東京ビジネスハブcurrently outside the Japan top 100
AI分野で出遅れたアップル。反撃の一手は(岡ゆづは)
Apple is reportedly exploring entry into the AI server market, planning to sell servers powered by its in-house M-series chips to AI companies such as OpenAI and Anthropic. The move would transform Apple from a chip buyer into a chip seller — a pivot reminiscent of NVIDIA's position in the AI infrastructure stack. NewsPicks reporter Oka Yuzuha (岡ゆづは) breaks down Apple's AI lag, the counteroffensive led by its new CEO, and the strategic calculus behind a potential NVIDIA partnership.
What this episode answers
Why is Apple considered to be behind in AI?
Apple's capital expenditure for the first half of 2026 was just $4.4 billion (roughly ¥700 billion), compared to Amazon's approximately $100 billion — a gap of more than 20 times. Apple Intelligence has also failed to make a strong impression, leaving the company looking behind on both infrastructure and products.
What exactly is Apple's AI server business model?
Apple plans to sell servers equipped with its M-series chips to enterprises and government agencies — a B2B model that flips Apple from a chip buyer to a chip seller, with a launch targeted for around 2029.
Why are Mac Studio and Mac mini selling out among developers?
OpenAI has purchased tens of thousands of units to cluster for AI model training and inference, and the machines' AI inference performance has won strong praise among developers; the top-spec Mac Studio has seen a 7–8 week wait.
Why would NVIDIA consider supplying technology to a potential competitor like Apple?
NVIDIA is alarmed that Google and Amazon have begun selling their own in-house chips externally, eroding its dominant market position; even if Apple's chips displace NVIDIA in the server, NVIDIA wants to secure sales of its NVLink Fusion interconnect technology.
What concerns remain about Apple entering the AI server market?
Apple has little B2B experience and has previously entered and exited the server market; whether it can build the enterprise-grade support infrastructure and rapid incident-response capabilities required by large customers remains an open question.
Apple's AI Lag — A Stark Gap in Capital Expenditure
Comparing capital expenditure for January–June 2026, Apple stands out as the lowest spender among the GAFAM group by a wide margin. Apple invested $4.4 billion (approximately ¥700 billion), while Amazon — the highest spender — invested roughly $100 billion, more than 20 times as much. On the product side, "Apple Intelligence" has failed to make a meaningful splash, and criticism has mounted that Apple has shown little innovation since the death of Steve Jobs.
Mac Studio and Mac mini — Hidden Horsepower
Among Apple's product lines, Mac recorded the highest recent sales growth rate — up 29% year on year, compared with 22% for iPhone. The driver is bulk purchasing of Mac Studio and Mac mini by AI developers. Both are compact, box-shaped machines roughly the size of two palms, sold without a monitor or keyboard, and typically stocked at the back of Apple Store shelves.
Word spread that OpenAI had purchased tens of thousands of units, connecting multiple machines in clusters for local AI model training and inference. Stock-outs followed. Even at over ¥1 million for a top-spec Mac Studio (M5 Ultra), units were unavailable in Japan until the end of November.
Apple has pursued semiconductor in-house development since around 2010 for iPhone and around 2018 for Mac. On top of that, AI inference chip technology developed for autonomous driving under the Apple Car project — disbanded in 2024 — has been carried forward into the current M-series.
The AI Server Business — From Buyer to Seller
What Apple is now exploring is a business selling enterprise servers powered by next-generation M-series chips — anticipated to be M7 Ultra or M8 Ultra — with a projected launch around 2029. The plan is to take the "AI-capable chip" credibility demonstrated in consumer Mac Studio hardware and translate it directly into server products sold to model-development companies such as OpenAI and Anthropic, as well as to government agencies.
Anthropichas a stated policy of not depending solely on NVIDIA for procurement and adopted Google's and Amazon's in-house chips early on. Oka argues that Apple's servers could become an attractive option for any organization looking to reduce its NVIDIA dependence.
NVIDIA and Apple — A Complex Mix of Competition and Cooperation
Apple and NVIDIA have historically had a strained relationship. During the Jobs era NVIDIA supplied GPUs for Apple products, yet Jobs reportedly suspected NVIDIA of misappropriating Pixar technology — a claim that has never been confirmed. Apple's preference for custom specifications versus NVIDIA's focus on general-purpose products also created friction.
Nevertheless, Apple is reportedly exploring adoption of NVIDIA's NVLink Fusion, a multi-chip interconnect networking technology. Why would NVIDIA license technology to a potential rival? Because Google and Amazon have begun selling their own in-house chips to third parties, steadily chipping away at NVIDIA's share. Even if Apple's server business prevents NVIDIA from selling its chips into that stack, NVIDIA wants to secure at least the networking equipment sales — a sign of the defensive urgency it now feels.
A CEO's Pet Project — and the B2B Hurdle
The person who has led this AI server initiative for roughly the past year is John Ternus, who was then Apple's head of hardware engineering. Having joined the company in 2001 and spent nearly 26 years on the hardware side, Ternus was appointed Apple's new CEO in September 2026. His predecessor Tim Cook was an excellent operator by stock price and revenue metrics, but opinions are divided on whether he planted the right seeds for the AI era.
Concerns remain, however. Apple has limited experience in B2B businesses, and its previous foray into the server market ended in retreat. Building enterprise-grade capabilities — instant response to outages, dedicated support — requires a very different skill set from making beautiful consumer hardware.
While other GAFAM companies are squeezing capital expenditure and free cash flow on AI infrastructure, even raising debt to fund investment, Apple may be able to compete on the same field simply by selling its own chips. Oka concludes that "looking back, this could be remembered as the smartest choice Apple could have made."
Takeaways
- Apple is exploring an AI server business that would use its proprietary M-series chips to sell to enterprises and AI labs, turning the company from a chip buyer into a chip seller in the mold of NVIDIA.
- Mac Studio and Mac mini — purchased by OpenAI in the tens of thousands for AI inference clusters — have proven the M-series chip's AI credentials and are driving the case for a dedicated enterprise server product.
- NVIDIA is willing to supply its NVLink Fusion interconnect technology even to a potential competitor because Google's and Amazon's in-house chip sales are already eroding its dominance, making networking-equipment revenue a necessary fallback.
This article is edited from a machine transcript of the first 30 minutes of the episode. It is not an official summary — please listen to the episode itself.