Major Shifts Across AI Infrastructure Funding, Energy, and Technology
인프라/플랫폼 | Sat Aug 15 2026 00:00:00 GMT+0000 (Coordinated Universal Time) | 7 sources
From Nvidia's $500 billion data center plan to hyperscaler bets on natural gas and Databricks' massive funding round, the AI infrastructure landscape is undergoing dramatic transformation.
Analysis
[Nvidia] announced $500 billion AI data center financing plan [5]
- Guarantees up to 25% value preservation for aging GPUs
- Apollo
- BlackRock
- Blackstone
- Brookfield
- Goldman Sachs
- and KKR participating
- Aims to build secondhand GPU distribution ecosystem
- 'Wrong way' risk controversy
[Databricks] raised $5 billion at $190 billion valuation [4]
- Initially targeted $1 billion but investor demand surged to $15 billion
- Led by Coatue
- with Blackstone
- MGX
- and T. Rowe Price participating
- Maintains $7 billion annualized revenue with 80% growth rate
- Funds to be invested in 100-person AI research team and hyperscaler cloud contracts
[Hyperscalers (Amazon, Google, Meta, Microsoft)] massive natural gas plant bets raise price surge risks [3]
- Meta planning 7.5GW in Louisiana
- Amazon 7.6GW gas plants in Texas
- Microsoft and Google also building gigawatt-scale plants in Texas
- Gas prices projected to triple in some regions (Noreva report)
- Fuel costs make up half of power costs
- potentially raising token prices
[Kog] developed inference engine maximizing existing GPU performance [2]
- Demonstrated 3
- 000 tokens per second on AMD MI300X and Nvidia H200
- Targets 30x improvement in LLM inference speed
- Open-sourced 2B parameter Laneformer model
- Leverages GPU memory bandwidth through software optimization
[Google] released HEIR, a private AI compiler based on homomorphic encryption [7]
- Enables AI inference on encrypted data
- Provided as open-source compiler toolchain
- Converts pre-trained AI models for encrypted inputs
- Targets regulated industries such as healthcare and finance
[Apple] negotiated payment deals with publishers for Siri AI news data [6]
- Proposed variable usage-based compensation model instead of fixed licensing
- Reviewing nine-figure budget (over $100 million)
- Siri AI slated for release this year
- Payment structure differs from industry standards
[Nine PBS · Iron Mountain] faced potential loss of 50TB archive after cloud storage provider vanished [1]
- Court ordered data return after loss of contact with OSS
- Iron Mountain provides only physical infrastructure and refuses data access
- Regional historical archive materials at risk of preservation
- Exposed risks in cloud storage contract structures
Sources
- [1] PBS station fears losing 50TB of data after being ghosted by cloud storage provider - Ars Technica AI
- [2] Kog is going deeper to squeeze more inference out of GPUs - TechCrunch AI
- [3] Hyperscalers might regret embracing natural gas if new forecast proves correct - TechCrunch AI
- [4] Databricks wanted to raise $1B, investors wanted $15B. It settled on $5B at a $190B valuation. - TechCrunch AI
- [5] Nvidia’s new $500B plan is risky but brilliant, especially for aging GPUs - TechCrunch AI
- [6] Apple in talks to pay publishers to provide Siri with current news: report - TechCrunch AI
- [7] Google is making private AI practical with homomorphic encryption - Hacker News