Semiconductor stocks just suffered their worst week since 2008 — while TSMC posted a record profit and Big Tech kept spending. Inside the great AI repricing.
Published 19 October 2026 · 13:10 GMT
The artificial-intelligence trade just lived through its ugliest week in nearly two decades. The Philadelphia Semiconductor Index — the benchmark tracking 30 of the world's biggest US-listed chipmakers, from Nvidia to AMD — fell for eight straight sessions through Friday, shedding about 16% in its longest losing streak since the 2008 global financial crisis (CNA, Oct 19).
Friday capped the damage on Wall Street: the Nasdaq Composite lost 2.14%, bringing its weekly slide to roughly 4.5% — its worst week since June — while the S&P 500 fell 1.09%, putting its weekly losses at almost 2% (CNN, Oct 16). On Monday the sell-off rolled into Asia: South Korea's KOSPI broke below 5,500 in early trading, opening down 3.17% at 5,488.41, while Japan's Nikkei 225 plunged as much as 1,493 points intraday (Bloomberg, Oct 19).
What makes this rout unusual is what did not happen: demand did not collapse. The world's largest contract chipmaker reported another record quarter in the middle of the storm. The market is no longer asking whether AI demand is real. It is asking whether anyone can afford to keep paying for it.
The verified ledger of the week is stark. In the United States, the eight-session SOX slide followed a bruising Wednesday: the S&P 500 fell 1.11% and the Nasdaq 1.88% — each the worst day since July — with Marvell down 8.7%, Oracle 8.8% and Dell 11.3% (CNN, Oct 14). Thursday brought the chip index's seventh straight decline even as the broader market steadied (CNN, Oct 15).
Asia bore the heaviest selling. South Korea's KOSPI closed Thursday at 5,554.22, down 7.57% on the day — a six-month low and its seventh consecutive decline (Yonhap Infomax). Samsung Electronics plunged 17.8% and SK hynix 24.3% from Monday to Friday, each the steepest weekly drop since the 2008 crisis; foreign investors dumped a net 11.21 trillion won (about US$8.4 billion) of Korean shares, nearly four-fifths of it in those two names (The Investor, Oct 18).
Japan and Taiwan were not spared: Kioxia plunged nearly 24% last week and TSMC fell about 5% — despite a net profit of NT$763.1 billion (US$23.5 billion), up 69% from a year earlier, its sixth consecutive record quarter, on revenue of NT$1.5 trillion, up 50.9% on 'extremely robust' demand (CNA, Oct 19).
The trigger underneath was inflation. US consumer prices rose 0.6% in September, the largest monthly increase since April, taking the annual rate to 3.7% (Al Hayat/AFP, Oct 15). The Federal Reserve raised rates by 25 basis points last month — its first increase since 2023 — and signalled more, with its next decision due October 28 (CNA, Oct 19). And the market found its stress test in CoreWeave, whose US$50 billion debt pile sent its shares down more than 15% on the week — including 12% on Friday, a day after Nvidia chief Jensen Huang publicly endorsed the company (CNA; CNN, Oct 16).
Unresolved is whether this is a correction or a repricing of the entire AI investment thesis. Nobody knows yet whether Microsoft, Amazon, Alphabet and Meta — expected to spend around US$600 billion this year on data centres, chips and AI infrastructure (CNA, Oct 19) — will hold that pace. Goldman Sachs forecasts hyperscaler capital expenditure near US$800 billion in 2026 and US$1.1 trillion in 2027, but those are forecasts, not commitments.
Nor is it clear whether financing stress stays contained: the cost of insuring Oracle, SoftBank and CoreWeave against default has widened, in some cases to records. A new variable arrived Monday — DeepSeek released a cheaper model, raising fresh questions about memory demand (Bloomberg, Oct 19), with its revenue impact unquantified. And the Federal Reserve's October 28 decision remains the macro hinge.
This matters because the AI trade had become the market's load-bearing wall. The SOX was up more than 70% year to date before the slide, and the chip-stock rout put roughly US$2 trillion of market value at risk in a single week. In Seoul, Samsung and SK hynix account for nearly half of the KOSPI's market capitalisation; their slide is the index's slide.
There is a deeper question in play, the one Saxo Singapore's Charu Chanana keeps returning to: 'There is an important difference between spending money on AI and making money from AI.' Technology companies are investing enormous sums in chips, data centres and electricity, and investors ask whether AI revenues will grow fast enough to justify it. As Chanana put it: 'The biggest risk is not necessarily that AI fails. It is that AI succeeds, but investors have paid too much for the profits it eventually generates.'
Strip out the daily noise and the structure is a timing mismatch. Geopolitically, China's memory challenger CXMT is advancing on denser architectures while DeepSeek's cheaper models question how much premium hardware the world needs — competition arriving as financing tightens. Economically, the mechanism is the discount rate: 'Higher rates make future profits less valuable in today's terms, which tends to hurt expensive growth stocks,' Chanana told CNA, while raising the cost of borrowing to build the infrastructure itself.
Anthony Tung, a professor at the National University of Singapore's School of Computing, located the deeper fault line: 'With AI, the value is real. The timing is where expectations run ahead.' Much of today's adoption sits in drafting, summarising and assisting programmers; use in core manufacturing and healthcare operations is limited. Yet capacity is being built for demand expected several years out. 'If adoption lags, there will be a correction in valuations and spending,' Tung said.
