Meta’s Muse AI May Boost Stablecoin Use, BlackRock Says

1. U.S., China Put AI Safety on Agenda, but Deal to Slow Development Looks Remote

On Wall Street on September 23, the prevailing view was that President Donald Trump and Chinese President Xi Jinping were unlikely to reach a substantive summit agreement to slow the pace of the AI race.

Treasury Secretary Scott Bessent said earlier that the agenda also included a communication channel for handling AI incidents that could escalate into national-security issues. Alok Wadhwani, head of the AI Center at the Center for Strategic and International Studies, said both leaders are worried about AI models developing stronger cyber capabilities.

Particular attention is focused on agentic AI systems that can make plans on their own and breach websites, raising the risk that they could slip beyond human control.

Even as safety talks advance, the technology rivalry is sharpening. The U.S. is restricting China’s access to Nvidia’s most advanced AI chips. Washington has also raised allegations that Chinese AI companies are using distillation to train on outputs from more advanced U.S. models. China has denied the claim.

Beijing has long used AI talks as a venue to criticize U.S. export controls on advanced semiconductors. Melanie Hart, senior director of the Atlantic Council’s Global China Hub, said China is pressing the U.S. to accept its domestic AI models and ease restrictions on chip access.

Whether Trump would accept those demands is the key question. Given Washington’s broader stance on technology controls targeting China, that looks unlikely.

Inside China, there is already a growing perception that U.S. sanctions have constrained development. Selina Xu, who researches China and AI policy at former Google Chief Executive Officer Eric Schmidt’s office, said Chinese officials see U.S. technology restrictions and chip export controls as having already forced a slowdown. In her conversations with policy officials, she said, the message has been: “We are already slowing down.”

The summit may still leave room for limited cooperation, including incident response and communication channels. A deal to slow competition in semiconductors and AI models, however, appears out of reach.

2. SoftBank’s Junk-Bond Sale Comes With Record Borrowing Costs

SoftBank Group is launching a bond sale worth more than $11 billion to fund AI investments, one of the largest high-yield offerings ever by a lower-rated borrower.

The deal consists of $10 billion of dollar-denominated bonds and 1 billion euros of euro-denominated notes. Indicative yields range from 8.75% to 8.87% for 3.5-year notes, 9.37% to 9.5% for 5.5-year notes, and 9.7% to 9.87% for 7.5-year notes.

Those are the highest levels for SoftBank dollar bonds of comparable maturities.

SoftBank has already committed to invest about $65 billion in OpenAI, the developer of ChatGPT. Proceeds from the sale are also set to be used for OpenAI investments and AI-related mergers and acquisitions.

Chairman Masayoshi Son has brushed aside concerns about AI infrastructure spending, saying AI-related industries could account for 20% of global output by 2040, or about $46 trillion.

The issue is the price investors are demanding. SoftBank carries a BB+ rating from S&P, the highest rung within speculative grade. Dollar bonds sold by BB-rated issuers typically yield about 6.55%. SoftBank is offering 8.75% to 9.87%, much closer to the roughly 8.5% average for B- rated issuers.

That suggests rating agencies may score SoftBank at BB+, but lenders are demanding a far steeper risk premium as they factor in heavy AI spending and rising leverage.

There is also concern that if the technology does not generate profits as quickly as hoped, enthusiasm for AI investment could become a burden on the broader corporate bond market.

Global AI-related bond issuance has already surpassed $575 billion this year. As debt-funded AI investment spreads, SoftBank’s elevated borrowing costs are being read as a sign of growing caution in the bond market over the AI boom.

3. IPO Delays Stir U.S. Market Anxiety as Investors Watch Anthropic’s Timing

A string of IPO delays in the U.S. is fueling fresh concern about the new-issue market. Investors are also watching closely for signs of when Anthropic, one of the biggest AI companies, will make its move.

Nuclear-services company Holtec Nuclear and insurer Bamboo Insurance recently postponed their IPOs, citing market conditions. Bamboo Insurance counts private-equity firm CVC Capital Partners as an investor.

September, after Labor Day, is usually when the U.S. IPO market becomes more active. This year, however, the market has been quieter than expected, with only three companies listing so far this month.

