The AI boom has created a very different semiconductor cycle from previous shortages. Instead of demand coming from smartphones and PCs, it is being driven by hyperscale AI data centres that require enormous quantities of high-bandwidth memory (HBM). Most industry analysts now believe this is becoming a structural shortage rather than a normal semiconductor cycle, with tight supply potentially lasting into 2027-28.
Why is memory suddenly the bottleneck?
Training and running large AI models requires vast amounts of HBM, a premium form of DRAM stacked alongside AI GPUs.
Key factors include:
- AI server demand continues to outpace new fab construction.
- Manufacturers are prioritising HBM over conventional DRAM and NAND.
- Building new leading-edge memory capacity takes 2-4 years and tens of billions of dollars.
- Major customers including hyperscalers increasingly sign multi-year supply agreements, leaving little spare capacity.
Potential shortage scenarios
| Scenario | Probability | Market impact |
| Tight but manageable | Medium | Higher chip prices, strong semiconductor earnings |
| Severe AI memory shortage | High | Memory prices surge, hardware shipments delayed |
| Extreme supply crisis | Low-Medium | AI deployment slows, wider technology disruption |
Biggest winners
| Company | Why it benefits |
| Nvidia | AI GPU demand remains exceptionally strong if HBM can be secured |
| SK Hynix | Global HBM market leader with pricing power |
| Micron Technology | DRAM and HBM margins could remain elevated for years |
| Samsung Electronics | Largest memory producer with increasing HBM exposure |
| TSMC | More advanced packaging and AI chip production |
| ASML | Long-term beneficiary as chipmakers expand capacity |
Likely losers
| Company type | Risk |
| PC manufacturers | Higher memory costs squeeze margins |
| Smartphone makers | More expensive components reduce profitability |
| Consumer electronics firms | Higher input costs and possible shortages |
| Smaller AI start-ups | May struggle to secure scarce AI hardware |
| Data-centre operators | Capital expenditure rises further |
Could this hurt global stock markets?
Possibly—but not immediately.
Short term
Higher memory prices actually support semiconductor earnings.
That is bullish for:
- semiconductor stocks
- AI infrastructure
- equipment manufacturers
Technology indices could therefore continue outperforming while shortages remain manageable.
Longer term
If shortages become extreme:
- AI project deployments could slow.
- Cloud companies may delay investment.
- Corporate AI spending could become more expensive.
- Valuations across AI stocks may come under pressure.
Rather than a collapse, investors could see greater volatility as supply constraints periodically limit revenue growth.
Analyst commentary
Recent industry commentary has become increasingly bullish on memory pricing:
- Wolfe Research expects DRAM prices to jump sharply this year because manufacturing capacity cannot expand quickly enough.
- Goldman Sachs believes long-term HBM supply agreements improve earnings visibility for memory producers, although expectations are now much higher.
- Nvidia (NASDAQ:NVDA) CEO Jensen Huang has warned that shortages across memory and advanced AI components could persist for several years.
Nvidia Q1 FY2027 earnings: AI spending boom keeps powering world’s most important chip stock
Risks investors should watch
- New fabs arriving faster than expected could create oversupply after 2027.
- A slowdown in AI spending would reduce pricing power.
- Geopolitical tensions affecting Asian semiconductor production.
- Export controls disrupting supply chains.
- AI infrastructure investment becoming economically unsustainable.
Investment opportunities for UK retail investors
Highest conviction
| Theme | Outlook |
| AI memory manufacturers | ★★★★★ |
| Semiconductor equipment | ★★★★★ |
| Advanced packaging | ★★★★★ |
| AI infrastructure | ★★★★☆ |
| Consumer electronics | ★★☆☆☆ |
Investors who already own AI leaders may also want exposure to the ‘picks-and-shovels’ of the AI boom—memory manufacturers, semiconductor equipment suppliers, and foundries—which could enjoy sustained pricing power if shortages persist.
Investor verdict
Unlike previous semiconductor shortages, today’s memory crunch is being driven by a structural shift toward AI infrastructure. Production is increasingly being redirected to high-margin HBM, leaving conventional memory markets tighter than normal. Industry forecasts increasingly suggest constrained supply could last well into 2027 or even 2028.
For UK retail investors, the biggest beneficiaries remain companies with exposure to AI memory, chip manufacturing equipment, and advanced semiconductor fabrication. The AI boom has created a very different semiconductor cycle from previous shortages. The greatest risk is not an immediate collapse in technology shares, but that prolonged shortages eventually slow AI deployment and make today’s premium valuations harder to justify. Until new capacity comes online, supply constraints are more likely to support the earnings of memory producers than derail the broader AI investment cycle.
Disclaimer: The author Steven Frazer has a personal interest in Nvidia.
You might also like:
