In response to the growing demand for AI infrastructure, major tech companies are increasing their investments. Meta plans to spend over $60 billion by 2025 to enhance its AI capabilities, while a new joint venture called Stargate, involving OpenAI, Oracle, and SoftBank, aims to invest up to $500 billion in US AI infrastructure. While some expect inference costs to eventually decrease due to increased competition, this is likely only for smaller-scale systems, with large-scale services facing higher expenses.
Key takeaways:
- DeepSeek's latest AI model outperformed those from OpenAI and Meta on third-party benchmarks, causing panic among US AI companies.
- DeepSeek offers significantly cheaper pricing for its models compared to competitors, with its R1 model costing $0.55 per million tokens versus OpenAI's $15.
- Meta's chief AI scientist, Yann LeCun, argues that the market's reaction to DeepSeek is unjustified, emphasizing that inference costs will rise as AI tools become more sophisticated.
- Major investments in AI infrastructure are being made, with Meta planning over $60 billion in capital expenditures for 2025 and a joint venture, Stargate, funneling up to $500 billion in AI infrastructure across the US.