I went in with a specific goal. Since January, I've sent our customers a monthly update on market volatility, and I've written here about how extreme things have become, including when RAM passed gold in price. The summit confirmed what we're seeing in the field, straight from the source. That kind of direct manufacturer insight is what keeps our guidance to customers sharp.
Everyone asks when the AI bubble will pop. Kingston’s answer was that the technology is not going anywhere. The real question is financial: when will the biggest players stop pouring money into AI infrastructure?
Nobody knows, and nobody expects it to happen soon. Hyperscalers are already placing memory orders into 2028, and that horizon could stretch toward 2030. Most hyperscalers are custom designing chips.
At the summit, Kingston explained that fabricators had spent roughly five years in an oversupplied market, losing money through much of that period. They are not eager to repeat that cycle. Capacity is tight and prices remain high. Consumer demand has dropped because buyers will not pay today’s prices, but that has not been enough to move the market.
Chipmakers are shifting wafer capacity toward the highest-density memory chips that AI demands. That capacity is coming away from lower-density parts used in watches, wearables, and many phones. Those markets are likely to feel the effect over the next few years.
I asked whether newer memory manufacturers in China might ease the pressure globally. Kingston’s answer was no. Much of what is produced in China is staying in China.
Memory was the first domino. On the summit’s last day, motherboard manufacturers announced increases of roughly 30%. We have also seen price movement in graphics-card supply, and PNY/Nvidia expects another increase in Q4. Some components have climbed 20% to 30% month over month. Even if that slows to 5% per month, it compounds faster than most budgets can absorb.
The impact will not stop at hardware. Data centers at this scale consume staggering amounts of power and water. Communities may push back, but the buildout moves elsewhere. We all still want our AI tools.
First, a shout-out: Kingston shares some of the most useful market information I get anywhere. Kingston walked attendees through roadmaps for its two primary product lines: SSDs and memory. They diverge across server and industrial, consumer, and gaming markets.
Some manufacturers have prioritized AI-related demand over consumer-facing lines. Kingston went the other way. They told us plainly that they will not abandon long-term partners for short-term data center profits. Nobody believes this market lasts forever, and you do not alienate your bread and butter chasing a spike.
A few other things impressed me:
What stayed with me most is Kingston's culture. It's rare for a company their size and reach to be so down to earth. The people are easy to work with and focused on more than the bottom line. In today's world, that's refreshing, and it's admirable.
Some of the best conversations at the summit were with other integrators. We compared notes on negotiating with distributors and using AI to make purchasing more efficient. One peer told me how “crazy” it was that everything had worked out for his company after they just waited.
That was not my experience. In late 2025, we saw early signs that the surplus was drying up. Early this year, we doubled our order volume across a rolling two-year horizon, locking in our place in the queue before allocations dropped to about 15%.
Then in summer, PNY told us it had 30 RTX 6000 Ada cards left, and I committed to all of them. Within days, other buyers were calling to ask for them. For some deployed systems, the newer Blackwell cards were not a drop-in replacement because of specific vGPU compatibility requirements. Knowing that difference is the job.
For mission-critical teams, here's what I'd prioritize now:
For more on how Kingston handles allocation, read our Voices Driving Innovation conversation with Jake Box.
The whole Radeus Labs team is heading to Orlando for I/ITSEC, the world's largest modeling, simulation, and training event. It brings together defense, industry, and academia around the technology behind warfighter readiness. Every one of those training and simulation systems depends on the kind of compute this market is squeezing.
If you're planning programs through this volatility, or working through a computing, recording, or a simulation challenge, we'd love to talk. Feel free to book time with out team at the event here.
See you in Orlando!
— Juliet Correnti is the CEO of Radeus Labs, A U.S.-based cleared facility and ISO 9001 certified manufacturer of computing and satellite communication systems for military, infrastructure, aerospace, government technology solutions, and public safety environments.