Google’s AI Brain Drain, SpaceX's Huge Quarter, Airtable’s 90% Collapse, US Data Fuels China AI

21 insights saved from this video by @business

@business· Business Models
One gigawatt of compute capacity costs at least $50 billion. Six gigawatts implies roughly $300 billion in capital expenditure.
@business· Business Models
Fifteen of fifteen US cabinet agencies reportedly run on Salesforce, making a vibe-coded replacement implausible.
@business· Business Models
Compute payback assumptions compressed from roughly four or five years to potentially one year as scarcity drove prices sharply higher.
@business· Business Models
Bending Spoons’ potential Airtable play is stark: cut most costs, abandon sales-led growth, and restore product-led growth.
@business· Business Models
Microsoft’s resilience is infrastructure: Active Directory holds identities, Excel supplies board numbers, Teams records compliant conversations, and Azure carries government authorizations.
@business· Business Models
Open-source and commodity models already handle 95% of everyday jobs, while frontier models matter when failure is costly.
@business· Business Models
Open models look cheap until you count training, fine-tuning, maintenance, safety guardrails, and specialized expertise.
@business· Business Models
Frontier models demand tens of billions, while open-source and open-weight models rapidly narrow the gap, making model investment riskier than infrastructure.
@business· Business Models
Premium frontier models resemble Apple: cheaper alternatives may have more users, yet superior experiences capture disproportionate monetization.
@business· Business Models
Open models may capture more tokens while frontier labs capture more economics, letting both sides of AI grow.
@business· Business Models
The winning enterprise cloud may orchestrate many models, not sell only one.
@business· Business Models
The strategic case for restricting AI-data exports rests on one question: is the data proprietary and military-relevant?
@business· Business Models
Export controls on EUV lithography machines show what a strategically meaningful technology restriction looks like.
@business· Business Models
Data labeling is largely a commodity, and China has both alternative labor and multiple ways to obtain it.
@business· Business Models
AI-generated tools may replace lightweight apps, but replacing CRM, ERP, or HR systems wholesale is far harder.
@business· Business Models
Before Airtable was acquired, it spun out Hyperagent so its AI-agent talent could chase a higher-upside opportunity.
@business· Business Models
AI model selection is becoming blended: cheap open-weight models for simple workflows, premium models for high-value tasks, specialized models for domains.
@business· Business Models
Anthropic, OpenAI, and Nvidia are the main buyers able to commit to compute at this scale, not Chinese open-source companies.
@business· Business Models
Cheap open models can match frontier systems for ordinary work while frontier models remain substantially ahead on sophisticated use cases.
@business· Business Models
Starlink severely disrupted HughesNet, ViaSat, AT&T, and Verizon across legacy satellite and telecommunications markets.
@business· Business Models
Google, Microsoft, and SpaceX balance compute sales against internal model development, while Anthropic and OpenAI avoid that channel conflict.

More from All-In Podcast

See everything from All-In Podcast

Post on X

Keep what you found in Google’s AI Brain Drain, SpaceX's Huge Quarter, Airtable’s 9…

A free account saves any of these cards to your Collection, and Korva resurfaces them so you actually remember.