The infrastructure wars are here. For a decade, NVIDIA owned AI compute. Now the cracks are showing—not in training, but in inference, where products actually live. We broke down why OpenAI signed a $10 billion deal with Cerebras while NVIDIA locked up Groq’s technology in a $20 billion licensing agreement. TrendForce confirmed the tension: OpenAI engineers want faster inference for Codex, and Cerebras claims up to 21x lower latency than NVIDIA’s Blackwell B200. The $30 billion question: who owns the next decade of AI compute?
Elon Musk thinks the answer is orbit. We analyzed the $1.25 trillion SpaceX-xAI merger—the largest in history—and the audacious plan to launch up to 1 million orbital data center satellites. Musk claims space-based AI compute will be cheaper than terrestrial within three years. The economics are questionable. The FCC filing is very real. And if he’s right, the entire infrastructure debate changes.
While the hardware wars rage, the model makers are drawing different battle lines. Anthropic just bought Super Bowl ads mocking ChatGPT’s upcoming advertising: “Claude will remain ad-free.” That’s an $8 million bet that users will pay to avoid sponsored responses—right as OpenAI confirms ads are coming to free and Go tiers. The infrastructure question is who controls the chips. The business model question is who captures the value. Both wars are heating up.
Latest from PulseMark
 |
OpenAI NVIDIA Alternatives: Why Inference Hardware Is Diverging
Cerebras, Groq, and the $30 billion battle for what happens after training. |
 |
SpaceX Acquires xAI: The $1.25T Vertical Integration Gamble
One million satellites. Orbital data centers. The most ambitious bet on AI infrastructure ever filed with the FCC. |
That’s Wednesday. Billions moving, battle lines drawn.
— The PulseMark Team
Get the Daily Pulse
Sharp analysis on what's actually moving in AI. No hype, no filler, no weekly digest.