Apple Goes All-In on AI Hardware, Memory Price Hikes Emerge as the Biggest Concern

Apple's latest earnings report reveals a clear strategic pivot: the company is positioning itself as the hardware supplier for the AI revolution. According to The Verge, Apple seems happy to let Anthropic and OpenAI burn cash training models while it focuses on selling increasingly expensive computers to run them. It's a clever calculation, but one issue that CEO John Tarnas admitted outright could make the whole plan tricky.

Apple's AI Hardware Business Model

From the earnings report, Apple has a very clear understanding of the division of roles in the AI era. It has no intention of competing head-on with OpenAI and Anthropic on model training. Instead, it chooses to be the one "selling shovels" — you go burn money training large models, and I'll sell the devices that can run them.

The brilliance of this strategy is that Apple leaves the most expensive, highest-risk part of the AI arms race to others while steadily capturing the premium from hardware upgrades. The report notes that the iPhone Duo has performed quite well in the market, and the price increases don't seem to have caused Apple any trouble. For consumers this is certainly not good news, but for Apple's financials, it's a well-oiled money-printing machine.

Memory Prices: The One Big Worry

The problem lies on the cost side. On the earnings call, Tarnas stated plainly that current memory pricing is "unprecedented," and he expects prices to rise further in the coming months.

That statement carries weight. Apple is renowned for its supply chain management capabilities — almost no company on Earth is better than Apple at playing this cost game. But even Apple has to publicly sound the alarm in the face of the current memory market. This means the magnitude and speed of memory price increases may already exceed the standard toolbox of responses.

For Apple's AI hardware strategy, this is a structural contradiction: AI devices are precisely the biggest memory consumers, and Apple wants to sell them at increasingly higher prices. If memory costs keep climbing, Apple will either have to compress profit margins or pass costs on to consumers — and the latter, given that price increases have proven "no problem," will most likely be the first choice.

Memory (DRAM) is one of the most cyclical categories in the global semiconductor market. Its prices are typically dominated by the three major manufacturers — Samsung, SK Hynix, and Micron — and even a slight imbalance in supply and demand can trigger violent fluctuations. Over the past two years, surging demand for high-bandwidth memory (HBM) from AI servers has led manufacturers to shift large amounts of capacity from consumer-grade DRAM to HBM and server memory, tightening the supply of standard memory used in phones and PCs. This structural capacity shift is the underlying reason for the current "unprecedented" memory price increases. For a company like Apple, which ships hundreds of millions of units annually, even a few dollars increase in memory procurement costs translates into billions of dollars in profit impact when multiplied by shipment scale.

OpenAI's "Data Center" Renaming Game

Another development worth noting comes from OpenAI. The company announced a new plan aimed at making its large computing facilities less of a burden on local power grids. But interestingly, these facilities are no longer called "data centers" — they're now called "compute campuses" and "compute facilities."

The Verge hosts were unequivocal about this: don't be fooled by the soft renaming. A new name doesn't change the enormous electricity consumption of these facilities — it's more like a PR-level repackaging.

Behind this naming strategy is a real pressure facing the AI industry: the energy consumption of data centers is becoming a focus of public and regulatory attention. Calling them "compute campuses" sounds gentler and more academic, but the physical electricity demand won't decrease by even a fraction.

Data centers typically refer to physical facilities that centrally deploy servers, storage, and networking equipment and draw large amounts of power from the grid. In the US, the siting, power consumption, and water use of large data centers are increasingly scrutinized by local governments and residents, and some states have begun discussing special electricity rates or grid connection restrictions for high-energy facilities. By renaming facilities "compute campuses," OpenAI can more easily package them as something akin to university or research campuses in regulatory communications and community relations, thereby reducing the public's intuitive resistance to "power hogs." But regardless of the name, their power supply capacity, cooling requirements, and grid connection application scale are all calculated based on physical load — renaming won't change the new generation and transmission capacity the grid needs to add.

GTA 6's Storage Anxiety

The last piece of news, though unrelated to AI, is shocking enough for gamers: Rockstar says GTA 6 may require 212GB of storage space on consoles. With only weeks until release, that number is enough to make players start thinking about deleting other games from their consoles.

This news indirectly confirms a trend: whether AI applications or AAA titles, demand for hardware resources is expanding rapidly. And the AI hardware track Apple is betting on sits right in the center of this trend — except that the trajectory of memory prices will determine how thick the final profit margin of this business will be.

Console game storage footprints have grown by an order of magnitude over the past decade. In the PS4 era, mainstream AAA titles typically installed at around 50GB, while this generation, with 4K textures, high-precision models, and more complex open-world streaming, 100GB+ has become the norm. 212GB means players need to reserve nearly a quarter to a third of their console's available space (about 800GB usable on a 1TB model), which directly conflicts with the system itself, other resident games, and applications. For hardware manufacturers, this actually benefits sales and upgrade demand for high-capacity SSDs — storage, like memory, is becoming a direct beneficiary of content expansion.

One-line Summary

Apple profits from AI hardware upgrades, but unprecedented memory price hikes are the biggest variable.

Paragraph Summary

The article analyzes the strategic pivot revealed in Apple's latest earnings report: Apple is not participating in the large model training race but positioning itself as the hardware supplier for the AI revolution, letting OpenAI and Anthropic bear the risk of costly training while earning premiums from device upgrades — with the iPhone Duo's performance supporting this strategy. However, CEO Tarnas admitted that current memory pricing is "unprecedented" and expected to keep rising, and AI devices are precisely the biggest memory consumers; cost pressure will force Apple to either compress margins or pass costs to consumers. The article also mentions OpenAI's PR repackaging of data centers as "compute campuses" and the trend of comprehensively expanding hardware resource demands reflected in GTA 6's 212GB storage requirement.