The market has also split the AI trade in two: the demand layer (chipmakers with order backlogs and record profits) and the financing layer (debt-heavy builders like CoreWeave). The first is dented; the second concentrates the stress. That is why a record TSMC quarter and a 16% chip-index slide can coexist — less a verdict on AI than an audit of who pays for it, and when.
Second-order effects are moving through credit and primary markets. CoreWeave's bonds trade at distressed levels — a reminder, as Chanana notes, that financing stress shows up before it appears in earnings. In primary markets, risk has become the filter: Firmus shelved its US$5 billion listing, and debt investors have grown selective toward technology financing while long-term rates stay elevated.
For chip-heavy economies the exposure is macroeconomic. South Korea's won weakened toward 1,344.6 per dollar and its financial authorities held an emergency meeting, pledging a proactive response (South Korea). Japan's market, carried to records by the same semiconductor names now leading the decline, faces the same concentration in reverse. If Big Tech disciplines its capital spending, the adjustment lands hardest on equipment and memory suppliers — which is why upcoming results from Nvidia, Samsung and SK hynix are now the market's most-watched prints.
Chanana named the four indicators that decide what comes next: inflation and interest rates, financing conditions, Big Tech's earnings and spending plans, and semiconductor demand. 'Investors will want to see not just whether companies such as Microsoft, Alphabet, Amazon and Meta continue investing in AI, but whether those investments are delivering higher revenues, stronger productivity and ultimately better profits,' she told CNA.
The calendar is unforgiving: the Federal Reserve decides on October 28, and Big Tech earnings arrive in the same window. Chipmakers' forecasts — Nvidia, Samsung, SK hynix — will show whether demand is holding and whether competition and new capacity are squeezing margins. The honest signals of a deeper downturn: major technology companies materially cutting AI spending, semiconductor orders weakening, or financing problems forcing data-centre projects to be delayed or cancelled. None has happened yet. Until one does, this remains what RHB Singapore's Alfie Yeo called it: 'a correction as the market feels overvalued.'
Wall Street reads this as a repricing of growth expectations, not a demand collapse. 'It's unclear that CoreWeave is going to get a sufficiently good return to pay down the debt that they borrowed,' Gil Luria of D.A. Davidson told CNN, calling the company 'the subprime tranche of the AI trade.' BlackRock's research arm downgraded its outlook for US stocks to neutral, citing 'potential volatility around the AI buildout' (TechCentral). JPMorgan analysts wrote that 'financing risks and IPO timelines' remain at the centre of the story.
Asia reads it as a paradox. Yonhap Infomax called it 'KOSPI's paradox': South Korean chipmakers are shattering earnings records while valuations sit at crisis levels, the KOSPI's price-earnings ratio near 8 times. 'Even strong earnings may no longer be enough when investors have already priced in years of exceptional growth,' said Saxo Singapore's Charu Chanana. 'The AI trade is moving from excitement about growth to questions about the cost of delivering that growth.' Singapore's central bank expects AI activity to keep backing expansion next year, at a slower pace; South Korea expects robust semiconductor exports to support solid growth; Taiwan expects AI demand to keep supporting exports and investment (CNA).
The Global South reads it as systemic risk. 'The consequences of an AI correction could be global,' the IMF warned in its latest global financial stability report (TechCentral, South Africa). The stress is visible in primary markets: Firmus, a Nvidia-backed Australian data-centre operator, shelved a US$5 billion IPO; Tencent's bond sale came under scrutiny; SoftBank sold the biggest high-yield bond deal on record, offering up to 9.75% on US$4.5 billion. In the Arab press, Al Hayat framed the sell-off around the Fed's October 28 decision and Asia's pain.
Investors no longer question whether AI demand is real — TSMC just posted a record NT$763.1 billion quarterly profit on 'extremely robust' demand. They question the returns on the roughly US$600 billion Big Tech will spend on AI infrastructure this year. Hotter US inflation revived Fed rate-hike bets, devaluing future profits and raising borrowing costs for debt-heavy builders like CoreWeave. The market is repricing who pays for the buildout, not denying it.
The Philadelphia Semiconductor Index fell about 16% over eight straight sessions through October 16 — its longest losing streak since the 2008 crisis. The Nasdaq lost 4.5% for the week, its worst since June. In Asia, Samsung fell 17.8% and SK hynix 24.3% for the week — each the steepest weekly drop since 2008 — while the KOSPI closed down 7.57% on Thursday. Foreign investors dumped a net 11.21 trillion won of Korean shares in one week.
CoreWeave is a US cloud company — a 'neocloud' — renting AI computing power to AI developers; it went public in 2025 backed by Nvidia. It borrowed to fund growth and now carries about US$50 billion in debt. Its shares plunged more than 15% last week and 12% on Friday, even after Nvidia's Jensen Huang endorsed it, as investors doubt its returns can cover its borrowing. Gil Luria of D.A. Davidson called it 'the subprime tranche of the AI trade.'
Four things, per Saxo Singapore strategist Charu Chanana: inflation and interest rates — the Fed decides October 28; financing conditions, especially stress among indebted AI infrastructure firms; Big Tech earnings and spending plans from Microsoft, Alphabet, Amazon and Meta; and semiconductor demand in upcoming results from Nvidia, Samsung and SK hynix. Genuine warning signs of something deeper: Big Tech materially cutting AI spending, weakening chip orders, or financing problems delaying data-centre projects.
