As a tech company, Expert needs to keep up with the latest in tech hardware in order to provide our clients with a reliable and speedy service, so we spend a lot of time planning and budgeting to replace components before they fail or become obsolete. Until recently, this was never a problem. Unfortunately, over the last few years things have changed dramatically and most of it is down to the rapid rise of artificial intelligence.

Initially we put price increases down to inflation and a weak NZ dollar, however it quickly became apparent that there was more to this than meets the eye, as supply had also become a real problem. Interestingly, the uptake in AI was in freefall. Coincidence? Yeah, nah.

So, I thought I’d do some in-depth investigating and here’s what I learned.

I posed the simple question ‘why do hardware components such as memory, drives, and Graphical Processing Units (GPUs) cost so much more now than they did before AI became popular and why is the supply so limited?

The short answer is that AI dramatically increased demand for the same components that gamers, businesses, and consumers already wanted, while those components remain difficult and expensive to manufacture.

Here's what's happened.

GPUs (graphic processing units) became the "gold rush" hardware

Before the AI boom, most high-end GPUs were purchased by gamers, video editors, engineers and scientists. Cryptocurrency miners also purchased GPUs during mining booms. Today, AI companies are buying them in unprecedented quantities to train and run large language models.

For example, a single AI data centre may purchase tens of thousands of GPUs in one order. A cluster of 50,000 GPUs can cost hundreds of millions of dollars, and some of the largest AI companies are deploying clusters containing over 100,000 GPUs. That level of demand far exceeds what the gaming market ever represented.

As a result, manufacturers prioritise enterprise AI customers, so consumer GPU supply tightens, and prices remain elevated because some businesses are willing to pay premium prices.

Advanced chips are incredibly difficult to manufacture

Modern AI hardware uses some of the most sophisticated manufacturing processes ever developed. Producing a leading-edge chip requires multi-billion-dollar fabrication plants ("fabs"), extreme ultraviolet (EUV) lithography machines costing hundreds of millions of dollars each, years of engineering and highly specialised packaging technologies.

Only a handful of companies can manufacture these chips at the most advanced process nodes. Because expanding capacity takes years and billions of dollars, supply cannot quickly respond to surging demand.

Memory demand has exploded

AI servers require enormous amounts of memory. A single AI accelerator may include 80–192 GB of very high-speed memory, and a large AI server can contain several terabytes of RAM. This increases demand for high-bandwidth memory (HBM), DDR5 server memory, and NAND flash storage.

Manufacturers naturally focus on the most profitable enterprise products, which can influence pricing across the broader memory market.

Storage requirements are enormous

Training AI models involves processing vast datasets. Companies often deploy petabytes of SSD storage, high-performance NVMe drives, and large-scale backup systems.

While consumer SSD prices have fluctuated over recent years, AI has contributed to stronger demand for enterprise-grade storage. SSD stands for Solid-State Drive. It is the primary storage device in modern computers, replacing older mechanical hard drives (HDDs) by using flash memory to save data instantly. Because it has no moving parts, an SSD makes computers boot faster, load programmes quicker, and run more quietly.

AI companies have much larger budgets

A gamer might spend NZ$2,000 on a graphics card. An AI company may spend hundreds of millions—or even billions—of dollars on hardware for a single project. When buyers are willing to pay almost any price to obtain scarce components, market prices tend to rise.

Supply chain disruptions didn't help

The AI boom followed several other events that had already strained hardware markets. First the COVID-19 pandemic disrupted manufacturing and shipping costs increased significantly. Semiconductor shortages affected many industries and geopolitical tensions created uncertainty around semiconductor supply.

Although many of those issues might have eased a little, they coincided with the rapid growth of AI demand.

New hardware is becoming more expensive to design

Each new generation of processors requires larger investments in research and development. This affects chip design, manufacturing processes, cooling systems, power delivery and even advanced packaging. Manufacturers recover these costs through higher prices, particularly for premium products.

Why memory and SSD prices still fluctuate

Unlike GPUs, memory chips are somewhat more commoditised. Prices can swing dramatically because manufacturers adjust production based on expected demand. Factors such as oversupply tends to reduce prices, while undersupply pushes prices up.

AI has increased long-term demand, but short-term pricing still depends on production levels and inventory.

Is AI the only reason prices increased?

No. AI is a major driver, but several factors contribute to this:

Will prices come back down?

They may, but probably not to pre-AI levels for the highest-end hardware. Several trends could help moderate prices, such as new semiconductor fabrication plants coming online, increased competition from more chip designers, greater production of advanced memory, and AI hardware becoming more specialised and efficient, which reduces reliance on the same general-purpose GPUs.

However, demand for AI infrastructure is still growing rapidly, and many technology companies continue to invest heavily in new data centres. That means high-performance GPUs and the fastest memory are likely to remain relatively expensive, while mainstream consumer SSDs and RAM should continue to experience more normal price cycles than AI-focused hardware.

Is it worth it?

So, there is light at the end of the tunnel for bespoke tech companies like Expert to pay slightly more realistic prices for hardware items. But like fuel, dairy and most other commodities these days, there’s a huge difference in the speed of prices going up compared to the speed of prices going down again.

If AI improves productivity at the expense of people in certain sectors losing their jobs, with those jobs disappearing, possibly forever, will the end result be worth it? Especially given the inevitable increases in costs, as companies try to claw-back the massive investment they are now being forced to make in technology hardware? Is the price we’ll be forced to pay to make things go a little bit (or maybe a lot) faster, worth it? We already know that AI doesn’t always get it right, all the time. But I guess that’s the price of progress and we’ve already passed the tipping point.

In case you need reminding, a tipping point is not just a quiz show on TV. A tipping point is the critical threshold or exact moment when a series of small, gradual changes builds up to cause a sudden, significant, and often irreversible shift in a system. Once this threshold is crossed, the ensuing change usually becomes unstoppable or snowballs independently.