Nobody can definitively answer your question considering all the variables at play.
We are inside the “AI boom” bubble and if you went solely by investor enthusiasm, it shows no signs of slowing down. If you move into tangible metrics like data center capacity, data center occupancy, chip sales, capital expenditures, and public demand there’s no evidence of a plateau.
The fundamental issue we have right now is that chip production isn’t sustaining increased growth and until growth starts to slow or chip manufacturing grows, we’re not going to see an improvement in parts pricing.
You asked if hardware will get cheaper in 2 years, well in Microsoft’s latest quarter they’d said they’d added 31 new data centers across five continents, added roughly 1 GW of capacity in that quarter alone, and is on track to roughly double overall capacity in two years, and Microsoft has said that customer demand is exceeding capacity.
Gartner has estimated that AI spending is going from $1.79 trillion in 2025 to $2.67 trillion in 2026 to $3.46 trillion in 2027 so based on forecasts, we’re not looking at a plateau any time soon.
Experts have said that 2028 is emerging as a plausible inflection point and that there will be a cyclical pause in semiconductor capital spending in 2028, and that growth will continue at high levels.
Based on all the raw data it does not look like the AI industry will be plateauing any time within the next 4 years.
Now that’s one piece of the puzzle, next we have to look at hardware production…
SEMI’s latest forecast has total installed semiconductor capacity growing about 5% in 2026 and 6% in 2027, while 300 mm capacity is expected to grow 7% in 2026 and roughly 8% annually from 2027 through 2029. Semiconductor-manufacturing equipment spending is also setting records, meaning this isn’t merely announced fabs, companies are actually buying the machinery required to increase output.
That and semiconductors are going to get smaller and more powerful so perhaps they’ll need less to do more (wishful thinking).
TrendForce expects DRAM to remain undersupplied throughout 2027. New fabs start contributing in 2027, but because semiconductor fabs take a long time to qualify and ramp, they don’t expect substantial output from that new capacity until 2028.
On the graphics side of things, TSMC is building 13 advanced-packaging fabs in Taiwan and they’ve increased their investments in Arizona by a few billion more.
So with all that being said, I don’t believe we’re going to see any price fluctuations in our favor until 2028 if we’re lucky, but more likely 2030.
I’m personally looking more at the used market for additional systems.