Nomura: AI race supports memory upcycle; 2027 long-term contracts and power bottlenecks determine pace of realization
Tech giants have begun using external financing to bridge the AI investment gap, but Nomura still expects the race to continue. Long-term contracts support 2027 memory shipment volumes and price ranges, but power delays could first cause chip inventories to accumulate, and supply and demand could also loosen in 2028.
Original source：Nomura International (Hong Kong) Ltd.First published on Unhidea：Original source date：
Key takeaway
Nomura believes tech giants will find it difficult to exit the AI investment race. Memory manufacturers will benefit from shortages and long-term contracts in 2027, but financing sustainability, power delivery and supply growth in 2028 are key constraints.
How have the sources of funding for AI investment changed?
Through 2025, tech giants’ operating cash flow could still cover capital expenditure. In 2026, the gap began to be filled by debt issuance, equity financing and suspended buybacks, with shareholders directly bearing the financing and dilution costs.
Alphabet reported its first quarter of negative free cash flow in the second quarter of 2026, at negative 5.9 billion dollars; Meta reported negative 2.1 billion dollars in the third quarter; and Amazon reported negative 7.6 billion dollars for the past 12 months through June 2026. Microsoft still had positive free cash flow of 19.6 billion dollars from month 4 through June of 2026, reflecting differences in their ability to withstand cash flow pressure.
Nomura expects the industry to require more debt and equity financing over the next two to three years. Even if individual investments are profitable, cash outlays may still outpace cash receipts as long as investment grows faster than returns on invested capital.
With cash flow under pressure, why does Nomura still expect the race to continue?
The giants prioritize scarce computing capacity for model training, research and their own products, sacrificing some near-term revenue in exchange for future competitiveness. A company that slows down first could cede its position to rivals and even jeopardize its existing core business, making a coordinated pullback difficult.
After Meta’s free cash flow turned negative in the third quarter, it still signaled active investment in 2027. Its board approved non-voting Class C shares, effective from January 2027, as a financing tool for AI investment. Nomura takes this as a sign that the board is more inclined to absorb the shortfall and keep investing.
Nomura assigns the “race continues” scenario a probability of about 75%: external financing continues, debt rises and the semiconductor supercycle persists. The giants’ cash spending becomes revenue for chipmakers, but financing burdens could weigh on the giants’ share prices; a closure of financing markets is what could actually force the race to stop.
Why are memory manufacturers better supported in 2027?
Most investment decisions for data centers coming online in 2027 were made two to three years earlier; agentic AI has since added demand. Nomura believes supply remains well below planned demand, and long-term contract negotiations completed in October 2026 have locked in shipment volumes and price ranges for the following year.
Long-term contracts raise the cost for customers to exit: canceling commitments triggers a deposit penalty equivalent to 50%–100% of the value of annual demand. Thus, even if the giants cut capital expenditure, Nomura believes the memory business is already well protected for 2027.
From 2028, additional supply and power constraints could loosen the market. Volumes and prices for 2028 are intended to be preset through long-term contracts in 2027 to limit price fluctuations; Nomura is positive about their protection of a floor for margins, but their effectiveness in stabilizing the cycle remains uncertain.
With demand strong, why might chip orders still be cut?
Power could become a bottleneck before chips. A Sightline Climate survey shows that more than 16 GW of new data center capacity is planned for 2026, of which 30%–50% could be delayed, above the 26% delay rate in 2025.
If data centers cannot be powered on in time, chips already produced could accumulate, subsequently depressing orders for the next quarter. Nomura believes this mainly delays and reshapes demand, but short-term orders will still see gaps even if the AI investment race continues.
Power constraints also increase the value of performance per watt. Nomura believes the latest chips and high-capacity HBM will receive greater attention and could generate demand to replace older GPUs. Expansion outside the United States could ease the bottleneck, but the pace and scale of expansion and geopolitical risks in the Middle East constrain this option.
