AI demand for memory and computing hardware has pushed up US core inflation by roughly as much as the tariffs imposed early last year, according to Minneapolis Federal Reserve researchers — a finding with direct consequences for African organisations that buy the same equipment in dollars.
Prices for video and information processing equipment rose 12.2% year-on-year through July, adding about 0.4 percentage points to core personal consumption expenditures inflation. That is comparable to the entire contribution from tariffs, which the Fed puts at 0.2 to 0.4 percentage points.
Core PCE inflation reached 3.3% through July, the highest since 2023 and, outside the pandemic, the highest since the early 1990s.
The reversal is the point
The single most consequential figure for African readers is not the 12.2%. It is what it replaced.
Between 2015 and 2019, prices for video and information processing equipment fell at an annual rate of 6.5%. Hardware got cheaper every year, reliably, and that assumption was built into everything — data centre business cases, IT refresh cycles, device affordability projections, the economics of cloud itself.
That has inverted by nearly nineteen percentage points. Equipment that was becoming steadily cheaper is now becoming steadily more expensive.
Africa already saw this coming
iAfrica reported the forecast in June, when refurbished-technology provider Qrent flagged Gartner projections of DRAM prices rising 125% and NAND flash 234% as manufacturers diverted capacity toward AI infrastructure.
Kwirirai Rukowo, Qrent’s managing executive for the Middle East and Africa, described businesses facing “a growing imbalance between operational demand and available budget,” with procurement increasingly driven by price rather than need, and companies delaying projects and extending the life of ageing equipment.
The compounding problem he identified is currency. Across most of the continent, IT hardware is an imported, hard-currency cost. A 12.2% dollar-denominated price rise lands considerably harder where local currencies have weakened against the dollar over the same period — which is most African markets.
What it does to the build-out
The timing is awkward for a continent in the middle of an infrastructure push.
Turner & Townsend already ranked Lagos ($10.50 per watt), Cape Town ($10.33/W) and Johannesburg ($10.06/W) among the world’s more expensive data centre construction markets, with AI-specific facilities costing 7% to 10% more than traditional ones. Those figures predate the current hardware squeeze.
McKinsey’s estimate that Africa needs $10 billion to $20 billion in investment to reach 2.2 gigawatts of capacity by 2030 was calculated on assumptions that no longer hold. So were the business cases behind the projects announced this year — MTN’s data centre platform for South Africa and Nigeria, Raxio’s $380 million capital base, Amaco’s proposed $1.5 billion Mombasa facility, Equinix’s phased Johannesburg build.
Equinix’s decision is instructive in this light. The company disclosed this month that it holds R7.5 billion in budget and land in both Johannesburg and Cape Town, with no development under way on either, pending demand assessment. Rising equipment costs make that caution look more considered.
GPU access was already identified as a binding constraint on Nigerian AI development in the 2026 Global Index on Responsible AI. Higher prices tighten it further, and they widen the gap between operators who can absorb the cost and those who cannot.
And on the consumer side
The squeeze reaches past data centres. Apple raised MacBook and iPad prices by 15% to 25% in June. Lenovo, Dell and HP followed. Smartphone makers and console manufacturers have also raised prices.
Device affordability is the primary barrier to internet access across much of Africa — ahead of coverage, ahead of data cost. A hardware price cycle that runs upward for several years works directly against digital inclusion in markets where a smartphone already represents a substantial share of monthly income.
Where the pressure is coming from
The mechanism is straightforward. Microsoft, Google, Meta, Amazon and their peers are all investing in AI at speed, and all need the same components: CPUs, GPUs, video RAM, storage, cooling. Supply has not kept pace, and prices have moved accordingly.
Goldman Sachs put global AI investment at roughly $1.019 trillion for 2026 and noted that 8% of the nominal increase in US AI hardware spending this year reflects cost inflation rather than real expansion. Its recommended indicators for tracking the cycle include memory procurement and GPU rental prices, both of which sit in ranges not seen since 2022.
Some African responses are already visible. Malawi’s first AI data centre is being built around a GPU-as-a-service model so organisations rent rather than buy. South African-founded Supascale AI launched a marketplace pooling underused GPUs. Africa GPU Hub operates shared infrastructure across several countries.
Those are sensible adaptations to expensive hardware. They are also an admission that owning it outright is moving further out of reach.
