Samsung Electronics and Taiwan Semiconductor Manufacturing Co. both reported record profits in their latest results, providing the clearest confirmation yet that the AI-driven semiconductor upcycle has not yet rolled over.

The two companies occupy complementary positions in the AI chip supply chain. TSMC manufactures the leading-edge logic that runs AI training and inference workloads; Samsung supplies high-bandwidth memory that pairs with those processors. When both post record earnings in the same window, end-customer AI capital expenditure is still growing, not merely rotating between suppliers.

What do the results say about the AI trade?

Investor anxiety about AI demand normalization has been building for several quarters. The worry is straightforward: hyperscalers have spent aggressively on accelerator clusters since 2023, and at some point they will digest that capacity and pull back. A simultaneous record at the largest foundry and the largest memory supplier is the cleanest data point available against that thesis.

The earnings also point to where spending concentrates. Foundry revenue records reflect demand for advanced-node wafers, the substrate for AI accelerators and high-performance CPUs. Memory records reflect tight supply of high-bandwidth memory, particularly HBM3E and the next-generation HBM4 ramp.

Why is Samsung's memory side the bellwether?

Samsung holds a leading position in DRAM and NAND, but the relevant product for AI is HBM — stacked DRAM that delivers the bandwidth AI accelerators require. Memory orders tend to lag accelerator order books by one to two quarters because HBM shipments follow accelerator build commitments. A record at Samsung therefore implies that accelerator volumes into data centers remain on an upward trajectory.

Samsung's foundry business, which competes with TSMC, remains a different story. The unit has spent years trying to close the yield gap on advanced nodes, including 3nm and the 2nm generation. Its foundry losses have narrowed, but it still trails TSMC on customer wins in AI accelerators. Investors evaluate the two businesses separately; the memory result is the AI signal.

How concentrated is the supply chain?

The earnings underscore how thin the AI chip supply chain has become. Three companies — TSMC, Samsung, and SK Hynix — supply most of the advanced logic and HBM that goes into frontier AI systems. Nvidia designs the dominant accelerator; TSMC fabricates it; Samsung and SK Hynix supply the memory. Earnings reports from these two companies therefore act as a real-time read on AI capital expenditure.

What could break the upcycle?

Three risks remain visible:

• Hyperscaler digestion. If Microsoft, Google, Meta or Amazon conclude they have over-built AI capacity, foundry orders would likely fall within two quarters.
• HBM supply additions. Both Samsung and SK Hynix are expanding HBM3E capacity, and a faster-than-expected ramp could compress memory margins.
• Export controls. Further restrictions on advanced chip trade with China would remove a meaningful slice of foundry revenue, though AI-specific demand sits mostly outside the restricted geographies.

Until one of these breaks, the combined signal from the two earnings reports points the same direction: AI infrastructure spending continues to outrun the rest of the semiconductor cycle, and the two companies best placed to capture it keep posting records.