Samsung to allocate 40% of DRAM output to AI memory by 2027

Fitch, Morgan Stanley see Asia-Pacific AI data-center growth

KB Securities analysts, led by Jeff Kim, wrote that Samsung Electronics is expected to dedicate 40 percent of its total DRAM production capacity to high-bandwidth memory products for artificial-intelligence applications in 2027, up from 27 percent in 2025. The analysts wrote that Samsung’s rapid expansion of HBM production could squeeze supplies of the company’s general-purpose memory chips, with the share devoted to those products projected to fall to 65 percent in 2026 and 59 percent in 2027, down from 73 percent in 2025. Samsung’s HBM4 products could account for about 80 percent of its total revenue next year, up from 40 percent this year, the analysts added.

Two analyst notes pointed to Asia-Pacific markets as alternatives to constrained U.S. data-center development. Fitch Ratings said Japan’s data-center sector is entering a new phase as AI workloads become more power-intensive. Access to electricity is increasingly influencing development timelines, site selection and project viability in Japan, the rating company said. Growing power and land constraints in Tokyo are encouraging a redistribution of future development toward locations better positioned to support AI infrastructure and rising electricity demand, Fitch added. Community resistance is becoming more visible in some urban developments, adding another layer of execution risk, and government policy is becoming increasingly supportive of broader geographic distribution of digital infrastructure, the company noted.

Morgan Stanley analysts wrote that Australia is well placed to benefit from constraints on how rapidly U.S. data centers can be built to meet surging AI demand. Australia could accommodate part of the demand overflow, the analysts said, pointing to its data sovereignty requirements and renewables-assisted capability to power the centers. Australia already ranks in the global top five for installed data-center capacity, they noted. Large-scale and household batteries are increasing Australia’s energy storage capacity and damping forward electricity prices, while U.S. data-center energy costs are becoming increasingly expensive, they added.

Meta’s stock gained 12 percent on Monday after Sensor Tower data showed the app for the company’s new Muse AI agent had become the most-downloaded free iPhone app in the United States, The Wall Street Journal reported. Shopify shares rose 7.3 percent to $137.92 after the company said it would allow Muse to complete purchases through Shop Pay, its one-tap checkout service. Deutsche Bank analysts said the collaboration is further evidence that leading AI platforms are integrating with Shopify’s commerce infrastructure. “We believe the partnership helps refute the recent bear thesis that AI agents and personal shoppers could disintermediate Shopify and its large payments business,” they wrote, adding that the deal demonstrates the value of Shopify’s “structured product data, merchant connectivity, checkout and the underlying infrastructure required to complete an order.”

Truist analysts projected that Muse will generate $28.5 billion in incremental revenue for Meta by 2030 in a base-case scenario, calling the product “the clearest attempt yet to build a non-ad revenue stream to complement its ad juggernaut and show ROI against its large CapEx.” In a separate note, Truist analysts raised concerns about the trust pitch Meta is making for Muse. “Muse asks consumers to trust Meta with more personal data than social media ever did, and it arrived twelve days after Meta agreed to an $18 billion settlement with 29 states over social media harms to children,” they wrote. “For a product whose entire pitch is trust, this is not the most favorable launch context.”

The AI trade’s continuing strength came alongside a public debate over how the technology should be regulated. The CEOs of OpenAI and Anthropic signaled that the U.S. government may need to take an active role, according to The Wall Street Journal. Treasury Secretary Scott Bessent publicly disagreed on CNBC. “What did they try to do last week? It was, ‘Well there’s a 10% chance that we could destroy the world, but we want the government to give us a liability shield.’ That’s good business for them, bad business for the American people,” Bessent said.

Capital Economics head of markets for Asia Pacific Thomas Mathews wrote that “we don’t think the fall in tech stocks’ relative valuations means the AI bubble has burst yet, and the rally may have a bit further to go.” If oil prices keep falling and bring yields with them, “a tech-led relief rally is therefore possible even though the index hasn’t fallen that much,” he added. Mathews said there is “potential for market-moving news at the U.S.-China meetings” but predicted “progress, and any resultant market reaction, will be minor.”

Private-equity firms have meanwhile reduced their investments in software compared with previous years, favoring asset-heavy businesses whose products and services are considered less prone to replacement by AI, according to a report by the law firm Sidley Austin. The technology sector represented roughly 13 percent of the value of U.S. buyouts this year through June, down from an average of about 30 percent in the five-year period through last year, the firm said. “The pullback in activity also illustrated potential vulnerability in software valuations,” Sidley wrote, adding that instead of resolving valuation disputes through negotiation of deal financials, “buyers and sellers now increasingly disagree on [the] more fundamental question” of how vulnerable a software business is to AI-driven disruption.

Outside the AI theme, Grab is “paying top dollar” to expand its buy-now-pay-later business six-fold through its takeover of Atome, Peter Milliken of Deutsche Bank research wrote. The two-tranche structure of the deal suggests Atome is expected to generate around $200 million in adjusted Ebitda by 2028, according to Milliken. “Grab likes the idea of using its combined data to improve both companies’ credit models, and scaling the foundational infrastructure across a much larger loan book,” Milliken said. Deutsche Bank retained a buy rating on Grab but trimmed its target price to $5.60 from $6.40. Grab shares closed at $2.80 on Friday.

Singtel’s shares could benefit from rerating catalysts tied to its GPU-as-a-Service business, RE:AI, according to Macquarie Capital research analyst Zhiwei Foo. Every S$300 million of contracts won by the unit is estimated to drive a S$0.12 increase in the stock’s fair value, assuming all else is equal, Foo wrote. Singtel may revisit its conservative Ebit outlook in November, with Foo estimating a 12 percent Ebit increase for the year ending March, above the company’s guidance. The results will likely be driven by Optus in Australia and the digital-infrastructure and AI businesses, Foo added. Macquarie retained an outperform rating on Singtel but lowered the target price to S$4.98 from S$5.29, citing lower fair values for regional associates and weaker foreign-exchange rates. Shares closed at S$4.33.