Finance

Three Prices of AI Capex: Chip Stocks Retreat 38%, Meta Bond Spreads Widen About 90 Basis Points, Storage Device Prices Rise 45.4%

Updated · 2026-11-09 11:04 · 6 sources cited

Three Prices of AI Capex: Chip Stocks Retreat 38%, Meta Bond Spreads Widen About 90 Basis Points, Storage Device Prices Rise 45.4%

Around Nov. 9, the pricing of AI capex is unfolding simultaneously across three layers—asset prices, financing costs and real-economy prices: the US semiconductor index has fallen 38% from its June high, Meta is paying a bond spread about 90 basis points wider than in April to finance AI, and China's data storage device prices rose 45.4% year on year in October. On the same day, A-share AI hardware and physical AI concepts strengthened in morning trading, while Hong Kong tech stocks rebounded in tandem—the same industrial chain is sending inconsistent signals across different markets.

Chip Stocks: Decline Exceeds Historical Average, but Duration Is Less Than Half

The US SOX semiconductor index fell from a high of 14,634 on June 22, 2026, to 9,011 on Nov. 5, a cumulative drop of 38% over 136 days. Bank of America's Vivek Arya team counted four major semiconductor downturns in 2011, 2015, 2018 and 2022, with an average duration of 253 days and an average peak-to-trough decline of 32%. That means the magnitude of this decline has exceeded the historical average, while its duration is less than half the historical average; on that basis, the team judges that this 'indiscriminate capitulation-style selloff' may be completed in early 2027.[2]

Valuation also leaves limited room. SOX currently trades at 13.7x forward P/E, versus 17.7x for the S&P 500, a 22% discount to the broad market; at the hottest point of the AI rally, chip stocks traded at a 15% premium to the market, a swing of nearly 37 percentage points from premium to discount. Forward P/E at past cycle lows was about 12x; on that basis, current valuation has only about 12% downside left to the historical bottom. Analysts also cautioned that real EPS cuts—not merely cheap valuations—are what would truly reassure investors, and that will take time.[2]

Capex is at the core of this round of concern. BofA data show US cloud capex is currently running at about $300 billion a quarter, or about $1.2 trillion annualized, while the market's consensus for full-year 2027 is $1.555 trillion; if capex stays at the current rate, actual 2027 spending will be about 23% below expectations. By segment, traditional hyperscalers are expected to spend $1.221 trillion in 2027, versus a current run rate of $982.5 billion, a gap of about 20%, while the gap for emerging cloud providers is larger, at about 35%. BofA estimates that if 2027 capex is cut by 20% versus expectations, AI semiconductor revenue would face more than 20% downside pressure and EPS would face 35-40% downside risk.[2]

History offers another reference: SOX declines typically lead actual semiconductor sales declines by 6-9 months, then average a roughly 42% rebound after earnings expectations are fully digested. On investment strategy, BofA lays out three scenarios—in the base case, it prefers leaders in computing chips, networking chips, semiconductor equipment, memory and analog chips; in a bear case, defensive EDA software, high-free-cash-flow analog chip companies and firms with smaller AI exposure are relatively favored; in a bull case, high-beta names such as memory, optics, quantum computing and CPUs would lead.[2]

Financing Side: Meta's $46 Billion Bond Sale and Wider Spreads

Meta Platforms completed a massive bond issuance on US midterm Election Day. According to Bloomberg, Meta ultimately raised more than $46 billion, including $25 billion in dollar bonds, 9 billion euros in European bonds, $25 billion in common stock and $12.5 billion in preferred stock, and plans to raise up to an additional $60 billion through an at-the-market offering program. The deal was not smooth: S&P's rating outlook downgrade, Microsoft's large $25 billion bond sale competing for funds, and political uncertainty from the election successively disrupted the issuance schedule.[6]

