U.S. government bonds are once again offering investors yields around 5%, but the long-term Treasury market may still struggle to compete with the potential returns from artificial intelligence infrastructure, according to a senior Goldman Sachs executive.

Anshul Sehgal, Goldman’s global co-head of fixed income, currencies and commodities, said investors looking for an asymmetric opportunity should favor the infrastructure supporting the AI boom, including data centers and so-called neocloud providers that rent computing capacity to companies developing and running AI models.

“Personally, I think the asymmetric expression is being long compute,” Sehgal said on Goldman’s “The Markets” podcast published Friday.

The argument comes at an important point for markets. The Federal Reserve raised interest rates last week and signaled that another increase could come later this year, pushing long-term Treasury yields higher. Yields on longer-dated government bonds have climbed to their highest levels in more than two decades, bringing fixed income back into focus after years in which investors had little compensation for holding government debt.

A 5% yield on a long-term Treasury provides a substantially more attractive income stream than bonds offered during the ultra-low-rate era. But Sehgal argues that the opportunity cost of choosing that relatively predictable return could be high if spending on AI infrastructure continues to accelerate.

His preferred trade is not necessarily a broad bet on technology stocks. Instead, he is focusing on the physical infrastructure required to support the expansion of AI computing, including data centers and neocloud companies. That is considered relevant as the AI investment cycle moves beyond model development and into a capital-intensive buildout of computing capacity, electricity, networking equipment, and data centers.

Sehgal also argued that higher interest rates have not necessarily undermined this investment cycle.

One reason is that higher rates increase income for savers and investors holding interest-bearing assets. Some of that capital can ultimately flow into the financial system and help fund the companies building AI infrastructure. Analysts note that this creates an unusual dynamic for monetary policy. Higher rates increase the cost of financing data centers and other capital-intensive projects, but they also increase the income available to investors providing capital to those projects.

For Sehgal, the potential upside from AI infrastructure outweighs the attraction of locking in today’s elevated Treasury yields.

While he expects bond yields could decline modestly from current levels, he said AI investments “can go up multiplicatively.”

That does not mean he expects the entire stock market to benefit equally from the AI boom.

“I think long compute just makes a lot of sense to me here,” Sehgal said. “But does that mean that the broader equity complex should do very well? It’s less clear.”

The statement appears to highlight one of the central changes taking place in the AI trade. Investors have to separate companies that benefit directly from rising demand for computing capacity from businesses whose exposure to AI is more indirect.

The infrastructure side of the market has already attracted enormous amounts of capital. Data-center developers, chipmakers, power providers and specialist cloud companies are all seeking to expand capacity as AI companies require large amounts of computing power for both training and inference.

The economics are also different from those of many software businesses. Building AI infrastructure requires large upfront investments, long-term power arrangements, and access to scarce equipment. The companies able to secure customers and maintain high utilization can potentially generate substantial cash flows from that installed capacity.

That is part of the reason investors have been treating “compute” as a distinct asset theme rather than simply another technology-stock trade.

Also, the higher-rate environment creates a test for the AI infrastructure boom. Data centers require billions of dollars in capital, meaning financing costs matter. If rates remain elevated for longer, projects with weak customer commitments or uncertain economics could become harder to justify.

Sehgal’s argument therefore rests partly on the assumption that demand for AI computing will continue to grow fast enough to support the enormous capital being deployed.

His comments also challenge the idea that rising Treasury yields automatically make the AI trade less attractive. If the AI infrastructure buildout produces returns that exceed the cost of capital, higher rates may simply change which projects get funded rather than halt the investment cycle.

Goldman Executive Dismisses Worst-Case U.S. Debt Concerns

Sehgal was also relatively sanguine about concerns surrounding the sustainability of U.S. government debt. He argued that investors may be overstating the implications of Washington’s large fiscal deficits because an increasing portion of government spending is now represented by interest payments to holders of Treasury securities rather than new government spending entering the broader economy.

“That does not accrue to labor. That accrues to capital. That accrues to the top decile of wage earners,” Sehgal said.

His argument is that the composition of government spending matters as much as the headline deficit. Interest payments transfer income to holders of government debt, rather than necessarily generating the same immediate economic stimulus as spending on infrastructure, wages or government programs.

Sehgal also pointed to the relationship between nominal economic growth and the deficit. Annual U.S. budget deficits have remained around 6% to 7% of GDP, he said, while nominal GDP has grown by roughly 6% annually over the past four years.

That combination has limited the deterioration in the debt-to-GDP ratio relative to what the size of the deficits alone might suggest.

“I don’t think the sustainability issue is really that credible in the long run,” Sehgal said.

The argument does not eliminate the risks associated with higher U.S. borrowing costs. If Treasury yields remain elevated, the government’s interest bill can continue to increase as existing debt matures and is refinanced at higher rates. That can put further pressure on future budgets.

But Sehgal’s broader point is that the market should distinguish between the level of U.S. debt and the economic consequences of servicing that debt.

