The short version
On October 1, 2026, Micron told analysts that customers have already committed to more than 75% of its 2027 output — and that most negotiations have moved on to 2028 volumes. Twenty-six long-term agreements now carry $32 billion in customer financial commitments, up from 16 agreements and $22 billion in June. Remaining performance obligations stand at roughly $150 billion, up from about $100 billion last quarter. CEO Sanjay Mehrotra said memory and storage supply-demand in fiscal 2027 and 2028 will be "much tighter" than 2026, adding that he cannot see when balance will return. For buyers without a long-term agreement, the message is blunt: the 2027 allocation window at the world's third-largest memory maker is three-quarters closed before 2026 ends. This article explains what got locked, how it squeezes the spot market, and what procurement teams without LTAs can still do. This article is for informational purposes only and does not constitute investment or procurement advice.
What Micron actually said on October 1
All figures below come from Micron's fiscal 2026 fourth-quarter earnings call on September 30, 2026 (US time), as reported by 财联社 and ChosunBiz. On supply-demand, CEO Sanjay Mehrotra said memory and storage conditions in FY2027 and FY2028 will be "much tighter" than 2026, and that the company cannot see the inflection point when balance returns — stronger language than his June statement that tightness would extend beyond 2027. On HBM, he said demand growth will keep outpacing conventional DRAM through 2028, and that most of Micron's 2027 HBM supply is already contracted at prices up sharply year over year. On long-term agreements: 26 signed (up from 16 in June), $32 billion in customer financial commitments (up from $22 billion, mostly cash deposits), some running through 2031 and covering over 35% of expected revenue through 2030. CFO Mark Murphy added that remaining performance obligations reached roughly $150 billion (up from ~$100 billion), more than 75% of 2027 shipment volume is committed, and most current discussions concern 2028 volumes. COO Manish Bhatia called memory "the primary constraint on AI development," noting data centers are now the largest memory market. New fabs in Japan and the US are expected to output first wafers around mid-2027 but need several quarters of ramp. Fact-grade note: these are company statements on a public earnings call — treat as company-disclosed, not independently audited. The 75% figure describes Micron's order book, not the whole memory market; Samsung and SK hynix have disclosed no equivalent numbers.
Why 75% locked matters more than the headline
A single supplier's order book doesn't set the market. What makes this number matter is the mechanism. First, allocation math: Micron is one of three DRAM makers that matter. When the number-three player removes three-quarters of its 2027 volume from contestable supply, the remaining pool — spot distributors, smaller contract buyers, second-tier OEMs — competes for a smaller pie at all three suppliers, because the same hyperscalers locking Micron capacity are locking Samsung and SK hynix capacity too. Second, cash deposits change behavior: $32 billion in mostly cash-backed commitments means locked buyers take their volume first in a shortage — that is what the deposit is for. Spot buyers face allocation priority inversion: contracted tons ship, spot tons wait. Third, the negotiation window has moved: Murphy's line that most discussions concern 2028 volumes means 2027 pricing for non-LTA buyers is largely a residual outcome — decided by what contracted buyers don't take. Fourth, HBM is the binding constraint: every HBM stack consumes wafer starts that could have been conventional DRAM, so buyers who never touch HBM still pay for the HBM boom through tighter DRAM supply — the same spillover mechanism described in our memory supercycle analysis, now with company-confirmed numbers attached.
The counter-argument: is the LTA wave overheating
Three reasons to treat the lock-up as possibly overstated. Double-ordering risk: cash deposits discipline buyers, but the 2021 cycle showed fear-driven over-ordering inflates apparent demand. $150 billion in remaining obligations is real liability on customer balance sheets — if end demand softens, cancellations and push-outs follow and "committed" volume re-enters the market. One supplier's book is not the market: Samsung and SK hynix serve different customer mixes, and a buyer locked out of Micron's 2027 book may still find room elsewhere — at a price. New capacity is real but slow: first wafers around mid-2027 plus several quarters of ramp makes late-2028 relief plausible. Our read: the lock-up is real enough to plan around and uncertain enough not to panic-buy at any price. The playbook below is built for that middle ground.
Procurement playbook: five moves without an LTA
One — qualify second-source DRAM now. If your BOM is single-sourced on a Micron part, the 75% figure is your deadline; cross-qualify Samsung or SK hynix equivalents while distributors still have sample stock. Two — treat 2027 as two markets. Price your 2027 BOM on residual-market assumptions (higher, more volatile), not 2026 contract comps; if finance still models 2026 pricing into 2027, this earnings call is the document that kills that assumption. Three — shorten RFQ validity windows to 7–14 days and re-quote on cadence; stale quotes in a fast-moving allocation market commit you to prices the market left behind. Four — build buffer on non-memory BOM: memory is the constraint, so use the current window to stock MCUs, passives, and power components before AI-adjacent demand spills further. Five — talk to your distributor about allocation tiers, not just price: get your tier in writing, because in allocation markets priority beats price.
Frequently asked questions
Does 75% locked mean I cannot buy Micron DRAM in 2027? No — it means three-quarters of planned 2027 shipments are promised to specific customers. The rest plus any production upside is contestable, but you will compete for it at market prices. Should I sign an LTA now? Only if you can forecast 2027–2028 demand within about 20% and stomach deposit terms; LTAs signed in panic at cycle peaks are how buyers end up over-committed when the cycle turns. How is this different from 2021? In 2021 the constraint was broad logistics plus a demand spike and corrected in about 18 months. This cycle's constraint is structural: AI data centers are the largest memory market, HBM consumes leading-edge wafer starts, and $32 billion in cash-backed commitments means contracted demand cannot easily evaporate. What about NAND? Mehrotra's "much tighter" language covered DRAM and NAND; DRAM is the sharper constraint — prioritize DRAM coverage first.