Orbital Apple Compute
Chips, Orbits, and the Limits of 'Unlimited Demand'
Nvidia's Jensen Huang disclosed that quarterly revenue is approaching one hundred billion dollars — not an annualized figure but a per-quarter total that places Nvidia in rough company with Apple, historically the most valuable corporation in the world. AI chip executives are publicly characterizing demand as 'unlimited' and deploying language like 'valuemaxxing' to describe their strategic posture. The numbers are real; the sustainability of the trajectory is what the industry is now quietly debating.
Sam Altman publicly accused Elon Musk of misleading investors about the commercial viability of SpaceX's orbital data center project, known as Starmind. Musk had previously called Altman a 'scammer.' Altman's counterargument is specific: computing in orbit costs orders of magnitude more per unit than ground-based alternatives, and no enterprise customer will pay that premium for marginal benefits. Goldman Sachs has forecast Micron as the top financial beneficiary of the orbital AI buildout, on the premise that high-performance memory is a bottleneck regardless of where compute is physically located. James Murdoch's SpaceX stake is now valued at seven point five billion dollars.
Microsoft CEO Satya Nadella offered a sobering counterpoint to the demand euphoria, warning that companies are effectively 'paying twice' for AI tools — once for subscriptions and again for compute costs, integration work, and productivity losses during adoption. The total cost of ownership, Nadella suggested, is substantially higher than headline license fees implied. Meanwhile, a philosophical dispute is opening up around open versus closed frontier AI models: Zhipu's founder argued publicly that frontier AI should be accessible to all, echoing arguments from Meta's camp. Palantir's CEO has been attacking AI labs' pricing models, with Galaxy's Mike Novogratz publicly backing that critique.
The antitrust dimension is now live as well. A looming Apple lawsuit related to ChatGPT's iOS outage over the weekend raises questions about whether the distribution relationship between Apple and OpenAI has exclusionary elements that disadvantage rival AI apps on iOS. The Sherman Act — the foundational U.S. antitrust statute passed in 1890 — prohibits not large market share itself, but the use of monopoly power to exclude competition through mechanisms such as predatory pricing, exclusive dealing, or anti-competitive product bundling. Whether the Apple-OpenAI arrangement crosses those lines will be one of the defining legal questions of the coming years.
Apple's M7 Ultra chip is reportedly designed to support one point five terabytes of memory, according to journalist Mark Gurman — a figure that would represent a meaningful step toward using Apple hardware for serious AI inference tasks currently requiring server infrastructure. The figure underscores a broader trend: memory, not just processing speed, is emerging as the key constraint in advanced AI workloads, which is precisely why Goldman Sachs is watching Micron and why Huawei's move to build a DRAM fabrication facility in Shenzhen — with 140,000 wafers per month capacity — is being closely tracked by semiconductor analysts.