Firmus Seeks $10 Billion as AI Model Prices Fall

A giant chip-financing plan and cheaper frontier models show AI competition shifting toward capital and operating costs.

🗄️ Firmus Seeks $10 Billion for Nvidia Chips

Decoded: Australian startup Firmus Technologies is in talks to raise roughly $10 billion to buy Nvidia (NVDA) chips for an Indonesian data center, Bloomberg reported on Sep. 23. The plan comprises about $7.5 billion of senior and mezzanine debt plus $2.5 billion of equity, with a possible five-year tenor. Bloomberg, Sep. 23

Why it matters: The proposed financing would test whether lenders will underwrite one of Asia's largest AI infrastructure deals before Firmus pursues an Australian IPO. For Nvidia, customer access to debt and equity is becoming as important as chip demand because financing capacity determines how quickly orders become installed compute.

🤖 OpenAI and Anthropic Cut the Cost of New Models

Decoded: OpenAI said GPT-6 Sol and Luna cost half as much as their predecessors, with API prices of $2 and $10 per million input and output tokens for Sol and $0.10 and $0.50 for Luna. Anthropic priced Opus 5.5 at $4 per million input tokens and $20 per million output tokens, 20% below Opus 5, while saying cache reads cost 60% less and output is more than 30% faster. Ars Technica, Sep. 22

Why it matters: Lower inference prices can broaden enterprise usage on platforms tied to Microsoft (MSFT) and Amazon (AMZN), but they also put pressure on model-provider margins. Investors should watch whether cheaper tokens produce enough additional volume to offset falling unit prices.

Stay decoded. See you tomorrow.

— The Get AI Decoded Team