it writes rules, KSL reports; the measure followed a permit filing for a data center with diesel backup generators in the Northpoint neighborhood on the city's rural west edge. Mayor Erin Mendenhall called it a "healthy pause," saying the city's zoning never considered data centers at the scale now being proposed.
-- By our count in the Data Center Fights Tracker, Salt Lake City, San Francisco, Rochester, Minn., San Joaquin County, Calif., and Springfield, Mo., all voted for new or longer pauses on Monday and Tuesday, and most of them have no pending applications, so builders now face rules written before they arrive.
See also: Data Center Fights Tracker: five pauses voted or extended on Nov. 16-17
2026-11-17 / KSL; Underclassic Data Center Fights Tracker / energy / local / #salt-lake-city-votes-to-pause-new-data-centers-f

Nvidia reports $115.8 billion in quarterly revenue and raises its growth forecast

Nvidia said on Tuesday that revenue for the quarter ended Oct. 25 rose 103% from a year earlier to $115.8 billion, with $107.6 billion from data centers, and forecast about $130 billion for the current quarter without any data center chip sales to China. Interim chief executive Colette Kress, reporting for the first time since Jensen Huang resigned, raised the forecast for revenue growth in the fiscal year to January 2028 to about 75% from 70%.
-- Nvidia's sales are the clearest measure of what the largest companies spend on central AI clusters, and Kress told analysts that cloud backlog has reached $2.5 trillion, up from $2 trillion three months ago.
2026-11-17 / Nvidia / chip / central / #nvidia-reports-115-8-billion-in-quarterly-revenu

San Francisco extends its data center moratorium to September 2027

San Francisco's Board of Supervisors voted unanimously on Tuesday to extend the city's zoning moratorium on new data centers until next September, giving staff 10 months to study them, the San Francisco Standard reports. A Planning Department report last week found that peak electricity transmission across the city is near capacity even without new data centers, and the city has no pending permit applications for them.
-- The city's own planners say new data centers would compete with housing for the same limited grid capacity, which is the trade-off behind most local pauses: a power line used by a cluster is not available for homes.
See also: Data Center Fights Tracker: the fight calendar
2026-11-17 / San Francisco Standard / energy / local / #san-francisco-extends-its-data-center-moratorium

Commerce requires licenses for overseas data centers that rent AI chips to Chinese firms

The Bureau of Industry and Security issued an interim rule on Monday requiring a license to send advanced AI chips to overseas data centers that supply significant computing power to companies headquartered in China and other restricted countries, and adding license requirements for shipments to Indonesia, Malaysia, Singapore, Thailand and Brazil. The rule, which the Federal Register lists for publication on Nov. 19, gives existing data centers serving such customers 90 days to obtain a license.
-- Cloud operators abroad must now check who owns their customers before renting out Nvidia-class chips, a compliance cost that lands on central providers; a model downloaded and run on one's own machine needs no such check.
2026-11-16 / Bureau of Industry and Security / chip / central / #commerce-requires-licenses-for-overseas-data-cen

SenseTime releases a 9-billion-parameter open model that fits in under 10 GB

SenseTime has published the weights of SenseNova Loop-9B, a text-and-image model that reuses a block of its layers twice to get the depth of a larger network, and says its official 4-bit version takes about 4.9 GB and runs in under 10 GB of graphics memory. The company says the model roughly matches the 27-billion-parameter Qwen3.8-27B on agent and data-analysis tests; those scores are its own, and the low-memory version does not yet run in the popular vLLM server.
-- A capable model in under 10 GB fits on an ordinary gaming graphics card or a recent laptop, so coding and data analysis that now carry a per-token bill could run at home for the price of electricity.
See also: Open-Source AI Guide: what runs on the computer you already own
2026-11-17 / SenseTime / efficiency / local / #sensetime-releases-a-9-billion-parameter-open-mo


Axis
local [########################------] central
Score: 79   clusters still winning   0
Wednesday's edition leads with Salt Lake City's six-month pause on data centers, one of five local pauses voted or extended in two days, alongside Nvidia's record $115.8 billion quarter, a new Commerce rule on renting AI chips abroad, and SenseTime's small open model. Local resistance and a cheaper home model balance a stronger-than-expected Nvidia, so the score holds at 79.

Forces

ForceMeterScorePullNote
CHIP / CAPEX
..
97 central Nvidia reported $115.8 billion in quarterly revenue, up 103% from a year earlier, and raised next year's growth forecast to about 75%, while Commerce now requires licenses for overseas data centers serving Chinese firms. Demand outran the new rules, so the gauge rises one point, to 97.
ENERGY
..
95 central Salt Lake City, San Francisco, Rochester, San Joaquin County and Springfield, Mo., all voted for new or longer pauses on data centers on Monday and Tuesday, and San Francisco's planners say the city's grid is near capacity. The pauses slow local siting but not national demand, so the gauge holds at 95.
EFFICIENCY
..
88 local SenseTime released SenseNova Loop-9B, an open model whose 4-bit version needs under 10 GB of graphics memory and which the company says matches a 27-billion-parameter rival. The scores are self-reported, so the gauge holds at 88.

Line charts (0=local 100=central 2026-11-03 to 2026-11-18)

Axis score line chart

Forces line chart

Market overlay (2026-10-20 to 2026-11-18)

Latest: BTC $64,091 / hashrate ~982.28 EH/s / token markers avg $5.2583/M / axis 79. These lines are different units. We compare timing, not dollars-to-hashrate.

Market overlay line chart

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 $64,091 / hashrate ~982 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 2026-11-15; blended = (3*input + 1*output) / 4; -- = open/self-host or no single public list)

#ModelOrgCCWeightsTierInOutBlend
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

Full 54-model board


If useful, support the broadcast.
bc1qfs3yw5qlq8sxs50crzh3g84ug27gvww9npu85n
BTC QR

Get the daily issue

One email a day. Unsubscribe anytime.

home  |  rss  |  page 2  |  page 3  |  archive  |  models  |  buyer's guide  |  subscribe  |  data  |  board  |  about  |  mcp  |  llms.txt