Nvidia, Broadcom, and AMD sold off in sympathy with SK Hynix's slump, raising fears that demand signals from the memory supply chain are flashing caution on AI infrastructure spending. The second-order question is whether this is a tradeable dip in structurally strong franchises or an early read on a peak-cycle air pocket.
Nvidia, Broadcom, and AMD sold off in sympathy with SK Hynix's slump, raising fears that demand signals from the memory supply chain are flashing caution on AI infrastructure spending.
The SK Hynix-led selloff puts NVDA, AVGO, and AMD at a fork: is this a supply-chain read on genuine AI capex deceleration, or a dip in franchises whose underlying demand is still compounding?
A broad AI re-rating rally driven by a blowout hyperscaler capex print would likely lift AMD more violently on short-cover, collapsing the spread against this pair.
CoverageFirst reported by GuruFocus at 8:25 AM ET · the only report so farHow this is decided →
SK Hynix — the primary HBM supplier to Nvidia's H-series GPUs — slumped, triggering a broad sympathy selloff across U.S. AI chip names led by Nvidia (NVDA), Broadcom (AVGO), and AMD. The read-through is straightforward: if HBM demand were accelerating, Hynix wouldn't be under pressure, so the market is pricing in at least a near-term pause in AI hardware orders.
The three names caught in the move are not equals. Nvidia posted $215.9B in revenue growing 65.5% YoY with a 71.1% gross margin and $4.90 diluted EPS — a financial profile that reflects dominant pricing power in AI accelerators. Broadcom's $63.9B top line grew 23.9% YoY at 67.8% gross margins, underpinned by its custom ASIC (XPU) business serving hyperscalers directly. AMD, with $34.6B in revenue up 34.3% YoY, carries a noticeably thinner 49.5% gross and 12.5% net margin, reflecting heavier competition and lower AI revenue mix than the other two.
The bull case rests on the idea that HBM supply constraints — not demand softness — are Hynix's real problem, and that Nvidia's Blackwell ramp cycle is still in early innings with $4T+ in hyperscaler capex commitments publicly on record. Broadcom's custom silicon business insulates it from GPU cycle timing, and AMD's MI300X is still gaining enterprise traction.
The bear case is that memory suppliers historically see demand shifts before GPU vendors do, and any pause in hyperscaler ordering ripples quickly into this concentrated supply chain. AMD is most exposed given its thinner margins, smaller AI revenue base, and the fact it's still winning share — meaning it has the most to give back if the cycle stalls. NVDA and AVGO, with superior margin structures, are better positioned to absorb a correction but are not immune.
The key things to watch: Hynix's earnings commentary on HBM order visibility, any revision to hyperscaler capex guidance from Microsoft, Meta, Alphabet, or Amazon in upcoming prints, and whether NVDA's next earnings guide shows any deceleration. Until those datapoints land, the selloff is signal-but-not-confirmation.
The pair trade — long NVDA vs. short AMD — exploits the quality gap exposed by this selloff: NVDA's 71.1% gross and 55.6% net margins dwarf AMD's 49.5%/12.5%, meaning NVDA absorbs a demand air pocket far better. If the selloff deepens, AMD's thinner margin buffer and smaller AI revenue base make it more vulnerable to multiple compression; if it reverses on strong hyperscaler capex commentary, NVDA's Blackwell positioning captures more of the upside.
The read above, as written. kept as written · closes shown from JUL 13 on
3-6 weeks, into next hyperscaler earnings cycle. Follow to be told when one lands.
Price context does not establish that the story caused the move.
NVDA's 65.5% YoY revenue growth, $4.90 diluted EPS, and Blackwell ramp — backed by publicly committed hyperscaler capex — suggest the Hynix selloff reflects HBM supply dynamics, not a genuine Nvidia demand pause.
SK Hynix has historically been a leading indicator of GPU order cycles, and AMD's razor-thin 12.5% net margin leaves it (and sector sentiment broadly) highly exposed if hyperscalers trim AI hardware budgets even modestly into 2H 2025.
Kept as written · your side, if you take one, is graded privately against licensed closes after 10 trading days · nothing here is advice · How the Wire is made →
Shares a name with this story — discovery, not a connection.
This page is kept as it was written on Jul 13. Later coverage joins it only when the company and catalyst evidence match, and what the stock did is shown from licensed end-of-day closes — never re-graded, never backdated. The judgment is yours.