Changes in financing costs are more telling. The longest-dated dollar bond in this sale, maturing in 2056, was priced at a spread 220 basis points over US Treasuries, about 90 basis points wider than the spread on similar Meta bonds issued in April. Oracle echoes this: after explicitly saying the borrowing would be only $10 billion and 'would not increase', and simultaneously issuing an equal amount of equity, it still had to pay a yield premium about 35 basis points above its existing bonds. Portfolio managers' judgment is that investors are demanding higher premiums and clearer explanations of how massive AI capex will actually translate into revenue and free cash flow.[6]

The issuance process itself also shows a shift in how mega-cap tech companies finance themselves. Before the sale, Meta's bankers pitched about 500 institutional investors in the US and Europe over two days by phone, with CFO Susan Li personally joining some meetings, and the company made clear it would not return to the unsecured bond market until at least the second quarter of next year. After completing the dollar bonds, Meta last Wednesday completed its first euro bond sale, 9 billion euros, attracting orders as high as 53 billion euros, setting a record for the largest order book for a single deal in the European corporate bond market.[6]

In aggregate, Bloomberg-compiled data show global investment-grade corporate and project financing has exceeded $415 billion so far this year; Goldman Sachs and Apollo are in talks with investors to sell portions of a roughly $40 billion chip financing deal for SpaceX. But signs of contraction are also clear: the latest dollar bond sales from Meta and Oracle were both below the previously customary level of more than $25 billion, as companies deliberately control the size of individual deals.[6]

China Prices: CPI Up 0.7%, PPI Highest Since June 2022

Data released by the National Bureau of Statistics on Nov. 9 showed China's October CPI rose 0.7% year on year, with the gain narrowing 0.2 percentage points from the previous month and flat month on month; core CPI, excluding food and energy, rose 0.8% year on year. Among components, AI-related hardware drew the most attention: data storage devices, tablets, computers and mobile phones rose 45.4%, 29.3%, 23.0% and 16.7% respectively, with the four items together contributing about 0.26 percentage points to the YoY CPI increase.[5]

Energy and food pulled in opposite directions. Gasoline prices rose 22.4% year on year, with the gain widening 5.8 percentage points from the previous month, contributing about 0.63 percentage points to the YoY CPI increase and driving the energy price gain to 9.9% from 7.3% last month; food prices fell 2.9% year on year, with the decline widening 1.2 percentage points, contributing about 0.49 percentage points to the YoY CPI decline, with fresh vegetables, pork and fresh fruit down 16.9%, 8.6% and 3.8% respectively. Service prices rose 0.7% year on year, with the gain down 0.1 percentage point from the previous month.[5]

Industrial prices rose faster. October PPI rose 5.7% year on year, with the gain widening 0.7 percentage points from the previous month, the highest since June 2022; it rose 0.7% month on month. Dong Lijuan, chief statistician of the NBS Urban Department, noted that higher prices in energy-related industries were the main driver of the PPI increase. Coal mining and washing rose 36.6% year on year, oil and gas extraction and petroleum, coal and other fuel processing rose 35.0% and 29.9% respectively, and chemical raw materials and chemical products manufacturing rose 16.4%; these nine industries together contributed about 6.14 percentage points to the YoY PPI increase, 0.64 percentage points more than the previous month.[5]

In the month-on-month structure, input and seasonal factors contributed most of the gain: rising international energy prices drove oil and gas extraction and coal mining and washing prices up 6.0% and 2.7% respectively; downstream refined petroleum product manufacturing, organic chemical raw material manufacturing, coal processing, and gas production and supply rose 6.8%, 6.8%, 3.2% and 1.5% respectively; these six industries together contributed about 0.61 percentage points to the MoM PPI increase, more than 80% of the total gain. AI's continued empowerment of industrial transformation also brought localized price increases, with electrical machinery and equipment manufacturing and computer, communications and other electronic equipment manufacturing both up 0.1%.[5]

A-shares and Hong Kong Stocks: AI Hardware Recovers, Physical AI and Space Computing Lead

