take-two

Meta's AI investment is already paying off in ad targeting improvements, but the market is spooked by leaked town hall comments about slower agentic progress and worries about incremental capex for frontier models beyond core ads ROI.

Nvidia's investment in CoreWeave has been highly profitable, and Take-Two argues current 'vendor financing' fears miss that hyperscaler GPU cloud businesses may have 60-80% inference margins and generate massive free cash flow in 12-18 months.

Jensen Huang stated on Bloomberg that Nvidia has enough supply chain capacity (including power and data center shells) to double revenue every year, but consensus estimates for next year are far below that trajectory.

The Chinese Kimi model is a 2.8 trillion parameter model requiring 64-GPU servers and big networked clusters — not an efficiency breakthrough but a capability leap that will spur compute demand, analogous to DeepSeek's reasoning models which increased rather than decreased GPU consumption.

Sam Altman stated at YC that the next 6 months will see dramatically more AI capability advancement than the prior 2 years, and both OpenAI and Anthropic are signaling recursive self-improvement is imminent, which would create an exponential compute demand surge.

AMD CEO Lisa Su dramatically increased the 2030 CPU total addressable market forecast from $120B to $220B in just three months, signaling she sees insatiable demand that serious public CEOs don't project unless orders are materializing.

Nvidia rallied the entire tech industry (Google Cloud, Amazon, OpenAI) to counter Anthropic's push for open-weight restrictions, while using its scale to lock up scarce supply chain components (optical, TSMC wafers, HBM) and invest in ecosystem companies like CoreWeave, Lumentum, and Coherent.

An SK Hynix executive revealed during their IPO roadshow that customers (likely including Nvidia) are requesting 5-6x more HBM than they can supply, prompting a plan to double capacity over five years — confirming structural memory shortage.

Amazon CEO Andy Jassy explicitly stated in his annual letter that the $200B capex isn't a hunch — they see concrete demand, know it will be insanely profitable, and expect free cash flow positivity in the medium-to-long term.

AT&T exemplifies enterprise AI adoption: 100 generative AI models in production, burning a trillion tokens monthly with double-digit growth, driven by competitive necessity — if they don't use AI for customer service and R&D, rivals like Verizon will disrupt them.