Market Commentary

The Smartphone Floor Is Rising: what access inflation does to the next billion phone buyers

Memory chips are being redirected from phones to AI data centres, and the cheapest handsets are vanishing first. This note reads the shipment, pricing and margin data to ask what happens to the people who were about to buy their first phone — and what that leaves open for the businesses built to sell to them.

By Technology desk × Folka

1. The disappearing rung

Global smartphone shipments fell 22% in the third quarter of 2025 to 255.3 million units — the largest quarterly decline IDC has recorded. Counterpoint's preliminary read puts the fall at 20%, the weakest third quarter since 2012. Both trace the cause to the same place: memory chip costs that IDC puts up roughly 350% on a year earlier. Samsung and Apple gained share while the market shrank, which is the first clue that this is not a general slowdown. It is a sorting.

This is not a price rise. It is access inflation — the cost of owning a phone climbing fastest for the people who have the least room to absorb it, until the cheapest devices stop being made at all.

Coin a term and a market tends to treat it as inevitable. It is not. Access inflation describes a pattern in the data, not a law of the category — and the rest of this note is about how much of it is structural, and how much is a choice makers are still free to make differently.

2. Why the chip, why now

The intuitive explanation — a bidding war over the same chip — is not quite what is happening. Memory makers are converting wafer capacity toward high-bandwidth memory for AI accelerators, and that memory can take roughly three times the wafer of ordinary phone memory for the same silicon input. A factory retooled for a different product does not snap back when a price cycle turns; it comes back when the economics of the new product stop beating the old one.

That gap between a price floor settling and volumes recovering is not a rounding error. It is the window in which a cohort of first-time buyers simply does not get a phone on schedule — and a market does not necessarily refill that gap once the chips are cheap again.

3. The bottom falls out first

The pain is not spread evenly across price points or geographies. Rest of World reported on 5 October that existing phone prices are up about 15% globally this year, against roughly 5% in the United States. The cheapest tier has taken the sharpest hit: shipments of phones under $100 fell almost 60% in the second quarter, Africa's shipments fell 34%, and Oppo's sub-$100 shipments in Southeast Asia fell 96%.

Read in sequence, this is not a story about a product getting dearer. It is a story about a product being withdrawn from the people who most needed its lowest version, one shipment report at a time.

4. Two ways to hold a price

Facing a memory cost that triples in a bad year, a maker has roughly two levers, and the data shows both being pulled at once by different companies. One is to hold the price and cut the specification — lower RAM is reportedly returning as a base spec in some budget launches. The other is to hold the specification and raise the price, which is closer to what the premium end is doing. Neither is free, and the costs land on different people.

Apple's figures make the second path legible: it is absorbing part of the shock in its own margin rather than passing all of it on, which only a maker selling to buyers who will pay $1,199 can afford to do. The quieter version of the first path — a worse phone at the same price — is harder to see in a shipment chart, because it never shows up as a decline. It shows up as a phone that does less for the same money.

5. What the spec cut costs

A cut specification is not a neutral trade. It removes a buyer from categories of use that are becoming the baseline elsewhere — on-device AI chief among them, since Apple's on-device models are reported to need 12GB of RAM, a figure no sub-$100 phone touches. The risk compounds: the cheapest phones get both more expensive relative to income and functionally further behind.

The larger risk for this tier is not that the bottom shrinks for a cycle. It is that makers stop finding it worth their while to serve, independent of whether the chip shortage itself eases.

6. What access inflation feels like

The GSMA's income figure gives the scale of this away from the shipment charts: an entry-level phone cost the poorest 20% of people in low- and middle-income countries 44% of monthly income at the end of 2025, and 76% in sub-Saharan Africa. Jonathan Donner's After Access (2015) argued that 'connected' has always understated how fragile access actually is — shared phones, budgeted data, a device nursed well past its working life. A price shock does not create that fragility. It tightens a condition that was already there.

A phone that costs most of a month's income is not replaced when it breaks. It is repaired, shared or handed down.