On Nov. 9, A-shares rose collectively in morning trading, with the three major stock indices higher at the open and the Shenzhen Component and ChiNext up more than 1%. As of publication, the Shanghai Composite was up 0.79%, the Shenzhen Component up 1.11% and ChiNext up 1.69%; Hong Kong stocks fluctuated higher in the morning, with the Hang Seng Index up 0.61% and the Hang Seng Tech Index up 0.81%, though intraday gains narrowed. Space computing, physical AI, commercial aerospace and AI short drama concept stocks strengthened sharply, AI hardware stocks rebounded notably, and optical modules, optical chips, optical communications and memory chips all rebounded.[3]

At the individual stock level, on the AI hardware side, Victory Giant Technology and Yangtze Optical Fibre and Cable both rose about 8%, while Foxconn Industrial Internet, WUS Printed Circuit, Eoptolink, Shennan Circuits, Dongshan Precision and Zhongji Innolight all advanced; in the space computing sector, China Satellite, Junda and Shunhao hit the daily limit; in physical AI, Swancor New Materials rose more than 12%, and Unitree Robotics-W, Hanwei Electronics and Fulai New Materials all surged, with reports that PetroChina will set up a joint venture with Unitree Robotics; in AI comic drama, Zhongwen Tou Holdings hit the limit and Kunlun Tech rose more than 10%. In Hong Kong, Alibaba, Xiaomi, Baidu, Hua Hong Semiconductor and Tencent strengthened, while Zhipu and MiniMax rose 5%.[3]

Strong stocks and weak bonds appeared together. Government bond futures fell across the board, with the 30-year main contract down 0.19%, the 10-year down 0.06%, the 5-year down 0.03% and the 2-year down 0.01%. Domestic commodity futures were mixed: platinum rose 3%, Shanghai silver rose 2%, Shanghai aluminum, Shanghai tin and Shanghai gold rose 1%, while polysilicon, rubber, Shanghai copper and stainless steel advanced; industrial silicon, alumina, soybean meal, rebar, glass, coke, iron ore and coking coal fell, manganese silicon and lithium carbonate dropped 1%, and asphalt fell more than 3%. A previous State Council executive meeting noted the need to improve the capital market's market-stabilization mechanism and enhance the stock market's intrinsic stability.[3]

Iran's 'Iron Gate' Task Force and 'Extremely Low' Probability of Another Strike

Geopolitical moves were concentrated on the Iran corridor. On Nov. 7 local time, General Cooper, commander of US Central Command, ordered the formation of a task force called 'Iron Gate', comprising about 20 civilian and military personnel and stationed in Tampa, Florida, to support ongoing US government operations and prevent cross-border flows of advanced weapons systems, dual-use technologies, sanctioned goods and 'illicit funds' through Iran's land borders and the Caspian Sea.[1]

On Nov. 8 local time, US President Trump posted on social media that the 'Iron Gate' task force operation against Iran was in full swing, and together with what he called the 'largest naval blockade in history', almost no goods could enter or leave Iran. On the same day, he also said that as long as talks over Iran's nuclear program and the reopening of the Strait of Hormuz made progress, the possibility of another US military strike on Iran was 'extremely low', with the key being whether Iran was willing to sign a 'very strong agreement', and stressed that the US could still take military action if necessary.[1]

Military preparations and negotiation signals run in parallel. Trump said the US military was ready to strike again, with three aircraft carriers currently deployed in the relevant region, and said the US had identified infrastructure targets in Iran including power plants and bridges, including 38 bridges that could be destroyed within an hour, but he did not want to take such action because rebuilding those facilities could take 10 years. Since the US and Israel jointly launched a large-scale military operation against Iran on Feb. 28, Trump has repeatedly made contradictory remarks about ending the Iran war; the US and Iran are now in a standoff, with little diplomatic progress.[1]

Diplomatic channels are still operating. According to Al Hadath, the news channel of Saudi Arabia's Al Arabiya, on Nov. 8, an Iraqi delegation recently visited Tehran, held talks with Iranian officials and conveyed a US message; the visit lasted about eight hours and took place after US envoy Tom Barrack visited Baghdad and Trump spoke by phone with Iraqi Prime Minister Al-Zaidi. Al Arabiya also quoted sources on Nov. 7 as saying the US had warned the Iraqi government that if armed factions in Iraq joined Iran's side in a renewed US-Iran conflict, their headquarters and bases would become strike targets.[1]

