财经外刊精选 · 每日精读 Five Deep Reads from Bloomberg, the Financial Times, Reuters & The Information — for intensive English learning

Stocks fell toward the lowest since July as the Federal Reserve signalled the possibility of further policy tightening after raising interest rates for the first time since 2023 to fight inflation.

Equities erased gains, with the S&P 500 dropping 1 per cent. The Dow Jones Industrial Average lost 1.7 per cent. Short-dated Treasuries underperformed, with two-year yields hitting the highest since 2024. The dollar climbed. Money markets priced in an about 50 per cent chance of a Fed hike in October.

The Federal Open Market Committee voted unanimously to increase the benchmark rate to a range of 3.75 per cent to 4 per cent. The so-called dot plot, which the United States central bank uses to signal its outlook for the path of monetary policy, suggests one more hike this year.

In his remarks to reporters, Fed Chair Kevin Warsh restated his concerns over inflation, saying too many categories of products and services were showing annualized price gains above 3 per cent on a 6- and 12-month basis.

“The debate now shifts from whether rates will rise again to how many hikes lie ahead,” said Seema Shah at Principal Asset Management. “The unanimous vote shows that rising energy prices and stubborn inflation have brought even the doves on board, making a one-and-done move highly unlikely.”

With markets already pricing multiple increases, policymakers will probably need to deliver at least one more hike to safeguard credibility, she added.

The history is clear that once the Fed begins raising rates, they do it multiple times, but the pattern is less clear about whether they will raise rates at consecutive meetings or leave rates unchanged at some of the meetings in between this meeting and future ones where they do raise rates, according to Chris Zaccarelli at Northlight Asset Management.

“Chairman Warsh and the rest of the committee are building a strong reputation through their laser focus on defeating inflation,” said Jeff Roach at LPL Financial. “Given the current economic circumstances, the committee delivered what was needed. Going forward, the inflation debate will likely center on the magnitude of improvement as geopolitical conflicts wane.”

Wednesday’s rate decision sends a message to the markets that the Fed isn’t just talking about inflation, it’s actually doing something about it, according to Alex Guiliano at Resonate Wealth Partners.

“While one 25-basis point hike isn’t likely to bring inflation down overnight, it could help to stabilize the bond market, which has a direct impact on borrowing costs,” he said.

Foreign investors are now buying more US stocks than government bonds, in a rare move that comes amid fears that inflation and America’s fast-growing debt pile are undermining the “risk-free” status of Treasuries.

International flows into US stocks reached 2.8 per cent of US GDP on average in the year to June, overtaking Treasuries — at 2 per cent of GDP — for the first time this century outside brief episodes in the Covid-19 pandemic and the aftermath of the global financial crisis, according to analysis of US Treasury data by Deutsche Bank.

The change reflects the ever-growing lure of the US stock market, with the S&P 500 on track for a fourth straight year of double-digit gains. Powering the rally has been vast AI investment, which is driving profit margins to new highs, according to FactSet data stretching back to 2009.

At the same time, Treasuries’ traditional role as the global risk-free asset is coming under increasing pressure as investors grow warier of lending money to heavily indebted governments and question the independence of the Federal Reserve, in a move that could transform the way the US dollar trades.

On Monday the 10-year Treasury yield breached 5 per cent for the first time since 2023.

This is a “huge shift in US asset markets”, said George Saravelos, global head of FX research at Deutsche Bank, reflecting the fact “the American private balance sheet is booming . . . but the public sector balance sheet keeps worsening”.

Saravelos added that the value of the dollar might now be more closely tied to flows into US stocks than to flows into Treasuries, suggesting the greenback could strengthen as risk appetite grows.

This would mark a major break from its previous pattern of trading, when it has tended to rally during so-called risk-off periods, as foreign investors dump riskier trades and flock to US debt as a haven. US assets were “no longer the ‘safe’ but the risky asset of choice”, Saravelos wrote.

“I have never heard so much concern and debate around what ‘risk-free’ actually means,” said Matt Rowe, senior portfolio manager at Man Group.

