A $950 Billion Week in South Korea
On July 24, Nvidia chief executive Jensen Huang, OpenAI's Sam Altman, and the heads of Samsung, SK Group, Hyundai Motor and Naver converged in San Francisco for a summit with South Korean President Lee Jae Myung. By the end of the day, roughly $950 billion in new AI agreements had been signed, positioning South Korea as central to the next phase of the global AI buildout.1 Among the resulting arrangements was an expanded Nvidia–SK Hynix partnership. Nvidia's Raj Mirpuri said "the expansion will include a co-develop opportunity for us on the next-generation SK Hynix AI memory, and this will help us secure a stable supply of HBM memory."1 For Nvidia — which competes directly with Advanced Micro Devices for AI-accelerator share — locking in high-bandwidth memory supply from one of a small number of credible HBM suppliers is as strategically important as any single chip contract.
The Correction Beneath the Buildout
None of that has insulated chip stocks from a rough few weeks. On July 28, the Nasdaq 100 fell to a 2.75-month low even as the Dow Jones Industrial Average climbed to a 1.5-week high, with markets describing "a deepening rout in chipmakers and AI-infrastructure stocks."2 Packaging and test supplier Amkor Technology added to the gloom that same day, guiding third-quarter net sales to $1.95 billion–$2.05 billion, below the $2.11 billion Wall Street consensus.2 Memory maker Micron Technology captured the volatility most starkly: its stock fell as much as nearly 40% from its all-time high before bouncing 10% off those lows by August 9, leaving shares still roughly 30% below the peak.3 Leveraged semiconductor exposure has swung even harder — ProShares' 2x-leveraged Ultra Semiconductors fund (USD) was up about 69% year-to-date and 196% over the trailing year as of June 5, a run steep enough that a subsequent slide was severe enough to be described as the fund having "imploded" 17% within days.4 That whipsaw is a reminder that leveraged semiconductor vehicles amplify both the AI rally and its unwinding.
Fundamentals: What Kept Moving While Stocks Fell
Set against that sell-off, the deal flow underpinning AI chip production barely paused. On July 29, GlobalFoundries signed a letter of intent with the U.S. Department of Commerce under which the CHIPS Research and Development Office is expected to award the company $300 million to advance next-generation silicon photonics, the optical technology used to move data at high speed inside AI and high-performance-computing data centers.5 AMD's Mark Papermaster said "silicon photonics and advanced packaging will be key to delivering the bandwidth, energy efficiency, and system-level connectivity required for the next generation of AI cluster infrastructure," and AMD said it welcomes GlobalFoundries' continued U.S.-based investment.5 Four days earlier, AMD had unveiled a separate collaboration with Cerebras: AMD Helios systems and the Cerebras Wafer-Scale Engine will run as a single disaggregated inference workflow, pairing AMD Instinct GPU throughput with Cerebras' fast token generation, with the joint offering set to reach Cerebras Cloud in the second half of 2026.6 That pairing gives AMD an inference-specific pitch distinct from Nvidia's memory-supply strategy in the same market.
Consolidation moved forward too. Skyworks Solutions and Qorvo named the leadership team for their combined company on July 28, effective once their pending merger closes. Skyworks' Bob Bruggeworth said the announcement "reflects the strong partnership that has shaped our integration planning efforts from the very beginning," adding he was "confident these leaders will help foster collaboration across our teams."7 In France, Kalray and Bull announced a collaboration on high-speed networking technology for AI and HPC infrastructure, including compatibility with the Ultra Ethernet standard. Kalray's Éric Baissus said the Bull agreement confirms the relevance of the company's technological vision and the quality of its teams' innovations, framing the tie-up as a step toward next-generation AI and HPC networking solutions.8
The Supply Chain's Mixed Signals
Beneath the headline deals, smaller suppliers show the AI buildout is not lifting every balance sheet equally. Vicor Corporation, which makes power-delivery components used in AI computing, reported trailing-twelve-month revenue of $471.7 million and net income of $136.7 million, with a market capitalization of $11.3 billion as of the July 6, 2026 market close.9 Its lead computing customer is separately set to transition from Gen 4 to Gen 5/VPD power technology in the second half of 2026, with a ramp expected before year-end.10 Solitron Devices posted a starker divergence between top and bottom line: net sales rose 101% to roughly $5.44 million in its fiscal 2027 first quarter, and the company said it expects sales to "continue to be at this level or greater for the remainder of the 2027 fiscal year" on the strength of a backlog that grew 28% to $23.34 million.11 Yet net bookings fell 48% to $1.45 million, and net income was dented by "an increase to the contingent consideration of $0.33 million related to the MEI earn out payment based on a large order received after the end of the fiscal quarter."11 Allegro MicroSystems, which describes itself as "leveraging more than three decades of expertise in magnetic sensing and power ICs to propel electrification, automation, AI data center, and robotics forward,"12 added semiconductor-industry veteran Brian White to its board on June 18. Director Joseph Martin said "Brian's public-company CFO perspective, semiconductor industry experience and governance background make him a strong addition to our Board."12
Edge AI and the Next Front
Deal-making has reached the client-device layer too. On June 2, Phison announced a collaboration with Intel pairing its Pascari aiDAPTIV memory-expansion technology with Intel Core Ultra Series 3 processors, aimed at letting AI PCs run larger local models. Phison's KS Pua said "AI PCs are evolving into platforms for more sophisticated local AI workloads, including agentic applications and larger MoE models that place increasing demands on memory capacity and responsiveness," adding that the Intel collaboration helps expand the memory available for AI workloads on Intel's AI PC platforms.13 It's a smaller deal than the Korea memory pact or the GlobalFoundries award, but it points to the same underlying dynamic: chipmakers across the stack, from HBM to edge memory, are still racing to add AI capacity regardless of where their share prices sit.
What to Watch
Via News's reading of this data is that the sell-off and the buildout are not contradictory signals but two different clocks — one tracking near-term earnings and positioning (the Amkor guidance miss, the Nasdaq's 2.75-month low, the leveraged-ETF whipsaw), the other tracking multi-year capacity and supply agreements (the Korea pact, the GlobalFoundries CHIPS award, the Skyworks-Qorvo merger). Micron's bounce off its lows — still roughly 30% below its all-time high — is the closest thing in this record to a stabilization signal, but it is a partial recovery, not a reversal.3 It is also worth noting that several of the deal announcements cited here originate from wire-distributed press releases whose source material Via News's own fact-checking has measured at claim-verification rates of roughly 23% to 48% historically — a reason to treat forward-looking guidance, in particular, with some caution even where the underlying agreements themselves are corroborated. Further out, competitive pressure on the roadmap side is visible in Alibaba's T-Head unit, which has scheduled its Zhenwu V900 AI chip for commercial release in the third quarter of 2027 and a follow-on Zhenwu J900 chip for the third quarter of 202814 — a reminder that the supply agreements signed this summer are being built against AI chip roadmaps that already stretch to the end of the decade.
