Nomura: There is a 75% probability that the AI arms race will continue, and the 2027 memory chip shortage is basically locked in.

Nomura: There is a 75% probability that the AI arms race will continue, and the 2027 memory chip shortage is basically locked in. The AI arms race is hard to brake, and the tight supply-demand structure for memory chips in 2027 is also basically locked in. Nomura's global memory research team noted in its latest report that tech giants continue to ramp up AI investment and are unwilling to be the first to exit the competition even if free cash flow comes under pressure; at the same time, long-term memory supply agreements have locked in 2027 volumes and prices in advance, providing support for the industry's prosperity. Nomura calls this competition a "Game of Chicken": any company that is the first to cut investment may fall behind in the next round of AI competition. As long as competitors are still raising their bets, maintaining investment becomes the more realistic choice, even at the cost of continuously burning cash flow. This competitive landscape is translating into opportunities for the semiconductor industry. TSMC and several tech giants have said AI demand still exceeds existing supply capacity; order lock-ins and default clauses in long-term agreements also provide some protection for memory suppliers. Nomura expects the probability that the AI investment race continues and the semiconductor super-cycle extends to be about 75%. What could truly break this setup is a closure of financing channels; power shortages may delay chip deployment and become another key risk. The AI arms race is hard to brake: no one is willing to be the first to exit Even if AI investment can deliver considerable return on invested capital (ROIC), as long as investment growth continues to outpace the pace at which returns are realized, companies may face negative free cash flow for a prolonged period. Statements from tech giants confirm this trade-off. Google CEO Sundar Pichai said computing power for frontier model training has higher priority than cloud business revenue; Microsoft CEO Satya Nadella said that if all new GPUs were used for Azure, the cloud business would grow by more than 40%; Meta CEO Mark Zuckerberg said that even if superintelligence development is later than expected, the additional computing power can be redirected to advertising and recommendation systems. Companies are investing not only for current revenue but also for future competitive position. Mature businesses such as advertising and cloud services can still generate cash flow, and continued investment by competitors makes it difficult for any company to be the first to slow down. As long as financing capacity remains, short-term cash flow pressure may not be enough to force tech giants out of the race. The 2027 memory shortage is basically locked in: supply and demand sides are converging in their judgments AI demand continues to exceed supply capacity. TSMC CEO C.C. Wei said on October 15 that AI demand still exceeds current supply; Amazon CEO Andy Jassy said that even if capital expenditure rises to $220 billion in 2026, the company still cannot meet all demand in 2026-2027, and part of 2028 demand has already been booked in advance. Microsoft and Alphabet management also said their cloud businesses are being constrained by capacity. Nomura noted that 2027 memory supply volumes and prices were locked in through long-term agreements (LTAs) in October 2026, and prices have rebounded. Even if customers cut capital expenditure, default clauses in contracts will raise the cost of canceling orders, and the defaulting party may need to pay a deposit equivalent to 50% to 100% of the annual demand value. Advance order lock-ins and default costs provide memory suppliers with strong demand visibility and some profit protection. However, this judgment mainly applies to 2027; supply and demand in 2028 and beyond will still depend on the pace of AI investment and subsequent contract arrangements. Worsening cash flow is unlikely to stop investment: tech giants turn to external financing Capital expenditure has begun to exceed operating cash flow at some tech giants. Nomura report data show that Alphabet's free cash flow in the second quarter of 2026 was negative $5.9 billion, Meta's was negative $2.1 billion in the third quarter; Amazon's free cash flow for the trailing 12 months ended June 2026 was negative $7.6 billion. Microsoft still had positive free cash flow of $19.6 billion in the same period, but Nomura expects it will also come under pressure later. Companies still plan to maintain heavy investment: Alphabet expects capital expenditure of $195 billion to $205 billion in 2026, Meta $135 billion to $145 billion, Amazon about $220 billion, and Microsoft about $175 billion in fiscal 2027. Financing has become an important means of filling funding gaps. The report shows that Alphabet raised $49.6 billion in June 2026 through common stock and mandatory convertible preferred stock; Meta issued $24.9 billion of bonds in the second quarter and suspended share buybacks. Meta's board also approved the issuance of non-voting Class C shares to fund AI investment. Nomura divides the future path into three scenarios: the race continues with about a 75% probability, capital expenditure slows with about a 20% probability, and financing fails with about a 5% probability. Among them, a closure of financing channels is the risk most likely to force companies into a passive contraction and thereby break the current investment landscape. Power bottlenecks may delay demand, but may not end the super-cycle In addition to financing risk, insufficient power supply may also limit AI infrastructure expansion. A Sightline Climate survey shows that about 30% to 50% of planned new data center capacity in 2026 faces delay risk, and more than 11 GW of capacity has no sign of construction starting. If data centers cannot connect to power on schedule, chips already produced may not be deployed in time, causing delivery delays and periodic order volatility. However, Nomura believes power shortages are more likely to delay and reshape demand rather than directly eliminate it: under constrained power supply, the per-watt performance value of next-generation chips and high-capacity high-bandwidth memory (HBM) may increase further, and older GPUs may also face replacement demand. For investors, the key is to distinguish between delayed demand and disappearing demand. As long as tech giants can continue to finance, the AI investment race may continue, and the 2027 memory supply side is also expected to benefit from advance order lock-ins; what truly warrants vigilance is a substantial contraction in investment and chip demand caused jointly by worsening financing conditions and power bottlenecks. This article is reprinted from "Wallstreetcn", edited by GMTEight: Feng Qiuyi.