Weekly Market Commentary for Long-Term Investors

The rotation held its shape but the intensity cooled, and by week's end a soft jobs report flipped the entire macro narrative on its head. The Dow led again with a 1.37% weekly gain, the S&P 500 added 0.58%, and the Nasdaq essentially flatlined at +0.05% — a third consecutive week of the real economy outrunning the AI trade. Healthcare extended its remarkable run with another +5.30%, followed by Communication Services (+4.55%), Consumer Cyclical (+4.06%), and Financials (+3.28%) as the broadening deepened across cyclicals and defensives alike. Technology (−0.35%), Utilities (−0.45%), and Energy (−1.09%) lagged. Then Thursday's payrolls report changed everything: the U.S. added just 57,000 jobs in June versus 115,000 expected, with April and May revised lower. The weak data gutted the rate-hike thesis that had dominated since Warsh's hawkish pivot — the dollar sank to a two-week low, gold jumped 1.5% to $4,188, and Treasurys rallied. Oil sat near prewar levels at $72 Brent, with futures in contango signaling a near-term supply glut. The market that spent three weeks fearing rate hikes closed the holiday week wondering if it had the story backwards.

Micron's blockbuster profits are, viewed from the other side of the transaction, everyone else's blockbuster costs. Memory prices have roughly quadrupled in a year, and the chip makers — Micron, Samsung, SK Hynix — are now to AI what oil producers are to the airlines: suppliers of an essential input that suddenly became far more expensive. The crucial twist is that AI companies can't easily pass the cost on, because they're still pricing to win market share, not to make money. So the burden lands on the model makers and hyperscalers, squeezing margins higher up the stack while profits migrate down to whoever controls the scarcest input. For investors, the lesson is that the AI boom isn't lifting all layers equally — and owning "AI" through hyperscalers may mean holding the exact part of the stack currently losing the margin war.

A single report about OpenAI delaying its IPO knocked 4% off Japan's market and 6% off Korea's — while U.S. broad indices barely moved. That divergence is the story: when a few AI and chip names dominate index weight, theme-specific news becomes a whole-market event. The selloff stayed contained to the AI complex this time, but it arrived against a mixed macro backdrop — the hottest core inflation reading since early 2023, a Fed official penciling in a year-end hike, and the biggest IPO ever briefly slipping below its listing price. For investors, the takeaway isn't that AI is finished. It's that AI-driven concentration has made the market's risk profile less diversified than headline index levels suggest — and that fragility is structural, not a one-off scare.

This first study explains the compute layer of AI infrastructure and why investors should not look at GPUs alone. It breaks down the role of GPUs, ASICs, and CPUs, explains the difference between training and inference, and shows why hyperscalers still invest heavily in custom chips even in a GPU-dominated market.

This second study explains the memory layer of AI infrastructure and why the next bottleneck often moves from compute to memory. It breaks down the roles of HBM, DRAM, and SSD, and shows why memory bandwidth has become a critical constraint in large-scale AI systems.

This third study explains the networking layer of AI infrastructure and why connecting chips matters as much as the chips themselves. It breaks down scale-up vs scale-out networking, compares NVLink, InfiniBand, and Ethernet, and shows why networking shapes cluster performance and scaling efficiency.