Every breathless headline about a new AI model glosses over a physical fact: these systems run on silicon, and that silicon is hungry for memory. The current scramble to build out AI infrastructure has set off a quiet chain reaction in the global components market — one that ends, improbably, at the checkout counter of a first-time smartphone buyer in Uttar Pradesh or Sindh. The mechanism is unglamorous and largely invisible, but its effect is real: the AI boom is exerting upward pressure on memory prices, and the most exposed victims are the cheapest devices on the market. Call it the hidden tax on the next billion users.
How AI demand reaches your phone
The link runs through a specific component: high-bandwidth memory, or HBM. This is the premium, stacked memory that sits alongside AI accelerators in data centres, feeding them data fast enough to keep their compute cores busy. HBM is expensive to make, complex to manufacture, and — critically — it competes for the same fabrication capacity as the ordinary memory inside consumer devices.
Memory makers have a finite amount of wafer capacity. Every wafer dedicated to HBM is a wafer not making the DDR and LPDDR chips that go into laptops, tablets, and phones. As reported by dentro.de citing technologist Simon Willison, HBM is expected to consume roughly 20% of total wafer capacity by the end of 2026 — a dramatic jump from around 2%. (This figure should be treated as directional and checked against memory-market primary sources, but the trend it describes is well established.) When a fifth of capacity is reallocated toward a single high-margin product, the supply of everything else tightens.
Tightening supply does what it always does: it pushes prices up. And because memory makers will always prioritise the highest-margin orders, the squeeze flows downhill. Data centres pay premium prices and get served first. Flagship phones, with fatter margins, can absorb some increase. The pressure accumulates at the bottom of the market, where there is the least room to manoeuvre. That is how an investment decision in a Texas or Taiwan data centre eventually reprices a handset designed to sell for less than a hundred dollars.
Why South Asia feels it first
The sub-$100 phone is a marvel of cost engineering. Its bill of materials is pared to the bone, and memory is one of the larger line items. On a device with margins measured in single-digit dollars, a meaningful jump in the cost of LPDDR or NAND flash is not a rounding error — it can be the difference between a viable product and an unviable one. There is simply no cushion to absorb it.
This matters enormously in South Asia, because the budget segment is not a niche here — it is the on-ramp to the entire digital economy. According to IBEF, India had roughly 125.87 crore wireless subscribers and around 85.5% smartphone-owning households as of early 2026. That base did not materialise from premium devices; it was built on affordable handsets that brought first-time users online. The next tranche of users — the genuinely unconnected — are by definition the most price-sensitive of all. They are exactly the buyers a memory squeeze prices out.
So a problem that looks like a procurement headache in a spreadsheet becomes, in this region, a digital-access issue. India’s ‘next billion users’ story — the one underpinning everything from UPI adoption to vernacular content to government digital services — assumes a steadily falling cost of entry. The reported impact of the HBM-driven squeeze on sub-$100 smartphones, devices critical to markets across South Asia and Africa, runs directly against that assumption. When the cheapest device gets more expensive or worse, the slope of the affordability curve flattens, and the people at the margin stay offline a little longer.
Who absorbs the cost
When input costs rise, someone in the chain has to eat them. There are three candidates: the OEMs who build the phones, the carriers who sometimes subsidise them, and the buyers who ultimately pay. In practice, the cost gets shared — but the split is rarely flattering to the consumer.
OEMs facing a memory crunch have two unappealing levers. The first is a price hike: keep the spec, charge more. The second is a spec downgrade disguised as a stable price: ship the same nominal price point but with less RAM or less storage. Both are forms of the same tax. A buyer who pays the same money for a phone that now has 2GB of RAM where 4GB was becoming the floor is paying — just in performance and longevity rather than rupees. Spec downgrades are arguably worse, because they are quieter and they shorten the usable life of a device, accelerating the upgrade cycle for people who can least afford it.
Carrier subsidies and financing can blunt the impact, and in some markets they will. But carrier economics in budget South Asian segments are themselves thin, and there is little appetite to absorb sustained component inflation indefinitely. That leaves the buyer holding most of the bill.
The timing makes it sharper. Much of South Asia’s budget-phone volume clusters around festive-season buying. If a memory price spike lands ahead of that window, OEMs face an ugly choice between protecting margin and protecting their price-point positioning during the year’s most important sales period. Get the timing wrong and a launch lineup can look uncompetitive at exactly the moment volume matters most.
What to watch
None of this is destiny, and a few variables will decide how hard the tax bites.
- Memory pricing cycles. Memory is famously cyclical — gluts and shortages alternate as capacity catches up with demand. The key question is whether AI demand has structurally changed the cycle or merely amplified a phase of it. If suppliers add capacity fast enough, the squeeze eases. If HBM keeps absorbing a growing share of wafers, the floor under consumer-memory prices stays elevated for longer.
- Domestic assembly and PLI effects. India’s production-linked incentive schemes and the broader push to localise smartphone manufacturing change the cost calculus — but mostly on assembly and final-stage value, not on the memory chips themselves, which remain a globally traded, import-dependent input. Local assembly can shave logistics and duty costs; it cannot conjure cheaper DRAM. Watch whether component-level localisation, including any domestic packaging or fabrication ambitions, ever reaches the memory layer. Until it does, India remains a price-taker on the input that matters most here.
- On-device AI: offset or accelerant? There is a genuine tension worth watching. On-device AI features are being marketed down-market, and they are memory-hungry — running models locally needs more RAM, not less. If budget phones are pushed to add memory to support AI features precisely when memory is scarce and dear, the squeeze worsens. The optimistic counter is that smarter on-device processing could reduce reliance on cloud round-trips and, over time, justify devices that do more with less. For now, the demand-side pressure looks more likely to compound the problem than relieve it.
The uncomfortable through-line is that the AI infrastructure boom and the digital-inclusion project are now competing for the same scarce resource. That competition is invisible to almost everyone involved — the data-centre operator never sees the priced-out buyer, and the buyer never hears the word HBM. But the connection is real, and it deserves to be named. The cheapest phones are where access begins. If the cost of AI’s appetite quietly lands on them, the bill is being paid by precisely the people the next-billion-users story was supposed to lift up.