Dollar and Yen: AI Selloff Has Not Yet Shaken the Dollar

Barclays' latest FX strategy report concludes that the AI-related equity selloff has not yet shaken the dollar's appeal. The report noted that existing data center construction contracts will proceed as planned and capex is unlikely to be cut sharply in coming quarters; hyperscalers show no sign of slowing their 'borrow-to-invest' plans and are expected to achieve positive cash flow by the end of next year. On the macro level, US private domestic purchasers' final sales growth is near 5%, the economy is running above potential output and inflation is still stuck near 3%; the bank believes the capex cycle will continue, with few signs of a sharp downgrade over the next year.[4]

Valuation and positioning corroborate this judgment. The dollar risk premium has fallen to about 1%, the lowest in more than a year; the dollar sentiment indicator remains in slightly positive territory, with no net short signal yet. Barclays said the market may see a 'US slowdown imminent' narrative and push EUR/USD briefly to around 1.14, but given the continued resilience of the US macro backdrop, this move is 'unlikely to last' both now and on a forward-looking basis. The euro's appeal as an 'anti-dollar' alternative is fading, with French OAT spreads widening sharply, the upcoming presidential election and a hung parliament making consensus difficult, and market expectations for the ECB's terminal rate revised to about 3%.[4]

The yen, meanwhile, is supported by safe-haven demand. Barclays offers scenario estimates: if the S&P 500's decline widens from about 6% currently to 10%, USD/JPY could fall about 1.5% toward 155; if the decline widens to 20%, the yen could appreciate 4% to 5%. However, since coordinated intervention in the summer, speculative positioning has shifted overall to net long yen, which constrains the pace of further yen appreciation. Asian currencies are clearly diverging: the Korean won is dragged down by tech stock volatility, with foreign net equity outflows of about $5.9 billion and retail net outflows of about $2 billion this month, and USD/KRW faces near-term upside risk with 1,325 as support; the Indian rupee benefits from the sharp drop in oil prices, low AI exposure in its stock index and about $2.4 billion of foreign net inflows during the month, but watch the expansion of the RBI's forward book and the decline in FX reserves; USD/INR is expected not to stay below the 200-day moving average of 94.22 for long.[4]

Common Trends and What to Watch Next

Putting these threads together, AI capex is a common variable running through stocks, bonds and prices: on the equity side, SOX's 38% drawdown is front-loading an earnings downgrade; on the financing side, the spread on Meta's dollar bonds maturing in 2056 is about 90 basis points wider than in April, reflecting investors' demand to price the payback period of AI investment; on the price side, four AI-related goods including storage devices together lifted China's YoY CPI by about 0.26 percentage points, while energy prices dominated the PPI increase, contributing about 6.14 percentage points through nine industries.[2][5][6]

On timing, the first quarter of 2027 is seen by BofA as the overlap point when SOX may bottom and EPS downgrades may begin to appear, and policy clarity at the US Congress level is seen as a potential confidence catalyst; Lam Research's $25 billion buyback, Nvidia's ongoing buyback and Micron's buyback set to start on Dec. 9 form valuation-side support. On the financing side, the indicators to watch are tech bond spreads and the issuance window—Meta has said it will not return to the unsecured bond market until at least the second quarter of next year, and after Democrats won majorities in both the Senate and House, about half of new US AI data center projects face stricter local review, which could both reduce borrowing demand and lengthen the investment payback period.[2][6]

Two threads—real economy and geopolitics—are also worth tracking. Energy was the main driver of the October PPI increase, while tighter Iranian land and sea channels and progress in talks to reopen the Strait of Hormuz will directly affect oil and natural gas prices; in turn, oil price swings affect CPI through gasoline prices and transmit to the dollar and yen through safe-haven sentiment and risk premiums. Whether AI capex can move from valuation discount to earnings delivery must ultimately be verified along this chain of energy, exchange rates and financing costs.[1][4][5]

Sources

← All Finance stories