Investors with this growing anxiety around Treasuries were “quickly faced” with similar problems, such as rising issuance in other bond markets, he added. “In an odd twist, equities seem to be competing for those funds [instead].”

Earlier this month the manager of Norway’s $2.3tn oil fund proposed slashing its holdings of US Treasuries by about $80bn and instead buying debt such as mortgage-backed securities backed by government agencies.

Marija Veitmane, head of equity research at State Street, said the bank is seeing the same pattern emerging among institutional clients, with greater allocations to stocks, seemingly at the expense of government bonds.

“If you are worried about the US government’s fiscal prudence, then comparing equity fundamentals with governments . . . companies look very, very solid,” Veitmane said. “Profit growth is strong.”

The rise in yields comes despite US Treasury secretary Scott Bessent’s plan to buy back long-term government debt, with the announcement last week of a $6bn purchase disappointing investors.

In contrast, the blue-chip S&P 500 is up about 12 per cent this year, as booming corporate profits have allowed traders to shrug off warnings about the fallout from the near closure of the Strait of Hormuz and surging energy prices.

The index chalked up 52 per cent year-on-year earnings growth in the second quarter of 2026, according to FactSet, or 34 per cent excluding Amazon and Alphabet, which reported significant one-off boosts from their stakes in other AI companies.

However, some investors and analysts remain wary that the stock market, which has made little progress this summer, could start to suffer if yields rise further, raising corporate borrowing costs and increasing the lure of fixed income investments.

American assets came under pressure early last year after President Donald Trump’s sweeping “liberation day” tariffs in April triggered fears that the era of US dominance in financial markets could be waning, hitting stocks, bonds and the dollar. This brief “sell America” sentiment quickly gave way to investor fear of missing out on further gains in AI stocks, propelling Wall Street to new highs. But the US Treasury market has remained under pressure, as it has been swept up in broader concerns about debt levels and inflation in a number of developed economies.

The sell-off in bonds was “top of mind in the equity market”, said Michiel Plakman, global head of equity at fund firm Robeco. “If there’s trouble in the bond market . . . usually that’s what troubles stocks.”

SEOUL/WASHINGTON (Sept 16): South Korea’s SK Hynix is in talks with Intel about a deal that would see it manufacture memory chips on US soil for the first time, three people familiar with the discussions said.

Under one potential scenario, SK Hynix would lease part of Intel’s long-planned chipmaking facility in Ohio, they said. Under another scenario, it could form a venture with Intel and major cloud firms that are keen to lock in memory chip supplies, two of the people said.

A deal would relieve pressure on Intel, which has been struggling, and mark a big win for the Trump administration, which has been ramping up pressure on chipmakers to build in the United States as relentless investment in AI and data centres fuels an acute shortage of memory chips.

But potential opposition from Seoul could be a major hurdle, the sources said. One of the sources described the talks as exploratory and stressed that no decisions have been made, adding that SK Hynix might also consider other ways to structure the deal.

Details about how a SK Hynix-Intel deal might be structured are being reported for the first time. Reuters was, however, not able to learn what types of chips SK Hynix may manufacture in Ohio if an agreement is struck. Its product lineup includes DRAM (dynamic random-access memory) chips used in servers, PCs and smartphones and NAND flash memory chips that provide long-term storage memory. It is also the leading supplier of high-bandwidth memory (HBM) used in AI processing units.

Any agreement to produce advanced memory such as HBM or even DRAM could run into opposition from the South Korean government as those technologies are considered sensitive, said the three sources who declined to be identified as the information is not public.

SK Hynix said in a statement to Reuters that it is “reviewing various measures, including establishing additional production bases, to strengthen the competitiveness of its memory business”, but “no matters have been determined at this stage”.

Intel declined to comment on what it called speculation, but added that it was continuing with investments in Ohio to ready the site.

Asked about a potential SK Hynix plan to produce chips in the US, South Korea’s trade ministry said any decision would be at the company’s discretion but if it involved a “national core technology”, it would be subject to a review under the Industrial Technology Protection Act. The US Commerce Department did not respond to a request for comment.

