SpaceX Signs $6.3B Reflection AI Deal, Nvidia Eliminate Data Center Water

Summary of SpaceX Signs $6.3B Reflection AI Deal, Nvidia Eliminate Data Center Water

by Lex Fridman Podcast Fan

19m•June 23, 2026
Listen to Episode

Overview of SpaceX Signs $6.3B Reflection AI Deal, Nvidia Eliminate Data Center Water

This episode is a fast-moving AI news roundup focused on major infrastructure and product shifts: a huge compute contract for Reflection AI, NVIDIA’s push toward nearly water-free data centers, Apple’s plan to embed AI directly into core iOS apps, wave-powered data centers, and a new Oxford/AISI study showing AI can outperform humans in persuasion tasks. The through-line is that AI progress is increasingly being driven by compute, energy, and deployment strategy, not just model quality.

Key Stories Covered

1) Major compute deal for Reflection AI

  • The transcript says SpaceX signed a $6.3 billion compute deal with Reflection AI, but the underlying context appears to be Elon Musk’s Colossus 2 data center ecosystem near Memphis.
  • Reflection AI is described as an open-source lab founded by former Google DeepMind researchers.
  • The deal gives Reflection access to NVIDIA GB300 chips through 2029.
  • It is the third major tenant at the site, alongside Anthropic and Google.
  • Payment terms cited:
    • Reflection: about $150 million/month
    • Anthropic: about $1.25 billion/month
    • Google: about $920 million/month
  • A notable detail: these contracts reportedly include a 90-day cancellation clause after the first three months.

2) NVIDIA Rubin servers aim to eliminate data center water use

  • NVIDIA unveiled Rubin AI servers that use 45°C coolant, which the speaker says could essentially eliminate water consumption for cooling in many climates.
  • Key technical points:
    • Coolant enters at up to 45°C and exits around 55°C
    • This design reduces reliance on mechanical chillers and fans
    • A conventional data center can consume around 2.6 million gallons of water per megawatt per year
    • The Rubin design is presented as bringing that close to zero
  • The speaker also highlights:
    • Lower cooling costs
    • Better rack density
    • More compute in less space
  • Broader argument: many of the public concerns about data centers focus on water usage, but this new cooling approach may make that issue far less relevant.

3) AI and data center politics

  • The episode frames data centers as a geopolitical issue, not just an infrastructure one.
  • The speaker argues:
    • The U.S. should build more data centers
    • Some opposition to data centers may be politically motivated
    • Energy is a bigger near-term bottleneck than water
  • A policy idea mentioned: if companies build data centers, they should also build on-site power generation using solar, gas, or wind.

4) Apple is embedding AI directly into iOS apps

  • Apple is reportedly weaving AI into iOS 27 core apps rather than pushing everything through Siri.
  • Apps mentioned:
    • Messages
    • Mail
    • Calendar
    • Shortcuts
  • Example features expected in developer beta:
    • Bill splitting from receipt photos
    • Agentic password rotation across websites
  • The speaker likes the direction, especially the password automation, though notes possible issues if Face ID/device access fails.

5) EcoWave Power and ocean-wave-powered data centers

  • A company called EcoWave Power is using NVIDIA AI to run a data center at the Port of Los Angeles powered entirely by ocean waves.
  • The system reportedly has no grid connection.
  • NVIDIA’s AI helps:
    • Predict wave patterns
    • Schedule compute workloads to match energy availability
  • The episode emphasizes that wave energy is:
    • Highly dense
    • Potentially much more scalable than it sounds
    • Better suited to AI-era compute scheduling when combined with forecasting models

6) Oxford / AISI study: AI persuades better than humans

  • A study from Oxford and AISI found frontier AI models can persuade better than expert human debaters and professional fundraisers.
  • The study covered around 19,000 real conversations.
  • Main findings:
    • AI raised 3x more donations for charity in some settings
    • It shifted policy opinions more effectively than elite human debaters
    • AI beat professional canvassers by 10.8 percentage points in actual donation size
  • Important nuance:
    • The advantage appears to come less from “superior reasoning” and more from speed, throughput, and information density
    • When AI was forced to match human writing speed and message length, its advantage dropped to nearly zero
  • Models mentioned as top performers:
    • Claude Opus 4.1 / 4.6
    • GPT-4o / GPT-5.4
  • Lower performers in the study:
    • Gemini 2.5 Pro
    • Grok 4.2

Main Takeaways

  • Compute is the new strategic asset: The biggest AI companies are effectively becoming infrastructure customers, renting massive clusters to train and serve models.
  • Cooling and power are now central AI constraints: NVIDIA’s new server design and wave-powered systems point to a future where energy efficiency matters as much as model architecture.
  • AI product integration is shifting: Apple is betting that AI should live inside everyday apps, not just as chatbot sessions.
  • Persuasion is an emerging AI capability: The Oxford/AISI study suggests AI can be extremely effective in real-world persuasion tasks, especially when it can communicate quickly and at scale.
  • Infrastructure debates are becoming political: Data centers are now tied to energy policy, land use, and international competition.

Notable Quotes / Claims

  • “NVIDIA’s Rubin AI servers… basically kills the data center water use issue.”
  • “AI doesn’t win debates because it’s smarter; it wins because of speed and throughput.”
  • “SpaceX has basically accidentally become the world’s biggest compute landlord”
    • This appears to refer more accurately to Elon Musk’s broader xAI/Colossus data center ecosystem.

Promotional Notes Mentioned in the Transcript

  • The host promotes AIChatDaily.com for daily AI news summaries.
  • The host also promotes AIbox.ai and an MCP integration for bringing AI box tools into models like Claude, ChatGPT, and Gemini.

Bottom Line

This episode paints a picture of AI’s next phase: less about flashy demos and more about massive compute deals, new energy systems, and deep integration into everyday software. The biggest story is that AI infrastructure is becoming an industrial-scale arms race, while the newest technical breakthroughs are aimed at making that growth cheaper, denser, and far less dependent on water.