Micron: 2027 Set to Be Even Tighter Than 2026 as Customers Get Less Than Half the DRAM They Need
KeyBanc Technology Leadership Forum, August 10, 2026
Micron's chief business officer delivered one of the most bullish supply-demand updates the memory industry has seen in years, telling investors at KeyBanc's Technology Leadership Forum that demand signals from customers have strengthened materially since the company's last earnings call, to the point where calendar 2027 is now expected to be tighter than 2026. Sumit Sadana, Micron's Executive Vice President and Chief Business Officer, was blunt about the mismatch: "We still don't have line of sight as to when the supply is going to be able to meet demand because demand continues to escalate at a very rapid pace over time."
Aggregate Demand Accelerating Since Earnings
Sadana said the shift in tone is not confined to one segment. "The aggregate demand across all of the segments is at much higher levels than it has been," he said, adding that customers are frustrated even after signing long-term contracts because they feel they are "leaving considerable opportunity on the table" across market segments, not just in data centers. In the data center specifically, he said Micron is "quite often" unable to meet more than half of customer demand, and characterized the signals from customers as multiyear rather than a single-year phenomenon tied to 2026 alone.
Strategic Customer Agreements: The New Financial Architecture
The most consequential structural disclosure centered on Micron's Strategic Customer Agreements, a mechanism Sadana said the company pioneered and has used more aggressively than any peer. As of the last earnings call, Micron had signed 16 SCAs carrying $22 billion in cash and cash-like commitments, of which $18 billion is pure cash sitting on the balance sheet upfront. Sadana confirmed the company has signed additional SCAs since then. Unlike the legacy long-term agreements that defined prior cycles, these are five-year, take-or-pay contracts running through the end of calendar 2030 for the bulk of covered revenue, with no contractual outs for customers. "Those LTAs were somewhat of a misnomer because there was nothing long term about them," Sadana said, contrasting the old 12-month handshake arrangements with the new binding structure. Pricing terms are also new territory: some volume carries floating, market-based pricing, while the majority operates under a price band with a ceiling tied to the pricing environment at signing (the initial 16 SCAs were priced off calendar Q2 levels). Sadana said the flow pricing embedded in these deals is calibrated to generate gross margin "well above any prior peak in the cycles of the industry." The agreements also carry an evergreen structure allowing additional years to be appended over time, which Sadana said reflects customers' own expectation that supply shortfalls will recur well beyond this cycle, extending into robotics and edge AI in the next decade.
Addressing the Skeptics: Why These Contracts Won't Break Like COVID-Era LTAs
Pressed by KeyBanc analyst John Vinh on why these SCAs won't be abandoned the way pandemic-era LTAs were once supply normalized, Sadana leaned on both contractual mechanics and incentive structure. Beyond the absence of contractual outs, he argued that customers who treat memory "tactically" during a shortage will simply find themselves starved of allocation the next time a shortage hits — and he was clear he expects more shortages to come, driven by AI's spread from data centers into consumer devices, autonomous vehicles, and robotics. "This is not the last time you're going to see a significant deficit of supply in the industry," he said.
HBM Trade Ratios Are Getting Worse, Not Better
Sadana quantified the DRAM supply drain caused by HBM production more precisely than Micron has in the past. Producing 100 bits of HBM3E currently requires giving up roughly 300 bits of standard DDR supply, a 3:1 trade ratio. That ratio worsens to nearly 4:1 by the time the industry reaches HBM4E, meaning every increment of HBM growth increasingly cannibalizes supply available for the rest of the memory market. He also predicted the HBM market itself is evolving into something closer to the ASIC business model — effectively single- or dual-sourced per customer or platform, given how R&D-intensive and time-consuming co-design and qualification have become. "It is not practical for our customers to be able to do business with all 3 HBM suppliers for all projects," he said, adding that this single/dual-sourcing dynamic is spreading beyond HBM into other parts of Micron's portfolio, which he framed as "incredibly powerful from an ROI perspective long term."
Despeccing Is a Supply Story, Not a Pricing Story
Addressing investor concern that server memory despeccing signals demand destruction from high prices, Sadana pushed back directly, attributing the trend entirely to supply scarcity rather than affordability. Customers are trading off average DRAM capacity per unit against total unit shipments simply because they cannot secure enough DRAM to do both, he said. Critically, he noted that aggregate demand is still rising even after accounting for this lower average capacity, and framed the situation as latent demand that will resurface as higher-capacity SKUs once supply allows — meaning any despeccing narrative should be read as a supply constraint rather than a demand signal for investors to worry about.
Physical AI: A Growth Vector That Could "Dwarf" Today's AI Opportunity
Perhaps the most striking forward-looking comment concerned physical AI and robotics, which Sadana said remains in its infancy but could eventually eclipse the current AI-driven memory boom. He pointed to humanoid robots as a specific example, noting each unit is expected to require hundreds of gigabytes of DRAM and terabytes of SSD storage to enable fast response times and functionality without constant cloud connectivity. Multiplied across an expected decade-plus of deployment scale, he said the aggregate capacity requirement "is going to just be astronomical," predicting the category will hit an inflection point in the early part of the next decade and become another area where Micron "will struggle to supply for years to come."
U.S. Manufacturing as a Pricing Lever
Sadana used the forum to sharpen Micron's competitive positioning against Samsung and SK Hynix's U.S. expansions, characterizing both as fundamentally different from Micron's footprint. He said Samsung's Texas investment is in logic foundry capacity, not memory, while SK Hynix's Indiana project is back-end assembly and packaging rather than front-end fabrication. Micron, by contrast, has front-end memory fabs coming online domestically — Idaho 1 in mid-2027, Idaho 2 by the end of 2028, plus New York and Virginia facilities. That distinction, Sadana said, has translated into a concrete pricing benefit: Micron expects a premium for U.S.-manufactured supply, and that premium is already built into its SCA pricing structures, reflecting customer willingness to pay for supply chain resilience.
Investment Ramp Reaffirmed at $250 Billion
Micron's U.S. capital commitment has been increased from $200 billion to $250 billion over the coming years, according to Sadana, supplemented by supply chain investments including a $500 million stake in GlobalWafers for raw wafer supply as part of a broader $3 billion supply chain commitment. Even with this scale of investment, Sadana reiterated that Micron cannot forecast when supply will catch up with demand, underscoring that the industry's structural bottleneck is less about capital availability and more about the multiyear lead times inherent in building and ramping leading-edge fabs from greenfield sites.