Manufacturing semiconductors in the US costs much more than in South Korea, according to two of the sources. Building in the United States entails significantly higher labour and construction costs, and much of the semiconductor industry’s supply chain is located within Asia, which also increases US costs, one of them said.

Even so, SK Hynix, which recently completed a secondary listing on the Nasdaq in July and currently only has a chip packaging facility under construction in Indiana, has good reasons to pursue fabricating chips in the United States.

Chey Tae-won, chairman of the SK Group conglomerate that SK Hynix is part of, said in July that the company faces enormous pressure and lobbying from customers and countries to supply more chips. “I think we need to build a factory in the United States. If possible, I believe we should build it,” he told reporters.

A deal with SK Hynix would also ease some of the financial strain on Intel, which is part-owned by the US government. Intel announced in 2022 that it would invest up to US$100 billion to build potentially the world’s largest chipmaking complex in Ohio, with production scheduled to begin in 2025. But completion of the site’s two plants has since been delayed to 2030 and 2031.

Pleasing both Washington and Seoul could be tough, however, particularly as SK Hynix and rival Samsung Electronics have been urged by Seoul to accelerate plans to build a new cluster of chipmaking facilities in South Korea’s southwest.

US Commerce Secretary Howard Lutnick has threatened to impose up to 100% tariffs on South Korean and Taiwanese companies unless they commit to increased production on American soil.

Both countries have also been wrangling over the execution of a huge investment commitment that South Korea made to the US last year in return for lower US tariffs. Some US$150 billion of a proposed US$350 billion has been earmarked for shipbuilding but the remaining US$200 billion is undecided.

Two of the sources said this has left SK Hynix in a difficult position, with the South Korean government seeking to use any US investment by SK Hynix as leverage in the talks, and Washington heaping pressure on the firm to quickly commit to a US expansion.

Crux AI, the new cloud computing venture tied to Blackstone and Alphabet, is lining up a $22 billion loan from a group of 10 banks to finance purchases of Google’s custom artificial intelligence chips, according to Bloomberg. The transaction represents one of the largest debt financings assembled specifically to fund the processors needed for AI computing infrastructure.

The lending group includes Goldman Sachs, Sumitomo Mitsui Banking Corporation, Barclays, BNP Paribas and Bank of Nova Scotia, with the banks working to bring additional lenders into the transaction through syndication.

The financing will be used to purchase Google’s Tensor Processing Units, or TPUs, specialized processors the technology company developed to accelerate machine learning and other AI workloads.

The structure is notable because the $22 billion debt package will be secured by both the value of the chips and Crux AI’s customer contracts, effectively turning expensive AI computing hardware and its associated contracted revenue into collateral for one of the largest infrastructure financings of the AI boom.

Several banks are also providing Crux AI with a separate $1 billion revolving credit facility, adding liquidity beyond the primary chip financing.

The debt could eventually be refinanced with longer-term capital from institutional investors through the investment-grade bond market, giving banks a potential path to syndicate or replace portions of the initial financing as Crux AI’s infrastructure becomes operational.

Crux AI emerged from a cloud infrastructure initiative announced by Google and Blackstone in May.

Blackstone committed an initial $5 billion of equity to the venture, which is intended to address surging enterprise demand for AI computing capacity.

The initial development plan calls for approximately 500 megawatts of data center capacity to come online in 2027, with additional expansion expected afterward.

The combination of $5 billion in initial Blackstone equity and the new $22 billion chip loan illustrates how quickly financing requirements are increasing as AI infrastructure moves from individual data centers toward much larger computing platforms.

Unlike traditional cloud infrastructure dominated by general-purpose CPUs and GPUs, Crux AI will make significant use of Google’s proprietary TPUs.

Google designed the chips specifically for AI and machine-learning workloads, giving Alphabet another avenue for expanding the ecosystem around its custom silicon beyond its own internal operations and Google Cloud.

For Blackstone, Crux AI extends the firm’s already significant exposure to data centers and digital infrastructure.

The world’s largest alternative asset manager has increasingly focused on the enormous capital requirements associated with AI, including the real estate, electricity, computing hardware and supporting infrastructure necessary to operate advanced models.

The scale of the new loan also demonstrates how those investments are creating opportunities for banks.

