The global PC hardware crunch is moving from a niche gripe among enthusiasts into a full-blown market shock, and for South African buyers it is landing at exactly the wrong time. What started as a headache for gamers and DIY builders is now being driven by the explosive demand for AI infrastructure, with hyperscalers soaking up memory, graphics capacity and now even CPU supply.
For anyone trying to build a serious desktop today, the numbers are becoming hard to justify. A machine with enough RAM to run a useful local AI model, a modern multi-core processor and a capable GPU is no longer a modest enthusiast purchase. In many cases, it is creeping into the price bracket once reserved for used cars, and that is before local mark-ups and import pressure are added.
As we reported earlier, the real problem is that the world’s biggest technology firms are buying up production in bulk. Microsoft, Meta, Google parent Alphabet and Amazon are spending at a scale that dwarfs the consumer market, and chipmakers are responding by serving the customers who can lock in the biggest orders, years ahead of time.
Reuters reported this week that SK Hynix is being courted aggressively by global tech firms offering direct investment in production lines, with one source describing available capacity as “essentially zero”. That tells you everything about the squeeze. The chip maker’s shares have surged about 154% this year, as investors pile into the firms supplying the AI boom.
The spending is staggering. Microsoft has guided the market towards US$190-billion in capital expenditure in 2026, with roughly US$25-billion attributed to higher chip and component costs. Meta, Google and Amazon are also pouring in eye-watering sums. Combined, the hyperscalers are now committing about US$725-billion to AI infrastructure this year, up from US$600-billion only a few months ago.
How the hyperscalers are driving the **global PC hardware crunch**
That flood of money is reshaping the semiconductor industry in real time. Memory makers such as SK Hynix, Samsung Electronics and Micron Technology have shifted wafer capacity away from commodity DDR5 and standard DRAM, and towards high-bandwidth memory used in AI accelerators.
In practical terms, that means the best production is being reserved for HBM3e chips for current-generation Nvidia and AMD hardware, while HBM4 is now ramping for next-generation silicon, including Nvidia’s Rubin platform. For ordinary PC buyers, the result is a nasty shortage and prices that keep ratcheting higher.
The situation has been described, only half-jokingly, as the “RAMpocalypse”. DRAM prices reportedly rose by about 172% in 2025, and even major hardware vendors are warning that memory is now taking up a much larger share of system cost. HP told investors on its latest earnings call that memory now makes up 35% of the bill of materials in a PC, up from 15% to 18% in the previous quarter.
A 32GB DDR5 kit that sold for under US$90 a year ago can now fetch as much as US$530. For anyone trying to spec a machine with 128GB or more for local AI work, the pricing is starting to feel absurd. In South Africa, the pain is amplified by limited supply, exchange-rate pressures and the simple fact that we sit low on the global allocation list.
Local retailers have already been warning customers for months. Back in March, Evetech, Dreamware Technology and Tech.co.za were reporting dramatic increases, with DDR5 up as much as 230% in a single quarter, DDR4 up 150% to 200%, and SSDs up 35% to 50%. Since then, conditions have worsened rather than improved.
Even a firmer rand has not been enough to cushion the blow. The global shortage has overwhelmed any currency gains, meaning South African consumers are still paying a premium for hardware that is increasingly being treated as strategic inventory by the world’s biggest cloud companies.
Graphics cards have been under pressure for longer, and the latest moves from Nvidia show where the priority lies. The GeForce RTX 5090 launched in 2025 at US$1 999, but some custom models are now selling for US$3 000 or more. Reportedly, production of the 5090 and 5080 has been cut by 30% to 40% as Nvidia shifts GDDR7 memory and packaging capacity towards its data-centre business.
That business now generates the overwhelming majority of group revenue, and it is where the biggest profits are. One high-end AI accelerator can sell for around US$40 000, while a flagship gaming card cannot. From Nvidia’s point of view, the choice is obvious. From the perspective of gamers, creators and enthusiasts, it is a bitter pill.
What is now making the situation worse is the shift in CPU demand. For the past two years, the AI boom was mainly about GPUs and memory. CPUs were less affected, and for buyers that meant prices remained relatively manageable. That is changing quickly as the industry moves deeper into agentic AI.
Agentic systems, which can plan tasks, call tools and carry out multi-step jobs with limited supervision, lean far more heavily on CPUs than earlier chatbot-style applications. Even basic inference still requires considerable CPU orchestration, and once this is scaled across hyperscaler workloads, the CPU side of the data centre becomes a major bottleneck.
That has been good news for Intel, which a year ago looked like a company in serious trouble. Its share price has more than quintupled since then, briefly touching US$114 this week and pushing its market value above US$570-billion.
Intel’s long-delayed 18A process node is now in volume production at its new Arizona fab, and the first commercial Panther Lake processors built on 18A have reportedly landed with decent reviews from technical outlets. Even more telling, Apple is said to be considering shifting some chip production back to Intel after ending its reliance on the company in late 2023.
That may be a vindication for Intel, but it is not good news for ordinary buyers. More and more capacity is being locked in for data centres and hyperscalers years in advance. AMD is also enjoying stronger pricing power, which limits the chance of consumer-friendly pricing returning any time soon.
The broader picture is stark. The global semiconductor supply chain is now being optimised for a relatively small number of extremely large customers building AI data centres, and that reality is distorting the market for everybody else. Whether the current investment frenzy turns out to be a bubble or not, there is little sign of an immediate correction.
For South African hobbyists, gamers and local AI builders, the message is hard to ignore. The economics of on-device computing have been broken, at least for now, by the scale of the data-centre build-out. Until the industry changes course, the global PC hardware crunch is going to keep punishing anyone who wants serious performance at a sane price.