Yes. Interesting push on the sovereign AI thesis from Palantir, NVIDIA, Microsoft. Sovereign agentic workflows for enterprises. This is critical. And this is being more and more accessible, for example using NVIDIA infrastructure, foundation models and recipes (check NeMo suite and Nemotron open models, software and recipes). This is the data flywheel thesis in action: - Data processing. This needs filtering, dedup, etc - Model customization (i.e LoRA, RL, ...) - Model evaluation - Guardrails for security / safety - Custom model deployment. - Enterprise data refinement, and repeat to step 1.

These frontier labs ability to recoup all the costs to train these models by selling tokens as a business model is laughable. None of these companies will ever get an ROI by doing this. They are most certainly in the business of stealing IP in order to develop their own applications and products.

Nice, all AI guys are finally competing to be the worst? 🤣 They're stealing books, IP, secret docs, nonprofits, and even your whole repo now. Amazing. Innovation in robbery. I'm sure if you were stealing someone else’s repo, you could literally be jailed. Tbh, local models are becoming so good. Invest in hardware. Don’t ask yourself how much you’re spending, but how much you would lose if your secrets were stolen.

Palantir CEO Alex Karp has just taken aim at OpenAI and Anthropic. He claims Fortune 500 companies are paying billions for AI "tokens" that create little real value. His argument is simple: If an AI model can genuinely generate $1 billion in value... Why charge for tokens instead of taking a share of the upside? It raises a bigger question: Are today's AI business models built around delivering value... or selling access? 🤔

🔥GLOBAL COMPANIES TURN TO CHINESE AI TO CUT COSTS DoorDash, Siemens, and Airbnb are adopting models from DeepSeek, Z .ai, and other Chinese firms as they seek cheaper alternatives to U.S. frontier labs, per FT. Chinese models have overtaken U.S. rivals in token usage this year, driven by lower costs and growing concern in Europe over dependence on American AI providers.

MORNING BRIEF ☕️: AI is entering its efficiency era. OpenAI, Meta, and xAI are redesigning models to deliver similar performance with less computing power, lowering costs and increasing competitive pressure across the industry. Cheaper intelligence will accelerate AI adoption. 🤖

Jensen Huang just broke down the entire AI industry as a five-layer cake: 1. Energy (the foundation) 2. Chips 3. Cloud/Infrastructure 4. Models 5. Applications (the top) Most founders, funding, and talent are jammed into the Models layer, the one with the fiercest competition. That intensity doesn’t create moats. It creates commodities. The harder everyone races, the faster intelligence becomes cheap and abundant. The real scarcity isn’t smarts anymore. It’s power. No algorithm builds a grid. No fundraising replaces a dam. Energy controls access at the bottom. Applications capture the revenue at the top. Hundreds of billions are pouring into the middle layers that sit between someone else’s electrons and someone else’s economy. Intelligence scales. Energy doesn’t. Build where the bottleneck and the payoff actually live.

Everyone's talking about AI replacing office jobs... Meanwhile, engineers are teaching spacecraft how to survive on their own in deep space. 🌌