The SOXX semiconductor index dropped roughly 25% in late July 2026, and leveraged semiconductor ETFs fell 62% from their highs. The selloff came despite record or better-than-expected quarterly results across the supply chain.1
Amkor Technology, a key packaging and test partner for AI chipmakers, reported results ahead of expectations, underscoring that assembly capacity used for AI accelerators stayed in demand even as investors repriced risk.1
SK hynix locked in long-term supply deals for AI memory, a move that secures high-bandwidth memory allocation for AI infrastructure buyers years ahead. That contract structure signals memory suppliers expect sustained AI demand regardless of stock-market sentiment.
Foundry consolidation accelerated on the state level. The U.S. government backed GlobalFoundries with a $300 million CHIPS Act R&D award and took a roughly 1% equity stake in the company, tying public funding directly to domestic chip-manufacturing capacity.
Supplier consolidation also moved forward outside foundries. Skyworks and Qorvo finalized leadership for their merger. "Today's announcement reflects the strong partnership that has shaped our integration planning efforts from the very beginning," said Bob Bruggeworth.2 Fewer independent RF and component suppliers concentrates more of the AI hardware supply chain into larger, merged entities.
Networking infrastructure for AI data centers also saw new investment. Kalray and Bull agreed to jointly develop next-generation high-speed networks for AI and high-performance computing. "As AI continues to redefine infrastructure needs across every sector, we are excited to offer technologies that deliver greater performance, flexibility and efficiency," said Éric Baissus.3
The pattern across these moves — record fundamentals, locked-in memory supply, state-backed foundry capacity, and supplier mergers — shows AI infrastructure buildout continuing on its own timeline. Investor risk-repricing in chip stocks has not slowed the underlying capacity commitments that determine how much AI compute gets built.