Long-term memory contracts are quietly rewriting how the AI hardware supply chain works. Micron has already signed binding multi-year agreements. Korean press reports that Samsung and SK hynix are negotiating similar deals with Google and Microsoft. And a separate, thinner report says Google is discussing pushing its memory commitments as far out as 2035. This piece separates what is documented from what is rumor, shows what the SEC-filing graph can and cannot see, and walks through what a longer lock-in does to the chain.

What Is Documented, What Is Reported, and What Is Chatter

Three tiers of evidence get blended together in most coverage of this topic. They should not be.

Tier 1: Confirmed on the record (Micron). With its fiscal Q3 2026 results, Micron disclosed 16 Strategic Customer Agreements (SCAs). The data center and consumer agreements run five years, roughly calendar 2026 through 2030; automotive agreements run about three. Together they cover roughly 20% of Micron's DRAM volume and about a third of its NAND volume over the period. Micron said 14 of the 16 carry roughly $100 billion in cumulative revenue at minimum contract prices, and it projects about $22 billion in customer cash deposits and related financial commitments. On the largest agreements, price is bounded: a ceiling tied to the calendar Q2 market price for existing products, and a floor for the full term. Micron did not name the customers.

Tier 2: Reported, unconfirmed (SK hynix, Samsung, and Google). In April 2026, TrendForce relayed Korean press reports. Aju News reported that SK hynix is pursuing a five-year commodity DRAM agreement with Google, with a possible two-year extension contingent on next-generation HBM supply. Hankyung reported that SK hynix's long-term negotiations include upfront payments of roughly 10-30% of contract value and a minimum price floor. That report did not tie those terms to a specific customer. Samsung is reportedly in late-stage talks on three-year deals with Google and Microsoft. TrendForce relayed all of this without independent verification.

Tier 3: Chatter (the 2035 figure). The claim that Google is discussing extending memory agreements through 2035 traces to a social media post attributing it to KeyBanc channel checks. We could not find a confirmation from Google, a memory maker, or the bank. It also does not match the reported terms: a five-year deal plus the reported two-year extension starting in 2026 runs to about 2033. Treat 2035 as an unconfirmed discussion point, not a contract.

What the Filing Graph Shows (and Doesn't)

ChainSifter maps relationships that public companies disclose in SEC filings. That makes it a useful test of how visible these contracts are. They are barely visible.

  • Alphabet (GOOGL) has 8 relationship edges in the graph, and none connect to a memory maker. Its edges are suppliers such as NVIDIA (NVDA) at 80% confidence, plus ownership and debt records. Whatever Google has agreed with memory suppliers is not named in the filings we ingest.
  • Micron (MU) has 54 edges. That includes 21 supplier-side and 16 customer-side relationships, 10 "receiving" records, 4 spend records, and debt and ownership entries. NVIDIA is the strongest customer edge at 92% confidence. No hyperscaler appears as a named customer edge.
  • Customer concentration is the only visible trace. Micron's most recent 10-Q excerpt in our evidence base states that revenue from one customer was 13% of total revenue for the first six months of fiscal 2026, versus 15% a year earlier, primarily in one segment. The customer is unnamed.

The lesson for anyone mapping this chain: the hyperscaler-to-memory-maker contracts are the most important relationships in the sector and among the least visible. Named-edge data tells you who the equipment, packaging, and foundry dependencies are. It does not tell you who holds the allocation.

How a Longer Lock-In Changes the Chain

The following is analysis, not reported fact. It describes the mechanisms that long-dated, price-floored, prepaid contracts would be expected to create.

  • Allocation moves out of the spot market. If a large share of capacity is committed years ahead, the remaining uncommitted supply is thinner and more volatile. Buyers outside the club, such as smaller OEMs, automotive, and consumer device makers, absorb the swings. One secondary report says Samsung has roughly 70% of its capacity allocated through 2031. That is unverified, but it illustrates the direction.
  • Visibility supports capex, and capex flows to the equipment chain. Prepayments and committed volumes are what let memory makers commit to new capacity. The beneficiaries are the toolmakers in Micron's graph, including Lam Research (LRCX), KLA (KLAC), and Applied Materials (AMAT). Equipment order patterns tend to confirm or contradict contract rhetoric before management commentary does.
  • Price floors shift risk to buyers. A floor protects the supplier in a downturn. A buyer committed to volumes at a floor price in a falling market holds an above-market obligation. Ceilings cap the buyer's pain in a shortage. The contract is a structured trade-off, not a free lunch.
  • Concentration risk moves, it does not disappear. Memory remains a three-supplier market. In ChainSifter's graph, Micron and SK hynix appear in each other's chains as both supplier and customer. Longer contracts lock buyers to that structure, so a yield problem or disruption at one supplier lands harder on buyers with no flexibility.
  • Counterparty and credit risk lengthen. Multi-year prepayments mean customers are lending to suppliers, and suppliers are betting on customers' demand through the term. Both sides carry exposure that did not exist under annual contracts.

What to Watch in Filings

  • Customer deposits and deferred revenue lines in memory makers' balance sheets, which are where prepayments should show up.
  • Purchase obligations and commitments footnotes in the 10-Ks of the large buyers, which is where multi-year memory commitments may surface in aggregate.
  • Customer concentration percentages in Micron and its peers, as a proxy for how much volume sits with a few buyers.
  • 8-K material definitive agreement items that disclose a contract's existence even when counterparty names are withheld.
  • Equipment maker order and backlog commentary, as the downstream confirmation that capacity is being built against committed demand.

SK hynix and Samsung are foreign issuers and do not file 10-Ks, so their side of these deals reaches the market through earnings releases and press, not the same filing evidence ChainSifter uses for U.S. companies. That is one more reason the Korean-press reports above should be read as unconfirmed.

Bottom Line

Binding multi-year memory contracts are documented at Micron. Comparable deals involving Google are reported but unconfirmed. A 2035 horizon is a single unverified claim that exceeds every reported term. What is certain is the direction: memory buyers and sellers are moving from annual pricing toward multi-year commitments, and the supply chain that serves them is adjusting. For the underlying relationship map, see our earlier analysis of the memory supply chain, HBM, and the MU-SK hynix link.

This article is for informational purposes and is not investment advice. Claims attributed to press reports or third parties are unverified. ChainSifter relationship data is derived from public SEC filings and is subject to extraction confidence limits.