Samsung, AI memory and the transition from deployment to productivity

Signal

For most companies, record profits are rewarded with rising share prices. For Samsung Electronics, the opposite happened.

The South Korean technology group announced preliminary second-quarter results showing operating profits reaching 89.4 trillion won, approximately 58.5 billion dollars, nearly nineteen times higher than a year earlier. Yet despite reporting one of the strongest quarters in its history, Samsung shares fell sharply in Seoul, pulling semiconductor stocks lower across Asia and Europe.

This was less a reaction to actual performance and more a pre-emption of future expectations.

🟦 Markets do not simply price current earnings. They price the sustainability of future earnings.

For much of the past eighteen months, investors embraced a straightforward narrative. Artificial intelligence requires accelerators. Accelerators require memory. Memory shortages create pricing power. Pricing power generates extraordinary profits.

Samsung, together with SK Hynix, became one of the principal beneficiaries of this dynamic. Yet the latest results appear to have triggered a different question.

Not whether demand for AI infrastructure remains strong. But whether financial markets have already priced in years of future growth.

🟦 The issue may no longer be whether AI infrastructure will continue to expand. It may be whether today’s shortages inevitably become tomorrow’s excess capacity.

Much of Samsung’s earnings expansion appears to have been driven by extraordinary memory pricing conditions.

Demand for HBM, DRAM and advanced packaging remains exceptionally strong as hyperscalers race to build AI datacentres at unprecedented scale.

But commodity cycles often peak precisely when current scarcity encourages future oversupply. Oil experienced this. Solar manufacturing experienced this. Lithium experienced this. Semiconductors have experienced it before as well.

Periods of exceptional profitability frequently trigger waves of investment that ultimately weaken the very supply constraints that created those profits in the first place.

Investors may therefore not be questioning demand. They may be questioning duration.

🟦 Perhaps the most significant shift is that markets appear to be moving from a deployment economy towards a productivity economy.

For much of 2024 and 2025, the dominant question was straightforward. Who could build the largest datacentres? Who could deploy the most accelerators? Who controlled access to advanced memory?

Increasingly, however, investors appear to be asking different questions. How quickly will enterprises integrate AI into workflows? Which business models can justify hundreds of billions of dollars in capital expenditure? When does infrastructure become utilisation? And when does utilisation become productivity?

Samsung’s record earnings suggest that demand for AI infrastructure remains extraordinarily robust.

The market response suggests something else. The next phase of the AI cycle may no longer be judged by the amount of infrastructure that can be constructed. It may be judged by the economic value that infrastructure ultimately creates.

---

Credit

Illustration: Altair Media / AI-generated artwork

Caption

When Record Profits Trigger Fear
Record profits rarely unsettle markets. Yet Samsung’s latest results may reveal a deeper transition: from an economy focused on building AI infrastructure towards one increasingly concerned with utilisation, productivity and long-term returns.