In 2017, Ethereum mining profits exploded. Miners bought every GPU they could find — Nvidia and AMD cards disappeared from shelves, prices tripled, and PC gamers were furious. Then, in early 2018, the crypto market crashed. Within six months, GPUs flooded back into the market at or below original prices. The boom had lasted approximately 18 months, left no permanent structural change in GPU pricing, and became a cautionary tale about commodity markets driven by pure speculation. Now, in 2026, the GPU and RAM markets are experiencing price inflation of a comparable or greater magnitude — but driven by artificial intelligence rather than cryptocurrency. The question is whether history will repeat itself.

The answer requires separating what looks the same from what is structurally different. And there are genuinely significant differences — ones that suggest the outcome will not be a clean replay of the crypto collapse. But there are also warning signs of speculative excess that cannot be ignored. This guide gives you the honest, data-grounded picture — and tells you specifically what it means for your laptop buying decision in Kenya in 2026.

In 2017, Ethereum's price rose from approximately $8 to over $1,400. Mining Ethereum was extraordinarily profitable, and GPUs — particularly Nvidia GTX 1080s and AMD RX 580s — were the tool for doing it. Miners bought every available GPU. Prices doubled and tripled on secondary markets. Retail shelves were empty. Laptop and desktop GPU availability collapsed for ordinary consumers.

Then Ethereum's price peaked in January 2018 at approximately $1,400 and began falling. By December 2018 it was at $83 — a 94% decline. The economics of mining flipped from extraordinarily profitable to actively loss-making at current electricity prices. Miners stopped buying GPUs overnight. They began selling their existing stock. A flood of used GPUs hit the market simultaneously. New GPU prices returned to MSRP within months. The crash was complete, fast, and total — because the underlying demand was purely speculative, had a clear profitability threshold, and could reverse direction in a single trading session.

Intellectual honesty requires acknowledging the real similarities. The AI hardware market in 2026 displays several characteristics that mirror speculative bubbles, not structural demand shifts — and dismissing these signals would be as wrong as dismissing the structural differences.

The DRAM price volatility is itself a bubble signal. When DRAM prices crashed 30% in weeks merely because AI capital flows slowed — without any change in actual deployed infrastructure or real-world compute needs — that is the behaviour of a speculative commodity, not a stable supply-demand equilibrium. Real supply-demand mismatches correct gradually as new capacity comes online; they do not crash 30% in a few weeks. This crash-and-spike pattern is exactly what commodity markets with speculative overlay do.

GPU performance gains have hit diminishing returns. The RTX 5090 uses twice the power of the previous generation for 40–50% more performance. Data centres report that doubling investment now yields less than 20% real throughput gains in many 24/7 AI workloads. When the cost of the next hardware generation exceeds its productivity benefit, rational capital allocation demands a pause. That pause, when it comes at scale, looks very similar to crypto miners' overnight stop.

AI revenue has not fully caught up with AI infrastructure investment. Microsoft, Google, Amazon, and Meta have collectively spent hundreds of billions of dollars on AI infrastructure. The consumer and enterprise revenue generated by AI products — Copilot subscriptions, AI-powered cloud services, API revenue — while genuinely significant, remains well below the scale of infrastructure investment. The gap between what has been spent and what has been earned creates a correction risk that is similar in structure (if not magnitude) to the crypto boom.

The most critical structural difference is the nature of the buyers. Cryptocurrency miners in 2017 were, individually, small rational actors responding to a price signal. When the price signal reversed, they all responded simultaneously, flooding the market with used hardware. The AI buyers of 2026 are Microsoft, Google, Amazon Web Services, Meta, and a cohort of well-funded AI startups — organisations with multi-year infrastructure commitments, enterprise purchase contracts, and strategic imperatives that do not dissolve in a single earnings quarter.

The second critical difference is what happens after the hardware is purchased. A crypto mining GPU stops being useful the moment mining becomes unprofitable — it cannot mine another coin at the same rate, and it generates no value sitting idle. An AI data centre GPU, once trained on a model, continues to generate inference revenue from every API call, every Copilot suggestion, and every AI-powered product query. The value of AI hardware is not destroyed when AI token prices fall — because AI is not priced like a token.

The AI hardware boom will not end like the crypto crash — but it will not be permanent either. It is something in between, and that matters specifically for how Kenyan buyers should think about their next laptop purchase.

The crypto crash was total because its demand driver (token profitability) could flip from positive to negative in a single day, and because the buyers were individual economic actors who responded rationally and instantly to that flip. The AI hardware boom's demand drivers are corporate infrastructure commitments measured in years, backed by real revenue, and managed by organisations that sign multi-year contracts with chipmakers. A Microsoft or Google does not cancel its AI infrastructure programme because DRAM prices fluctuate or because one quarter's AI revenue misses analyst estimates. The structural demand floor is genuinely different in kind, not just in magnitude.

However, the 30% DRAM crash that occurred when AI capital flows merely slowed — not stopped — is a clear signal that the speculative premium baked into current prices is real and will eventually deflate. New fabrication capacity will come online. AI model efficiency will improve (we are already seeing this with smaller, more efficient models doing work that previously required much larger ones). The combination of supply relief and efficiency improvement will trigger a meaningful correction from peak prices. Just not overnight, and not back to 2023 levels.

The question of whether the AI hardware boom ends like the crypto crash is ultimately a question about the nature of the underlying demand. Crypto mining demand was pure arbitrage — profitable one day, unprofitable the next, with no floor below profitability. AI infrastructure demand is structural — embedded in multi-year corporate strategy, backed by real enterprise revenue, and representing a genuine long-term shift in how computation is consumed. These are meaningfully different, and they predict meaningfully different outcomes.

What they share is a speculative premium — the portion of the price that reflects hoarding, anticipation, and fear of missing out rather than immediate real demand. That speculative premium will deflate. The question is whether it deflates violently (the crypto scenario) or gradually (the commodity cycle scenario). The evidence overwhelmingly points toward the latter: a partial correction over 18–36 months as supply catches up, settling at prices permanently above the pre-AI baseline. If you are a Kenyan buyer making a decision today, our full range of laptops at today's pre-inflation prices on quality refurbished hardware represents the most rational response to this environment.