In the same week that Nvidia locked down a long-term memory supply agreement and the U.S. government agreed to help fund next-generation silicon photonics for AI data centers, chipmaker and AI-infrastructure stocks sold off hard enough to drag the Nasdaq 100 to a 2.75-month low.1 That gap — strengthening technical and supply fundamentals on one side, a sharp equity repricing on the other — is the real story of the AI chip sector this summer, and the underlying releases show why neither half of it is exaggerated.
The Supply Deals Say the Buildout Is Real
On July 24, South Korean President Lee Jae Myung flew to San Francisco for a summit that Jensen Huang, Sam Altman, and the heads of Samsung, SK Group, Hyundai Motor and Naver all attended; by the end of the day roughly $950 billion in new AI agreements had reportedly been signed.1 Out of that summit, Nvidia expanded its partnership with SK hynix. "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," said Raj Mirpuri.1 That is Nvidia moving to lock in the high-bandwidth memory that its GPUs depend on — not a speculative bet, but a supply-chain hedge for a company that is still the center of gravity for AI compute.
Five days later, 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 accelerate next-generation silicon photonics — the optical technology GF describes as underpinning the AI and high-performance-computing data centers driving the global economy.2 AMD, a GF customer, welcomed the move: "As AI systems scale, moving data efficiently is as critical as increasing compute performance. 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," said Mark Papermaster, adding that AMD "welcomes GF's continued U.S.-based investment."2 This is federal money following the same bottleneck the industry keeps naming: not raw compute, but the interconnect and memory bandwidth needed to feed it.
The buildout isn't confined to data-center silicon, either. Phison and Intel announced a collaboration bringing Phison's Pascari aiDAPTIV memory-extension technology to Intel Core Ultra Series 3 AI PCs, aimed squarely at a shift already underway on the client side. "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," said KS Pua, describing how aiDAPTIV expands the memory available to AI workloads on Intel's platforms so OEMs, developers and end users can run more capable AI applications locally while preserving privacy and infrastructure efficiency.3 On the inference-serving side, AMD and Cerebras announced a joint architecture pairing AMD Instinct GPUs' throughput with the Cerebras Wafer-Scale Engine's fast token generation into a single disaggregated inference workflow, with Cerebras planning to deploy AMD's Helios systems in its own data centers and make the combined solution available through Cerebras Cloud in the second half of 2026.4
An Ecosystem Widening Past the Hyperscalers
Via News's mapping of the entity relationships behind these releases shows Nvidia's footprint extending well past the familiar cloud giants.5 Its own agentic-AI tooling — the NeMo Agent Toolkit, NeMo Guardrails, A-IQ, Nemotron 3 Super, NeMo microservices and the DGX Spark hardware line — feeds into deployments at customers as varied as Mount Sinai Health System, the State of Alaska Legislative Affairs Agency and Yum! Brands, while distributor TD Synnex supplies Nvidia hardware into the channel and HPE has built its own Agentic Trend Analyzer on top of Nvidia's stack.5 AMD remains Nvidia's most direct competitor across this same landscape.5 The picture that emerges is demand diffusing into ordinary enterprise and even government buyers, not concentrated in a handful of frontier-lab contracts — which is precisely the kind of durable demand signal that supply deals like the SK hynix expansion are meant to protect against.
Then Wall Street Sold the Sector Anyway
None of that stopped a broad late-July selloff. On July 28, the Nasdaq 100 fell to a 2.75-month low even as the S&P 500 (+0.05%) and Dow (+0.738%) were roughly flat to higher, with the divergence explicitly attributed to "a deepening rout in chipmakers and AI-infrastructure stocks."6 Contributing to that day's pressure was Amkor Technology, which forecast third-quarter net sales of $1.95 billion to $2.05 billion — weaker than the $2.11 billion analysts had expected.6 Leverage amplified the move: the ProShares Ultra Semiconductors 2x fund (USD) was up roughly 69% year to date and 196% over the trailing year as of June 5, 2026, a scale of gain that also means a scale of downside once sentiment turns — the same fund's own reporting flagged a 17% implosion in a single stretch.7 Micron Technology captured that swing directly: its stock fell nearly 40% from its all-time high starting in July before bouncing 10% off its lows by August 9 — leaving it still roughly 30% below that peak.8 None of this reads as a verdict on AI demand itself; it reads as leveraged, sentiment-driven money unwinding faster than the underlying supply agreements are changing.
Consolidation and Customer-Side Uncertainty
Two structural shifts are compounding the volatility. Skyworks Solutions and Qorvo named the executive leadership team for their combined company on July 28, with Bob Bruggeworth saying the announcement "reflects the strong partnership that has shaped our integration planning efforts from the very beginning" and that he is confident the named leaders "will help foster collaboration across our teams."9 And on the customer side, Apple announced that Tim Cook will step down as CEO on September 1, 2026, moving to executive chairman, with hardware engineering chief John Ternus taking over — the company's first CEO transition since Cook succeeded Steve Jobs.10
That succession lands alongside a separate technology transition already flagged in the power-management supply chain: Vicor Corporation, which reported trailing-twelve-month revenue of $471.7 million, net income of $136.7 million and a market capitalization of $11.3 billion as of the July 6, 2026 close,11 disclosed that its lead computing customer will transition from Gen 4 to Gen 5/VPD technology in the second half of 2026, with a ramp expected before year-end.12 A major customer's node and product-generation transitions are exactly the kind of near-term execution risk that can weigh on a supplier's numbers even when the customer relationship itself is healthy.
Further down the AI-adjacent supply chain, smaller names told a more upbeat story. Solitron Devices reported fiscal 2027 first-quarter net sales up 101% to roughly $5.44 million, with backlog up 28% to $23.34 million, and 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 that backlog,13 though net income was trimmed by a $0.33 million increase to contingent consideration tied to a large order received after quarter-end.13 Allegro MicroSystems added 30-year semiconductor veteran Brian White to its board, with Joseph Martin citing his "public-company CFO perspective, semiconductor industry experience and governance background" as Allegro — by its own description — leverages three decades of magnetic-sensing and power-IC expertise to serve electrification, automation, AI data-center and robotics customers.14 And in France, Kalray and Bull announced a partnership on Ultra Ethernet-compatible high-speed networking for AI and HPC infrastructure, which Kalray's Éric Baissus called confirmation of "the relevance of our technological vision and the quality of the innovations developed by our teams."15
Via News's Read
Taken together, Via News's analysis of this set of releases points to a market that isn't questioning whether AI infrastructure demand is real — the SK hynix memory lock-in, the federal photonics investment, and the AMD-Cerebras inference partnership all argue it is — but one that is separating structural buildout from company-specific execution risk: an incoming Apple CEO, a major customer's node transition at Vicor, and a guidance miss at Amkor. It's worth noting that several of the individual releases behind this piece are wire-service company statements — the class of source this dossier measures at roughly 22% to 33% claim-accuracy on independent fidelity checks — which is a reason to weigh forward-looking language in press releases (backlog projections, integration confidence) more cautiously than filed, as-reported figures like Vicor's trailing financials.
What to Watch
Ternus formally becomes Apple's CEO on September 1, 2026, the fleet's first leadership succession test since Jobs handed off to Cook.10 The same date marks the expected ramp of Vicor's lead customer onto Gen 5/VPD technology.12 The Skyworks-Qorvo leadership team is named but contingent on the transaction closing. GlobalFoundries' $300 million photonics award is still at the letter-of-intent stage, not finalized. And after its 10% bounce off the lows, Micron remains roughly 30% below its all-time high — the clearest live test of whether the late-July selloff was a leveraged overreaction or the start of a deeper repricing.8