That argument leaves a striking contrast for investors. Treasury yields near 5% now provide a meaningful income return and a potential source of capital preservation, while AI infrastructure offers considerably greater potential upside but comes with substantially greater execution, valuation, and demand risk.

Sehgal’s preference for “long compute” ultimately marks a bet that the AI infrastructure cycle is still in its expansion phase and that the demand for computing power will grow faster than the market’s ability to supply it.

If that happens, the companies controlling scarce computing capacity, data-center space, and related infrastructure could capture a significant share of the economic value created by AI, even if the broader equity market becomes more difficult for investors.

UBS Chief Executive Sergio Ermotti has warned Swiss lawmakers that excessively stringent capital requirements could weaken the bank’s competitiveness and ultimately impose costs on customers and employees, days before parliament is due to decide how far to tighten regulation following the collapse of Credit Suisse.

Switzerland’s upper house of parliament is scheduled to vote on Wednesday on new capital rules for UBS, which acquired its troubled rival in an emergency takeover arranged by Swiss authorities in 2023.

The debate has become a test of how far Switzerland should go to prevent another banking crisis without imposing requirements that could make its largest bank less competitive internationally.

The Swiss government has proposed rules that would require UBS to hold about $20 billion in additional capital. The measures are intended to strengthen the bank’s ability to absorb losses and reduce the potential burden on taxpayers if it were to encounter another severe crisis.

Ermotti, however, said UBS could accept some tightening but rejected the government’s proposal to require the bank to back its foreign subsidiaries with 100% Common Equity Tier 1 capital.

“It’s a mistake to believe the additional costs will only be borne by shareholders,” Ermotti told Neue Zuercher Zeitung. “Customers and employees will be affected, too.”

The dispute centers on how UBS should capitalize its foreign operations and what form that capital should take.

A parliamentary committee in the upper house agreed last month to a compromise under which UBS could satisfy half of the requirement using Additional Tier 1 capital, a form of bank capital generally considered less expensive than Common Equity Tier 1.

UBS estimates that it would need to raise about $13 billion in AT1 capital under that proposal.

Ermotti said the requirement would be painful but manageable for the bank.

A tougher alternative has since gained attention among lawmakers. During an upper house debate on Thursday, there were indications that some parliamentarians were considering requiring UBS to fund 90% of its foreign units with CET1 capital.

Ermotti said that proposal would go too far.

“We can live with a black eye, but two black eyes and a broken nose is too much,” he said. “Yet that’s exactly what the demand for capital backing of 90% or 100% comes down to.”

The disagreement goes to the heart of the lessons Switzerland has drawn from the Credit Suisse collapse.

Credit Suisse failed after years of financial and management problems culminated in a loss of market confidence in March 2023. The Swiss government and regulators arranged its takeover by UBS to prevent a disorderly collapse that could have threatened financial stability.

The emergency rescue also exposed the limits of Switzerland’s existing framework for dealing with a systemically important bank.

The government now wants UBS to have a larger capital buffer, reducing the likelihood that public funds would be needed in a future crisis. UBS, meanwhile, argues that forcing it to hold significantly more expensive capital could reduce returns and make it less competitive against international rivals that operate under different regulatory regimes.

The disagreement has become increasingly public.

UBS Chairman Colm Kelleher said last week that the bank would have to consider its future in Switzerland carefully if the new rules became so restrictive that it could no longer compete effectively.

Ermotti’s latest comments bolster that warning while drawing a line between regulation UBS can absorb and requirements the bank considers excessive.

The outcome could have consequences beyond UBS’s shareholders. Higher capital requirements can increase a bank’s resilience because shareholders provide a larger cushion against losses. But capital is also more expensive than debt and other forms of funding, meaning banks can seek to recover higher costs through pricing, reduce certain activities or accept lower returns.

That is the trade-off Swiss lawmakers are now being asked to weigh.

Ermotti also argued that Swiss regulators and policymakers should examine their own role in the Credit Suisse failure. He said the Swiss Financial Market Supervisory Authority, known as FINMA, and the Swiss National Bank bore some responsibility for the bank’s demise.

The parliamentary decision will therefore determine more than the immediate amount of capital UBS must raise. It will help establish the regulatory framework governing Switzerland’s largest bank after the country’s most serious banking crisis in decades.

However, the issue for UBS is whether the additional protection demanded by policymakers can be achieved without materially damaging the economics of its global business. But the Swiss authorities face the challenge of ensuring that the next banking crisis, if one occurs, does not again leave the state facing the choice between rescuing a major institution and accepting potentially severe consequences for the financial system.

China is putting the brakes on a rush of humanoid-robot companies seeking stock-market listings, as regulators scrutinize soaring valuations and question whether revenue tied to state-backed projects represents sustainable commercial demand.

People familiar with the matter told Reuters that Chinese regulators have used informal “window guidance” to discourage some humanoid-robot companies from pursuing IPOs, raising the bar for listings in one of the country’s most aggressively promoted technology sectors. One person said humanoid IPOs had effectively been frozen for now, while another described the move as a sector-specific slowdown rather than a formal ban.

