Analysis
Nvidia's own numbers, on their face, look like the best quarter any semiconductor company has ever posted: $81.6 billion in total revenue, up 85% year over year, with data-center revenue alone at $75.2 billion, up 92%, and its Blackwell architecture sold out through the middle of 2026. Yet as of July 6, Nvidia's stock is up just 3.2% for the year -- making it, by a wide margin, the worst-performing major name in a chip sector that is otherwise on fire.
The contrast with its peers is stark. AMD has gained roughly 171% year-to-date, and Micron -- a memory supplier, not even a chip designer competing head-on with Nvidia's GPUs -- is up around 305%. For a company still viewed as the single biggest beneficiary of the AI buildout, that gap is the story: investors are no longer rewarding Nvidia simply for being the incumbent, and are instead chasing the names seen as having more re-rating room left.
The structural threat is Broadcom, which designs custom AI accelerator chips (ASICs) for Alphabet and Meta rather than selling merchant GPUs the way Nvidia does. Nvidia still commands an estimated 81% of the overall AI chip market, but Bloomberg Intelligence has forecast custom silicon growing at a 27% compound annual rate through 2033, nearly double the projected 16% growth rate for merchant AI accelerators like Nvidia's -- meaning the largest hyperscalers are increasingly building their own chips rather than buying Nvidia's, even as they remain Nvidia's biggest customers for now.
“AMD has gained roughly 171% year-to-date, and Micron -- a memory supplier, not even a chip designer competing head-on with Nvidia's GPUs -- is up around 305%.”
Compared to the last major inflection in chip-market leadership -- Nvidia's own displacement of Intel as the AI-era's default silicon vendor -- this shift is quieter and slower, playing out inside hyperscaler capex budgets rather than in a single dramatic product launch, but the direction is the same: customers with enough scale and capital eventually build in-house rather than keep paying vendor margins indefinitely.
Nvidia is not standing still on the supply side. The company announced a new multiyear memory technology partnership with SK hynix aimed at securing next-generation memory supply for AI-factory-scale deployments, a defensive move to keep memory bottlenecks -- rather than compute bottlenecks -- from constraining its own growth as demand keeps outstripping supply.
For infrastructure investors, the split matters: Nvidia's fundamentals are not in question, but its valuation re-rating has already happened, while AMD, Micron and Broadcom are being priced for a future in which Nvidia's 81% share erodes meaningfully. The bear case for chasing AMD and Micron's run is that a sector-wide slowdown -- the exact scenario the Bank for International Settlements flagged in its AI-bubble warning -- would hit the higher-multiple, faster-moving names harder than it hits Nvidia's already-discounted stock.
What to watch: whether Broadcom's custom-ASIC design wins with Alphabet and Meta expand to additional hyperscalers, whether Nvidia's SK hynix memory partnership meaningfully eases supply constraints, and whether Nvidia's next earnings report narrows or widens the valuation gap with AMD and Micron.
