SpaceX Wants $40B for Nvidia Chips; Musk Still Says No to TSMC

SpaceX is reportedly aiming to raise $40 billion explicitly to buy Nvidia chips—Starlink revenue as collateral for silicon. The fundraise, first reported by National Technology News, would mark one of the largest single AI infrastructure bets by a private company. No pricing or timeline disclosed, but the optics are clear: Musk wants to own compute, not rent it, and he's willing to lever the satellite business to get there. Meanwhile, Intel confirmed it's still building Musk's Terafab chip plant despite ongoing TSMC talks, which Musk himself publicly ruled out a day later. The contradiction is the point—he's running parallel paths until one yields first silicon.

Lotte Hi-Mart, South Korea's consumer electronics chain, will deploy humanoid robots in at least one logistics center by 2028. No headcount, no vendor named, but the timeline matters. Two years is tight for site prep, integration, and operator training if the bots aren't already in late-stage pilots. Doosan Robotics separately landed a ₩98.9 billion state contract to build collaborative robots using domestic AI chips—part of Seoul's push to close the loop on robotics supply chains. South Korea is treating humanoids like semiconductors: strategic, subsidized, and non-negotiable.

dorsaVi, an Australian wearables firm, raised $2.09 million to sell human movement datasets to humanoid makers. The thesis: robots need biomechanical ground truth to train gait, balance, and manipulation models. dorsaVi's sensors have been used in sports rehab and workplace ergonomics; now they're pitching the same data as training corpus. It's a microcap bet, but the angle is legit—sim-to-real transfer still breaks on edge cases, and real human motion capture is expensive to generate at scale. If humanoid deployment actually happens, someone will pay for this data.

HD Construction Equipment secured a ₩390 billion order to supply power generation engines for a U.S. data center. Korean export, American buyer, diesel backup at hyperscale. Diesel costs and AI chip demand were cited by a Reuters-quoted market expert as twin inflation risks—fuel for gensets, fuel for training runs. Power is now the choke point everyone agrees on. Hitachi Vantara told Fierce Network the AI infrastructure cycle echoes the fiber boom of the late '90s: lots of buildout, uncertain payback timing, eventual consolidation. The comps aren't comforting.

U.S. banks' lending commitments signal confidence the AI capex cycle can sustain itself, per KBW. But bond markets are less sanguine. T. Rowe Price warned that AI-related debt issuance could push Treasury yields higher as companies stack leverage to fund chip buys and data center builds. Kristina Hooper at Invesco flagged the same risk: surging yields make the math on long-duration AI projects uglier. Temasek named AI spending and inflation as the two biggest risks for 2025. Michael Burry compared the current moment to the denial phase of 2000 and 2008, flagging data center financing specifically. The buildout continues; the credit market is starting to squint.

Tesla's Q3 deliveries beat production, FSD got approved in Denmark, and the company holds over $25 billion in unrestricted cash. Musk told Investor's Business Daily that Tesla has enough AI chips for near-term roadmap needs—no panic buying, no shortages. It's a rare all-clear in a market defined by chip anxiety. Nvidia stock price predictions for 2030 are now a genre unto themselves; Yahoo Finance ran another one this week. The hype cycle and the capex cycle are no longer in sync.

India is getting AI data center equipment manufacturing plants, per IMARC Group—no names, no capex figures, just feasibility studies and site surveys. The U.S., South Korea, and China lead in core AI sectors, according to a Chosun Ilbo analysis, while everyone else scrambles for middleware, assembly, and services. Sovereign AI isn't just about models—it's about who owns the steel, the power contracts, and the component fabs. The Robot Report framed it as 'same landlord' dynamics: different flags, same dependency on Nvidia allocation and utility-scale electricity. The race is physical now.