A year ago, the idea that Micron might be compared with Nvidia on market value would have sounded far-fetched. On May 9, that comparison stopped looking absurd.
Micron shares jumped 15% in a single session to close at $746.81, lifting the company to a market value of roughly $853 billion. Over the past 12 months, the stock has risen nearly 700%, outpacing even Nvidia's run.
CEO Sanjay Mehrotra summed up the shift in an interview: "Memory has become a strategic asset for customers."
For a 47-year-old memory maker, that phrase matters more than the stock chart. It says the industry is no longer treating memory as a commodity input that rises and falls with the cycle.
The numbers are already startling
Micron's second-quarter figures show how quickly the business has changed:
- Revenue reached $23.86 billion, almost triple the $8.1 billion reported a year earlier.
- GAAP net income came in at $13.79 billion, or $12.07 a share.
- Third-quarter guidance rose to $33.5 billion, plus or minus $750 million.
The company is also preparing more than $25 billion in capital expenditure for fiscal 2026. Micron's previous capex peaks were around $12 billion, so the new plan is not a routine increase. It is a doubling of the company's old investment ceiling.
The third-quarter outlook is even more unusual. Moving from $23.8 billion to $33.5 billion in one quarter implies roughly 40% sequential growth, a slope rarely seen in semiconductors, especially from a mature manufacturer with nearly five decades of operating history.
Memory has moved into allocation mode
This is not a normal memory-price upcycle.
Mehrotra has said that AI demand could consume more than 50% of global memory supply in 2026. That leaves the rest of the market, including phones, PCs, cars, servers and consumer electronics, competing for what remains.
HBM, the high-bandwidth memory used alongside AI accelerators, is even tighter. All HBM supply for 2026 has already been locked into multi-year fixed-price contracts. Buyers that need more capacity are effectively negotiating for 2027 and beyond. Micron, SK Hynix and Samsung are all discussing multi-year allocations with customers that want guaranteed supply.
That is why the rally has spread beyond HBM. Sandisk, a NAND-focused supplier, rose 16% on the same day because scarcity is spilling into the broader storage and memory market.
Mizuho lifted its target from $545 to $740
Analysts have moved quickly. Mizuho's Vijay Rakesh raised his price target to $740 from $545, citing one central reason: agentic AI is pushing memory demand higher.
The logic is straightforward:
- AI agents need long-horizon reasoning and larger context windows.
- Larger context windows require more memory for each inference request.
- Inference workloads can be more than 10 times larger than training because users run them every day, while training is episodic.
- Every new agent application therefore becomes a continuous consumer of memory.
Training is different: once a model is trained, that specific demand comes off the system. Inference runs around the clock. As AI moves from training into large-scale agent deployment, memory starts to look less like a one-time sale of compute capacity and more like a recurring infrastructure toll.
That is the substance behind Mehrotra's phrase. Memory companies used to be treated as cyclical stocks. Micron is now trying to be valued like a long-duration AI infrastructure supplier.
Bernstein added the caution
Not everyone is chasing the trade. Bernstein warned that price momentum could slow before the third quarter of 2026, pointing to early signs of softer demand in some segments even as short-term contract prices keep rising.
The question is real: if HBM prices are locked by multi-year contracts while capacity keeps expanding, does supply catch up first, or do alternatives arrive first?
Nvidia's Blackwell Ultra uses HBM3e, and Rubin is expected to move to HBM4. SK Hynix and Samsung are accelerating HBM4 production as well. If all three major suppliers open capacity in 2027, and CXL memory pooling lets systems do more with less HBM, part of today's scarcity premium could fade.
That is a 2027 question. For the rest of 2026, AI data-center operators face a simpler reality: if they want HBM, they need to sign early and wait in line.
The bigger signal
Nvidia has been the central AI stock for three years because compute was the bottleneck.
But once Mythos-class AI agents with long-horizon reasoning are deployed at scale, the bottleneck starts moving from the GPU to the HBM sitting beside it. An H200 can deliver enormous compute, but if memory capacity and bandwidth are constrained, inference throughput suffers as data moves in and out.
The key question for the next 12 months is not only how much higher Nvidia can go. It is who controls the HBM supply chain. Micron, SK Hynix and Samsung may become an even narrower choke point.
A 47-year-old memory company does not rise nearly 700% in a year by accident. It rises because money is being redistributed across the AI stack. The next beneficiaries are becoming easier to identify.
Sources: Micron Stock Rally: AI Memory Shortage Sends Market Value Past $850 Billion (ts2.tech); Micron zooms past $700 billion market cap as rally in memory stocks accelerates (CNBC); CocoLoop; Wall Street AI chip love moves from Nvidia to Intel, AMD and Micron (CNBC)