Underclassic   rss   archive

Issue 39: OpenAI Books Malaysian Compute

Permanent copy of issue 39. Today's issue is always the front page.

OpenAI signed a multi-year deal with Nvidia-backed Firmus for dedicated compute at two Malaysian AI factories, lifting Firmus contracted capacity above 900 MW. Anthropic has $517 billion of compute leases covering 14.8 GW on a 10-year book, according to a tally The Information compiled. Mistral raised 3 billion euros at more than 21 billion euros to fund open-weight models and the factories they run on.

Date: 2026-09-08
Updated: 2026-09-08 06:25
Status: LIVE


Broadcast

Sachin Katti, OpenAI's VP of compute strategy, said the Malaysian sites will serve demand in the region and around the world. Firmus will put Nvidia Vera Rubin NVL72 racks on its HyperCube platform, prefabricated in New South Wales. Anthropic's largest slice is an 11 GW Amazon and Google arrangement worth more than $300 billion. Mistral said it is building open-weight models, the compute they run on, and the products on top so customers are not locked to one vendor's roadmap.

Watch


Stories

OpenAI signs Malaysian compute deal

OpenAI signed a multi-year deal with Nvidia-backed Firmus for dedicated compute at two AI factories in Malaysia, making OpenAI an anchor customer. Firmus said the pact lifts its contracted capacity across customers above 900 MW, with Vera Rubin NVL72 racks on its HyperCube platform.
Asia-Pacific power and Rubin racks are being reserved for one lab's API, not for a desk.
2026-09-08 / Firmus / chip, energy / central

Mistral raises 3 billion euros

Mistral raised 3 billion euros in a Series D at a post-money valuation of more than 21 billion euros, led by Samsung Electronics with the EU-backed Scaleup Europe Fund and PSG Equity as co-leads. The Paris lab said the money will expand frontier research, scale training compute, and grow infrastructure around open-weight models.
Open weights are the local pole. The round still pays for factories the lab will own.
2026-09-08 / Mistral / efficiency, chip / mixed

Anthropic locks 14.8 GW of compute

The Information tallied Anthropic compute leases at $517 billion over 11 months, equal to 14.8 GW of capacity to be spent over 10 years, Seoul Economic Daily reported. The largest slice is an 11 GW Amazon and Google arrangement worth more than $300 billion; before last October the lab had 1 to 2 GW.
A decade of leased gigawatts is the IPO collateral. It is not a local runtime.
2026-09-07 / Seoul Economic Daily / chip, energy / central


Axis
local [####################----------] central
Score: 66   clusters still winning   up 1
OpenAI is booking Vera Rubin racks in Malaysia. Anthropic has 14.8 GW of leased compute on a decade-long book. Mistral raised to fund open weights and its own factories. The spend still sits in the cluster.

Forces

ForceMeterScorePullNote
CHIP / CAPEX
..
84 central OpenAI is booking Nvidia Vera Rubin NVL72 capacity at two Firmus sites in Malaysia. Anthropic added 14.8 GW of leased compute in 11 months.
ENERGY
..
87 central Firmus contracted capacity now exceeds 900 MW. Anthropic's 14.8 GW book is a decade of power, not a desk.
EFFICIENCY
..
45 local Mistral is still shipping open weights and selling the stack as something customers can run themselves. The new money also funds factories.

Line charts (0=local 100=central 2026-08-23 to 2026-09-08)

Axis score line chart

Forces line chart

Market overlay (2026-08-10 to 2026-09-08)

Latest: BTC $78,690 / hashrate ~928.83 EH/s / token markers avg $5.3033/M / axis 66. 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 $78,690 / hashrate ~929 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)

#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 Flash DeepSeek CN open
~284B total / ~13B active (MoE)
commodity $0.44 $1.32 $0.66

Full 20-model board


If useful, support the broadcast.
bc1qfs3yw5qlq8sxs50crzh3g84ug27gvww9npu85n
BTC QR

Previous issue  |  This week  |  Next issue  |  archive

home  |  rss  |  archive  |  models  |  data  |  board  |  mcp  |  llms.txt