Morgan Stanley used its August global technology webinar to connect three topics that are often discussed separately: memory, hyperscaler financing and SpaceX. The bank’s central argument was that capital formation tied to AI infrastructure is changing how semiconductors and communications are financed, while also shifting the way memory is priced.
In the memory segment, the report focused on three questions: how long-term agreements, or LTAs, are changing pricing mechanisms, where the current cycle sits, and how inventory should be interpreted. As described in the note, LTAs are long-term supply contracts in which memory makers and customers lock in capacity and pricing frameworks in advance, gradually displacing the traditional spot-based model.
LTAs move from broad intent to detailed disclosure
Morgan Stanley called LTAs the clearest structural shift in this memory cycle. The discussion, in its view, has moved beyond whether companies should sign LTAs at all. The questions now are how much capacity to lock up, which customers get allocation and what pricing mechanism is used.
Samsung provided one of the most specific disclosures in the report. The company plans to place 60% to 70% of capacity under rolling five-year LTAs. It has five agreements already signed and five more in final negotiations, with customers including Amazon Web Services, Microsoft, Google, Meta and Oracle.
SK hynix has signed about 10 agreements, though it did not disclose allocation ratios. Micron is targeting more than 50% of revenue under LTAs and has signed 16 agreements. Its customer base includes four large customers and three mid-sized customers, and it has received about $22 billion in prepayments and commitments.
SanDisk is targeting more than 50% of fiscal 2027 wafer capacity and roughly three-quarters of fiscal 2028 wafer capacity under LTAs. It has signed eight customers, with minimum revenue commitments of $93.9 billion and remaining performance obligations, or RPO, of $59.8 billion. Kioxia is aiming to place 50% of sales under LTAs by 2028, guided by a pricing principle the report summarized as price first and profit first.
The pricing model itself is changing. Morgan Stanley said the market is moving away from pure spot exposure toward bilateral structures that include both price floors and price ceilings. Customers may also need to provide prepaid deposits or equipment-backed guarantees. In that setup, price floors give suppliers a revenue base, while ceilings provide customers with cost visibility.
DRAM and NAND pricing is still rising, but closer to peak levels
The bank said 3Q26 DRAM contract prices rose about 15% quarter over quarter, below an earlier expectation of 20%. NAND contract prices rose about 20% from the prior quarter, though the pace of gains is also narrowing. Morgan Stanley expects the slowdown to continue in 4Q26, adding that DRAM contract prices year over year should come off cyclical peak levels.
That does not mean the cycle is over, according to the note. Morgan Stanley said DRAM next-12-month price-to-earnings multiples usually lead next-12-month earnings per share by about two months. It also said the breadth of memory earnings revisions has retreated from extreme levels since late June. In its reading, those conditions are more typical of the middle-to-late phase of upward earnings revisions, not the end of the cycle.
The shift in pricing structure matters as much as the direction of spot prices. Under LTAs, customers still transact at contract prices even if spot prices fall. Morgan Stanley argued that the combination of price protection on both sides should, in theory, leave memory suppliers with lower earnings volatility than in any previous cycle.
Inventory is rising, but Morgan Stanley says the structure is different
Inventory weeks for both DRAM and NAND increased in 3Q26. Supplier inventory remains low, but module makers and downstream customers have built DRAM and NAND inventory over the past several quarters.
On the surface, that can look like classic inventory accumulation. Morgan Stanley drew a distinction. In an LTA framework, customers lock in capacity ahead of time, which amounts to deliberate inventory building. Because these agreements often include prepaid deposits and non-cancellable purchase commitments, the bank said the inventory is backed by real demand rather than speculative hoarding.
The report contrasted current conditions with 2018 and 2022. In those earlier cycles, inventory accumulation came with sharp price declines and forced production cuts by suppliers. This time, Morgan Stanley said, LTAs set both price floors and minimum purchase levels. Even if end demand weakens, suppliers have much better visibility into cash flow than they did in prior downturns.
SpaceX and credit markets are used as evidence of rising capital intensity
Outside memory, Morgan Stanley used SpaceX and credit-market data to reinforce its broader argument about AI infrastructure. The bank said SpaceX posted second-quarter results above expectations and it maintained an overweight rating with a $300 price target.
According to the report, SpaceX is targeting at least $100 billion in total ARR by the end of 2026. For 4Q26 alone, ARR is expected to reach about $22 billion, above Morgan Stanley’s previous estimate of $19.1 billion. Starlink V3 satellites are set to begin deployment starting with Starship’s 14th flight, and the bank expects 1,000 satellites in orbit by the second quarter of 2027, up from its prior estimate of 840.
On capital spending, the note said guidance points to flat second-half spending on a sequential basis, but about $18 billion above Morgan Stanley’s prior forecast. Its sum-of-the-parts target value breaks down into more than $80 for the Space business, $118 for Connectivity, $6 for X and Grok, and $166 for Enterprise AI. The bank also viewed the pullback after the shareholder lockup expiration as an entry opportunity.
In credit, Morgan Stanley’s research team said the combined not-yet-activated lease commitments of five investment-grade TMT companies now exceed $1 trillion. Those five companies — Microsoft, NVIDIA, Google, Meta and Amazon — carry credit ratings ranging from A- to Aaa. Their five-year CDS spreads range from 57 basis points for Amazon to 171 basis points for SpaceX.
The bank said contract-backed and chip-collateralized debt financing is becoming a new area of focus. Large technology companies, in its view, are increasingly using their balance sheets to provide financing support to the AI supply chain instead of relying only on equity financing or internal cash flow. That, Morgan Stanley argued, is another sign that the capital intensity of AI infrastructure is reshaping the technology industry’s funding structure.
Morgan Stanley says the market may still be using an old framework
The report’s broader conclusion is that the logic of the memory cycle is being rewritten by LTAs, while the financing side of AI infrastructure is being reshaped by the balance sheets of major technology companies. Morgan Stanley said market pricing of the memory cycle may still reflect the old framework used for traditional commodity cycles, even as the structural changes in this cycle suggest valuation methods may need to adjust.
The original Chinese article stated that the piece was a compilation and interpretation by Chaoxiang Research of a third-party broker report from Morgan Stanley dated Aug. 11, 2026, combined with public market information. It also noted that the ratings, price targets, earnings forecasts and related judgments cited in the article were the views of the broker’s analysts and represented only the position of that institution, not that of Chaoxiang Research, and did not constitute investment advice.
The byline on the source article was Rita.