Anthropic’s next steps are drawing particular scrutiny. The company had been expected to make its filing public as early as late August, but it has yet to do so.

It recently unveiled a new AI model with lower costs and is also responding to debate over the risks tied to rapid advances in AI technology. Market participants have speculated that Anthropic could pursue fundraising on a scale matching or exceeding SpaceX’s record IPO.

Stock indexes have kept climbing. The S&P 500 has gained 1.1% in September, and the Nasdaq Composite has risen more than 3% to a record high.

That makes the IPO delays unusual. Jay Ritter, director of the University of Florida’s IPO Initiative, said it was surprising to see IPOs postponed even with the Nasdaq at an all-time high. “It’s not as if the market has dropped sharply,” he said.

Bloomberg reported that more than 20 companies have publicly filed for U.S. IPOs since July, but only nine have actually listed. Smart-ring maker Oura plans to raise as much as $2.2 billion in an IPO next week. Data-center infrastructure company Acceleration Holdings is seeking $720 million on the same day.

Funds raised in the U.S. IPO market this year have reached $161.4 billion, the highest level since 2021, thanks to a handful of large deals. SpaceX’s $86.2 billion listing and SK Hynix’s $26.5 billion American depositary receipt sale boosted the total.

But success in a few large transactions does not mean every company can list on favorable terms. Five of the 10 biggest IPOs this year are now trading below their offer prices.

The weighted average return for IPO stocks this year is 13%, trailing the S&P 500’s gain of about 14%. Federal Reserve rate increases are another headwind.

Higher rates reduce the present value of future earnings and raise interest costs for companies that need borrowing, including businesses building data centers. With geopolitical risk, oil prices and monetary policy all adding uncertainty, companies with ample cash are not rushing to go public during a volatile period.

4. ‘A Diesel Export Ban Would Push Prices Even Higher’

U.S. diesel prices have hit a record $6.53 a gallon, up 75% from a year earlier, according to American Automobile Association data.

President Donald Trump said on the sidelines of the United Nations General Assembly in New York that he could support a diesel export ban. The logic is that keeping more diesel at home could lower overall fuel prices. He said a decision would come “quickly, one way or the other.”

Experts warn the result could be the opposite. Gbenga Ajilore, chief economist at the Center on Budget and Policy Priorities, said an export ban would not increase domestic supply. Instead, he said, it could reduce supply and drive prices even higher.

Ajilore said the main drivers of higher diesel prices are the Iran war and instability around the Strait of Hormuz. The fundamental solution, in his view, is to end the war and reopen the strait, a critical oil-shipping route between Iran and Oman.

Bespoke Investment Group also said export restrictions could hurt global diesel supply and refinery profitability. Gulf Coast refineries could cut runs or idle plants if they run short of storage space, the firm said.

The U.S. has experience with export limits, having restricted crude exports from 1975 to 2015. At the time, multiple studies, including from the Energy Information Administration, concluded that allowing exports would not raise U.S. gasoline prices and could even lower them.

Gasoline prices did not surge after the rules were lifted in 2015.

Many analysts say the benefits of the old export ban flowed more to refiners than to consumers. Crude trapped in the domestic market traded at more than $10 a barrel below international prices, while gasoline and diesel prices continued to track global supply and demand.

That let refiners buy cheap crude and sell refined products at international prices, widening margins. Consumer prices, however, did not fall enough to deliver the same benefit.

The same logic applies to diesel. Diesel and gasoline are produced together in the refining process, making it difficult to increase diesel output alone.

The U.S. is the world’s largest diesel exporter. If those export volumes disappear from the global market, prices abroad could rise and eventually feed back into the U.S. market.

Some analysts also say New England’s problem is rooted less in exports than in transportation constraints. The region has few refineries and imports a significant share of its diesel from Canada.

It could be supplied from the U.S. Gulf Coast, but pipeline capacity is insufficient.

A central obstacle is the Jones Act, enacted in 1920. The law restricts cargo transport between U.S. ports to ships built in the U.S., owned by Americans and flying the U.S. flag.

