🧠 Food for thought

Implications, context, and why it matters.

Blackstone is tackling the energy bottleneck for AI

• The venture centers on power supply alongside chip deployment.
• New data centers often stall at the grid. In some markets, new connections take 7 to 10 years 1.
• Blackstone formed a joint venture with PPL Corporation, a US utility company, to build, own and run new gas-fired combined-cycle plants for Pennsylvania data centers under long-term energy services agreements (ESAs) 2.
• Demand is large. More than 13 gigawatts of data center projects are in advanced planning inside PPL Electric Utilities’ Pennsylvania service area 2.
• The deal adds to Blackstone’s US$85 billion global data center platform and links digital infrastructure with power investments for the long term 1.

The AI cloud is being unbundled for specialized workloads

• The venture lays out a different model for AI infrastructure beyond hyperscalers such as Amazon, Microsoft and Google.
• For Google, it opens a dedicated route for custom hardware such as Ironwood 3. Ironwood is its seventh-generation Tensor Processing Unit (TPU) built for inference, the process of running AI models to generate results for users 3.
• That differs from the multi-cloud setup used by Anthropic, an AI startup. It spreads workloads across Google TPUs, Amazon Trainium chips and Nvidia graphics processing units (GPUs) to manage costs and limit lock-in 4.
• The new company will likely focus on inference services, a large ongoing expense in AI, rather than model training 3.
• The market has reached a stage where specialized, asset-heavy providers can challenge large cloud companies in selected parts of the stack.