Permanent copy of issue 48. Today's issue is always the front page.
Anthropic signed a lease for its first Australian data centre at a $31.9 billion campus in western Queensland, Premier David Crisafulli told parliament on Wednesday. Nvidia said its Vera Rubin NVL72 rack posted up to 3.7 times the inference throughput of last year's GB300 NVL72 on MLPerf Inference v6.1. Reports say Apple is building an enterprise AI server with two or four future M8 Ultra chips, aimed at 2029. Mozilla put Mistral Small 4 into Firefox Smart Window beta in the United States, Canada, and France.
Date: 2026-09-17
Updated: 2026-09-17 06:12
Status: LIVE
Broadcast
A $31.9 billion computer hall on Queensland farmland still draws from the same grid a household pays into. Nvidia's new rack does not shrink the hall; it fills it with more tokens per watt of rented power. Apple's reported server would take the chip that already sits on a Mac Studio and stack it in a company closet three years from now. Firefox can switch the model behind a search window, and the conversation still leaves the desk.
Watch
Stories
Anthropic signed a lease for its first Australian data centre at the $31.9 billion Western Downs digital park near Dalby, Queensland Premier David Crisafulli told parliament on Wednesday. Singapore-based Zerra DC is to build the campus, which will connect to the Braemar power station, with a first stage planned to open next year; the local council has not yet approved the project.
-- A $31.9 billion hall on farmland still bills Claude to the grid a Queensland household shares.
2026-09-16 / The Guardian / energy / central / #anthropic-leases-queensland-data-centre
Nvidia said Wednesday that Vera Rubin NVL72, in its first MLPerf Inference v6.1 preview, delivered up to 3.7 times the throughput of GB300 NVL72 on Qwen3-VL and up to 2.5 times on DeepSeek-R1. A four-rack GB300 NVL72 submission with 288 GPUs reached 99 percent scaling efficiency, Nvidia said, citing MLCommons entries dated 16 September.
-- A rack that prints 3.7 times last year's tokens is more cluster in the same floor space, not a model that comes home.
2026-09-16 / Nvidia / chip / central / #nvidia-rack-leads-mlperf-inference-test
Apple is working on an enterprise AI server with two or four future M8 Ultra chips and a planned 2029 debut, Ars Technica reported Wednesday, citing The Information. The Information said the project has backing from new chief executive John Ternus, that Apple has talked with Nvidia about NVLink Fusion to connect the chips, and that the effort could still be cancelled.
-- The chip that already runs on a Mac Studio would, in this plan, sit in a company rack three years from now.
2026-09-16 / Ars Technica / chip / mixed / #apple-plans-m8-ultra-ai-server
Mozilla said Wednesday that Mistral Small 4 is coming to Firefox Smart Window beta for users in the United States and Canada, and that the beta is expanding to France with French-language support. Mistral said conversations are not saved on Mozilla's servers by default and that partners including Mistral agree to zero data retention; users can still pick other models.
-- Switching the model in a browser is a choice of landlord, not a copy that lives on the phone.
2026-09-16 / Mozilla / efficiency / mixed / #mozilla-puts-mistral-in-firefox
Axis
local [#####################---------] central
Score: 70
clusters still winning
up 1
Anthropic signed a $32 billion Queensland lease. Nvidia posted a 3.7 times jump on a rack-scale inference test. Reports say Apple wants an M8 Ultra server in 2029. Mozilla put a French open-weight model in Firefox. The cluster still holds the power and the training run.
Forces
| Force | Meter | Score | Pull | Note | ||
| CHIP / CAPEX |
|
87 | central | Nvidia said Vera Rubin NVL72 posted up to 3.7 times GB300 throughput on MLPerf Inference v6.1. | ||
| ENERGY |
|
92 | central | Anthropic leased a $31.9 billion campus in western Queensland that will connect to Braemar power station. | ||
| EFFICIENCY |
|
48 | local | Mozilla put Mistral Small 4 in Firefox with zero data retention at the partner; the model still runs off-device. |
Line charts (0=local 100=central 2026-09-01 to 2026-09-17)
Market overlay (2026-08-19 to 2026-09-17)
Latest: BTC $76,298 / hashrate ~953.86 EH/s / token markers avg $5.3033/M / axis 70. 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 $76,298 / hashrate ~954 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