This is why the IDC recovery date matters more than it looks like it should. A market that waits until 2028 for volume to return is not pausing a purchase decision for three years. It is asking the same fragile arrangement — the shared phone, the budgeted data — to hold for three years longer than it was built to.

7. Where price and access meet

Two things are converging on the same households: the cost of owning any phone, and the cost of a phone capable of what newer software now assumes. Neither alone explains the sub-$100 collapse; together, they do. A phone can be affordable and still be incapable, or capable and still be unaffordable — the group genuinely excluded is the one for whom both are true at once.

Every maker choice in section four pushes a phone out of one circle or the other: cutting spec keeps a phone affordable but pushes it out of capable; raising price keeps it capable but pushes a buyer out of affordable. Neither move enlarges the overlap. Only cheaper memory, or a different design assumption about what a budget phone needs to do, would.

8. Our view

We think the headline risk here is being misread as a supply-chain story that resolves on IDC's mid-2027 price timetable. It is better read as a demand-side sorting that may not reverse even once memory prices settle. Transsion's figures are the clearest evidence: its volume fell, its average price rose, and its profit rose 46% in the first half, partly on chips it had bought before costs spiked. That is a maker discovering that selling fewer, dearer phones to the buyers who remain pays better than holding volume at the bottom. Once that discovery is made, it tends to stick — margin logic does not reverse itself just because a chip gets cheaper again.

Big tech's capital spending is the force underneath all of it, and it is not aimed at phones at all. The $470 billion to $870 billion swing in hyperscaler spend is a bet on AI infrastructure that treats memory as an input to accelerators, not handsets. Phone makers are not a party to that decision; they are exposed to its side effect, through a shared supplier base they do not control. That asymmetry — one industry's capacity decision reshaping another industry's product line — is the real mechanism, and it is largely invisible from inside the smartphone market itself.

The human cost sits exactly where Donner's work said fragile connectivity always lived: in the gap between being covered by a network and being able to afford a device that reaches it. The GSMA's 3.4 billion figure was already large before this squeeze. What changes now is the direction of travel — a cohort that was closing that gap is, for some stretch of years, further from closing it than before.

We would weight the maker-incentive evidence (Transsion, Apple) above the supply-timeline evidence (IDC's 2027/2028 dates) when judging how long this lasts, because incentives are stickier than supply cycles. A chip shortage ends. A maker's conclusion that the bottom tier isn't worth chasing does not automatically end with it.

9. What would change this read

Three uncertainties sit underneath this note, and each is checkable rather than rhetorical. Whether the low end returns once memory eases is not guaranteed by the price timeline alone — it depends on whether makers who left the tier see a reason to re-enter it. Whether cheaper, newer entrants fill the gap depends on whether any maker finds a wafer-efficient way to serve the bottom while hyperscalers keep bidding for the same supply. And whether on-device AI's RAM requirements fall over time, as software efficiency work sometimes achieves, will decide whether the capability gap in section seven narrows or hardens.

Nothing in the current data answers that question yet — it sits past IDC's own recovery window. What the shipment and margin figures already show is enough to say the sorting is real, underway, and not evenly felt.

How we read this

The shipment collapse, the memory-cost rise, the price timetable and the income-share figures are measured, drawn from IDC, Counterpoint, Omdia, GSMA and company results as cited through the brief. The sorting read — that this is a demand-side split rather than a cycle that reverses on its own, and the Transsion margin-discovery argument in particular — is Folka's editorial interpretation of those figures, not a measured finding. The 2028 volume-recovery date and any claim about the overlap not rebuilding are projection. The one figure that would most change this read is the actual wafer or unit-cost trajectory for ordinary phone memory from 2026 onward: if that cost falls faster than hyperscalers' AI-memory demand grows, the sub-$100 tier could recover well before IDC's 2028 estimate, and the sorting would look temporary rather than structural.

Grounded in Folka’s corpus: The Chip Chokepoint · Intelligence That Stays Home