The regulatory caution was triggered in part by the volatile performance of Unitree Robotics, one of China’s best-known humanoid and quadruped robot makers. Its shares surged more than fivefold in their Shanghai debut last month before retreating 55% from their peak.

The episode has exposed a tension at the heart of China’s robotics industry. Beijing has made “embodied intelligence”, or AI systems capable of perceiving and acting in the physical world, a strategic emerging industry. That policy support has helped channel private capital and local-government funding into humanoid robotics, but it has also encouraged a rush of companies and investors into a sector where commercial deployment remains relatively limited.

The regulatory response therefore appears aimed less at slowing the underlying technology than at imposing a higher threshold for companies seeking access to public markets. The approach seems to be part of Beijing’s plan to develop a domestic robotics industry while avoiding the kind of speculative excess that can leave retail investors bearing the losses when expectations move faster than actual demand.

Leo Wang, a venture capitalist at Qianchuang Capital, described the investment wave as “campaign-style innovation”, a term used in China for periods when companies and capital rapidly concentrate around a policy-favored industry.

Wang said enthusiasm around embodied AI had exceeded that seen during China’s earlier internet and new-energy investment waves. Industrial-robot manufacturers have been shifting towards humanoid machines, while startups have attracted rapidly rising valuations. Some founders, he said, have attracted dozens of potential investors within weeks and resisted conventional due diligence. Some private-market projects have already suffered valuation cuts of 30% to 50%.

At the center of regulators’ concerns is the quality of revenue.

One person close to humanoid-robot investors said authorities were examining whether income generated through local-government-backed projects could be sustained. Robot data-collection centers and joint ventures, in which local governments can provide 80% to 90% of initial investment, have generated significant revenue for some companies.

Such arrangements can help companies build sales figures, support private-market valuations and satisfy financial thresholds for an IPO. But regulators are now asking a more fundamental question: how much of that demand comes from customers buying robots because they need them, rather than from government programmes designed to promote the industry?

That is expected to materially alter how some companies are valued. The person estimated that valuations at some robotics companies could fall 60% to 70% if revenue associated with data-collection centers were removed.

That scrutiny is also spilling into the public market. Mech-Mind Robotics CEO Shao Tianlan alleged in a WeChat post this month that some highly valued embodied-AI companies were generating revenue through data-collection centers, related-party transactions and other arrangements that may not be sustainable as they race towards IPOs. Shao declined to comment beyond the post.

Shares in Mech-Mind have fallen nearly 20% from their debut-day high on September 1, another indication that investors are becoming more selective even as the broader robotics narrative remains strong.

At least half a dozen Chinese humanoid-robot companies are preparing to go public, including Deep Robotics, X Square Robot and AGIBOT. None immediately responded to Reuters’ requests for comment on whether regulators had slowed their listing plans.

The scrutiny marks a significant change in the financing environment. For much of the year, the major question for investors was how quickly China’s humanoid-robot industry could scale. The focus is now shifting towards whether individual companies can prove that their robots are being deployed at meaningful volumes, generate repeat orders, and eventually produce sustainable profits.

Ruiying Zhao, a senior research analyst at S&P Global Market Intelligence, described the change in sentiment as a move from “blanket euphoria to selective rationality”, with investors paying greater attention to whether realized commercial value justifies high valuations.

That shift does not necessarily undermine China’s broader robotics ambitions. Instead, it could force the industry to move from demonstrations and policy-backed deployments towards measurable commercial use.

The situation is gaining wider interest as companies attempt to turn impressive demonstrations into industrial businesses. A robot that can run, dance or perform a complex maneuver can attract enormous attention, but the economics of the sector will ultimately depend on whether manufacturers can sell large numbers of machines that reliably perform useful tasks in factories, warehouses, logistics operations and other commercial environments.

A senior banker involved in Asian equity offerings said investors remained willing to finance robotics companies but were demanding more evidence of deployment, volumes and valuations.

“What’s the use case? Is it just people’s robots dancing around? Is it working in factories?” the banker said. “The volume hasn’t really caught up with the hype.”

That question goes to the heart of China’s next phase of the humanoid-robot boom. The country has already demonstrated that policy support, manufacturing capacity and abundant capital can accelerate the development of an emerging technology. The harder test is whether those advantages can produce an industry whose revenues are driven increasingly by independent customers rather than government-backed programmes.

Beijing is now facing the challenge to maintain momentum in a technology viewed as strategically important while preventing the capital market from getting too far ahead of the underlying business. Unitree’s rapid rise and subsequent fall have provided an early warning of what can happen when public-market expectations move faster than commercial adoption.

The result is likely to be a more demanding IPO environment for China’s humanoid-robot sector. Companies may now have to show not only that their technology works, but that customers are willing to pay for it repeatedly and at sufficient scale to justify the valuations attached to the industry’s next generation of public companies.