The platform giants fled the AI price war into hardware. The boom followed them.

In October, a Chinese cloud and software company put a pair of glasses on its website and said out loud what six rivals only imply. Alibaba's Qwen Glasses N1 and N1 Pro, at $450 and $520, are part of what the company describes as a deliberate move from data centers and software into consumer electronics — a full-stack bet on the object a person wears on their face ^[1]. That sentence is the key: it is the same bet Apple, Samsung, Google, Meta, Huawei, and Xiaomi are all making, only stated. And at $520, it is also just a pair of glasses. What Alibaba is running toward, the rest are running from. What it is running from keeps getting cheaper. OpenAI cut the fee on its Luna model by 80% to hold share against Chinese rivals, and Anthropic, xAI, and Meta all matched ^[2]. DeepSeek and Moonshot then distilled OpenAI's GPT-6 Astra outputs — trained their own models on its answers, a shortcut Google and Anthropic argue reproduces advanced capability far faster and cheaper than building it from scratch ^[3]. Citrini Research declared the traditional AI trade over and told clients to rotate into the real economy, as the Kospi fell 5% on Samsung and SK Hynix ^[4]. The one nuance is DeepSeek, which has begun raising its programmer prices as traction grows. The price of the software is collapsing, so hardware is where the industry is running. The run takes two shapes. There is the folded phone. Huawei's Mate XT 2 tri-fold launched at $2,980, running its first flagship chip in six years on an architecture built to dodge US semiconductor restrictions, and Xiaomi shipped a foldable to preempt Apple's debut — both pricing premium against a projected 14% slide in the global phone market ^[5]. Then there are the glasses that orbit the phone. Apple scrapped its lightweight Vision Pro headset and is instead developing display-free AI glasses for 2027, built to lean on the iPhone and a sharper Siri ^[6]. Google and Samsung launched Android XR glasses through Warby Parker and Gentle Monster ^[7]. Meta is scaling its Ray-Bans toward ten million pairs a year ^[8]. Seven players, one hedge: a device that belongs to one ecosystem, before the software inside it gets cheap enough to be anyone's. The trouble is that the same AI boom that made software cheap is inflating the exact components every phone needs. Gartner projects data-center demand will push DRAM and SSD prices up roughly 130% by the end of 2026, lifting phone prices 13% and cutting shipments 8.4% ^[9]. Apple's stated reason for its own price hikes is that memory chip costs are rising on demand from the hyperscalers — the cloud giants building out AI data centers ^[10]. The hedge and the boom it hedges against run on the same chips. Two predictions, made months apart, cannot both be right. Foxconn's chief executive said the AI boom will accelerate hardware upgrades, because devices need more processing power to run AI workloads ^[11]. Gartner's Ranjit Atwal calls it the other way.

This is the steepest contraction in device shipments witnessed in over a decade. — Ranjit Atwal

The mechanism is price: more expensive devices mean buyers hold them longer, stretching the upgrade cycle the whole pivot depends on ^[9]. The shipping evidence leans toward Atwal. Counterpoint projects a 14.3% drop in unit shipments on the memory crunch ^[12], and UBS reports wait times for the iPhone 18 Pro are flat versus last year ^[13]. The category everyone is converging on is barely growing. Global foldable shipments are projected to reach 20.2 million units this year — a 0.5% increase ^[14]. And the flagship everyone was watching stumbled out of the gate. Apple is replacing some iPhone Duo units over cracking screens that a teardown tied to hinge pressure on the display, while hinge supplier Fine M-Tec denies responsibility and Apple says only a small number are affected ^[15]. Samsung ran billboards mocking the defect as its own Z Fold8 surged more than 20% in Korea ^[16]. The demand was real before the crack. Pre-orders pushed delivery estimates to January 2027 within half an hour, forcing two-per-customer limits ^[17]. But the payoff is margin, not volume. Apple's US iPhone loyalty hit a record 96.4%, while Samsung's climbed to 90.1% and Google's to 86.8% — lock-in tightening across every ecosystem at once ^[18]. Investors have rerated Apple to 37 times earnings, crediting that lock-in and a Services segment running at 76% gross margin ^[19]. Even the people who try to leave mostly can't: Guardian readers who switched to dumbphones bounced back because their banking, transit, and healthcare live inside smartphone apps ^[20]. The one wing actually growing is the glasses. Shipments are projected to triple from 6 million to 20 million units in a year, with Meta holding 73% ^[8]. What constrains them is other people. Norway and Australia have banned them, boycotts have spread through the US and UK, critics note Meta's recording light can be switched off, and the public has given the devices a nickname Bernstein's analyst records without adopting: "pervert glasses" ^[21][22]. So privacy now dictates the design rather than trailing it. Google and Samsung built a physical camera switch, a capture light, and face redaction into the Android XR glasses, pitching them as the most privacy-forward device on the market ^[7], and smaller makers sell camera-free models with physical lens covers ^[23]. The backlash may even suit the frame maker: Bernstein's Luca Solca argues it keeps low-margin electronics from cannibalizing EssilorLuxottica's high-margin lens business ^[22]. Meta's own pitch is the fear of being left behind.

In the future, if you don’t have glasses that have AI you’re probably going to be at a pretty significant cognitive disadvantage compared to other people. — Meta Platforms Incorporated

But the object being shipped is smaller than the fear. The industry's newest differentiator is a mechanical switch that physically covers the camera — trust built as hardware, because the question came from other people.