U.S.-built tankers cost about four times as much as foreign-built vessels, and higher labor costs have left the market short of ships. Daily charter rates for Jones Act tankers are about $90,000, roughly 10 times the $9,000 cost for foreign-flagged ships.

As a result, transporting the same fuel from the Gulf Coast to the U.S. Northeast costs as much as $6 a barrel. Shipping it to Canada on a foreign-flagged vessel costs about $2 a barrel.

That is why the argument is gaining traction that New England’s diesel problem would be better addressed by tackling transportation costs and infrastructure bottlenecks rather than banning exports.

Ipek Ozkardeskaya, senior analyst at Swissquote, said restricting U.S. exports could trigger a sharp jump in global diesel prices. If U.S. refiners lose part of the export market, they may cut output, which could eventually tighten domestic supplies of other fuels as well.

5. BlackRock Says Stablecoins Could Be the AI Era’s Payment Tool

BlackRock, the world’s largest asset manager, said stablecoins could become a core payment tool for machine-to-machine transactions in the AI era. As AI assistants begin shopping, buying data and securing computing resources on behalf of humans, they will need money and payment networks that machines can use directly.

In its September report, “The Machine-Native Economy,” BlackRock said AI and digital assets share a common trait as “machine-used intelligence” and “machine-used money.” It said the two industries are likely to converge in earnest as agentic AI spreads. Those systems can plan and execute multistep tasks without human intervention.

AI understands information by breaking text into tokens and converting it into numbers. Digital assets work similarly by tokenizing money and ownership of assets so machines can verify and transfer them.

If a $100 stake in a money-market fund is tokenized, for example, a machine can read and process transaction details, including the owner and the number of units.

In that setup, an AI assistant can check balances, review transaction terms and complete settlement automatically. That opens the door to large volumes of small payments with minimal human involvement.

BlackRock said existing card payments and bank transfers are not well suited to the high-volume payments AI agents may need to make. Current payment networks are built around account opening, identity verification, per-transaction fees and long settlement times.

If AI assistants are to conduct thousands of small transactions a day, they will need payment rails that operate around the clock, charge low fees and settle instantly.

APIs used by AI assistants to obtain data are part of the backdrop for that shift. When humans search on websites, they often get free services supported by advertising or booking fees. AI systems, by contrast, receive data through machine-only gateways known as APIs.

That process can involve per-call charges for airline fares, stock prices, weather data, paid articles and calls to other AI services.

If an AI assistant is asked to find the cheapest trip to Tokyo, for instance, it may query airfare and hotel rates from hundreds of providers through APIs. Each request may cost little, but once those searches and payments are repeated hundreds of times, it becomes impractical for a human to authorize each one.

That is where stablecoins could emerge as a payment tool AI can use directly.

Stablecoins are digital assets designed to keep one coin worth around $1 by backing it with safe assets such as the dollar or U.S. Treasuries. Unlike Bitcoin or Ether, whose prices are far more volatile, stablecoins are seen as better suited to pricing goods and making recurring payments.

BlackRock said stablecoins could become transaction-centered digital assets because of price stability, 24-hour automated payments, fast settlement and programmability. According to the report, stablecoin market capitalization had surpassed $300 billion as of September 2026.

Transaction volume topped $11 trillion in 2025, and the compound annual growth rate since 2020 was about 80%.

Compute capacity used to run AI could also become a new asset for trading. BlackRock estimates cumulative AI capital spending from 2025 through 2030 will exceed $5 trillion.

Over time, it said, a market could emerge in which compute capacity is tokenized, traded, posted as collateral or used to hedge price risk.

The actual buyers of compute are likely to be companies operating AI services. Those firms could lock in prices in advance to protect against future increases in computing costs.

Institutional investors could invest in swings in compute prices or hedge that risk. AI agents, meanwhile, could compare computing resources in real time, choose the cheapest option for a given task and pay automatically.

BlackRock said both agent payments and the market for trading compute remain at an early stage. Still, the report’s central point is that as AI adoption expands, demand could rise structurally for AI-native payment tools, payment rails and markets for buying and selling computing power.

New York-based correspondent Park Shin-young nyusos@hankyung.com