AI infrastructure has traditionally been financed through combinations of corporate balance sheets, project finance, private credit and equity investments. But the extraordinary value of modern AI processors is creating new forms of asset-backed lending in which chips themselves can serve as collateral.

In Crux AI’s case, lenders have another potential source of protection through the venture’s customer contracts.

Long-term commitments from customers buying computing capacity can provide predictable future cash flows, potentially allowing lenders to underwrite the financing more like infrastructure or equipment finance rather than unsecured lending to a technology startup.

That could become increasingly important as the cost of AI hardware grows into the tens of billions of dollars for individual projects.

The Crux AI deal therefore represents a broader evolution in how the AI buildout is being financed.

Banks, asset managers and private-credit firms are increasingly developing structures that allow technology companies and infrastructure operators to access enormous amounts of capital without carrying all of the investment directly on their corporate balance sheets.

Chip financing is emerging as one part of that market alongside loans secured by data centers, power infrastructure and long-term cloud contracts.

The demand behind those transactions continues to grow as technology companies race to secure computing capacity.

AI developers require increasingly large clusters of specialized processors to train frontier models and operate inference workloads for millions of customers. At the same time, enterprises are beginning to incorporate AI into more everyday business applications, creating additional demand for cloud computing.

Those trends require investment not only in chips but also in data centers, networking equipment, cooling systems and electricity generation.

Crux AI’s financing shows how quickly that spending can scale. A single cloud venture is preparing to borrow $22 billion primarily to purchase specialized processors before accounting for the billions of dollars needed for the broader physical infrastructure surrounding them.

The transaction is also another significant financing mandate for Goldman Sachs and the other global banks involved at a time when AI infrastructure has become an increasingly important source of investment banking and lending activity.

For Alphabet, the venture can potentially expand demand for Google TPUs while increasing the availability of computing infrastructure based on its proprietary architecture.

For Blackstone, the project offers another way to participate in the expansion of AI infrastructure through a business that combines digital assets, computing hardware and contracted cloud revenue.

Crux AI ultimately sits at the intersection of several of the largest investment themes driving technology markets: custom AI silicon, cloud computing, hyperscale data centers and infrastructure finance.

The $22 billion chip loan, additional $1 billion revolving facility and Blackstone’s initial $5 billion equity commitment provide an indication of the amount of capital that may be required as the venture works toward bringing its first 500 megawatts of capacity online in 2027.

Sept 16: Apple is considering returning to the enterprise server market, combining its own chips with Nvidia’s networking technology, The Information reported on Wednesday, citing people familiar with the matter.

Cupertino, California-based Apple is developing an AI server that could be offered with either two or four planned M8 Ultra chips, its highest-performance processors, according to the report.

Such a move would signal Apple’s push into AI infrastructure — targeting the fast-growing inference market as enterprise workloads shift from training AI models to running them — while expanding its strategy of using in-house silicon beyond consumer devices.

The server is not expected to reach the market until 2029 and could still be canceled or launched without Nvidia technology, The Information reported.

It added that the initiative was backed by the then-hardware chief, now CEO, John Ternus when it began about a year ago.

Apple is considering Nvidia’s NVLink Fusion technology to connect the M8 chips, with some staffers believing Nvidia offers the best available connectivity solution for the project, according to the report.

NVLink Fusion gives chipmakers ready-made connectors and specialized memory so custom AI chips can plug into Nvidia’s larger data-center systems.

Apple and Nvidia did not immediately respond to Reuters requests for comment. Reuters could not independently verify the report.

Apple last month refreshed its Mac lineup with M6, M5 Pro, M5 Max and M5 Ultra chips, saying the 2-nanometer M6 delivers up to four times faster AI performance, while the higher-end chips target developers and professional AI workloads.

The project would represent Apple’s first dedicated server hardware offering since it discontinued its Xserve line in 2011 after shifting its focus toward standard Mac hardware and consumer devices.

The report also suggests a thaw in relations between Apple and Nvidia. The companies have had a strained relationship for nearly two decades, following disputes that culminated in Apple severing ties after widespread MacBook failures linked to Nvidia graphics chips.