NEW YORK, December 31, 2025, 17:33 ET — After-hours
- Nvidia sought talks with TSMC on boosting H200 GPU output as China demand jumped, sources told Reuters.
- Mega-cap AI names traded softer late, with Nvidia, Microsoft and Meta down modestly in extended trade.
- Investors head into 2026 watching AI capital spending, U.S.-China tech controls and the pace of chip supply.
Nvidia has approached Taiwan Semiconductor Manufacturing Co about ramping production of its H200 graphics processing units (GPUs) — chips used to train AI systems — after Chinese tech companies placed orders for more than 2 million units for 2026, far above the roughly 700,000 Nvidia has in stock, sources told Reuters. Beijing has yet to clear imports and the Trump administration only recently allowed H200 exports to China subject to a 25% fee, the people said.
The latest supply-and-policy test lands as investors try to gauge how much of the AI trade can carry into 2026 after a year dominated by spending on chips and data centers. For another year of strong double-digit returns, markets need “everything firing on all cylinders,” said Sam Stovall, chief investment strategist at CFRA. Reuters
U.S. stocks ended the year lower on Wednesday, with the S&P 500 down 0.74% and the Nasdaq Composite down 0.76% in thin holiday trading, Reuters reported. Markets are closed on Thursday for New Year’s Day.
In after-hours trade, Nvidia was down about 0.5% at $186.50, while Microsoft slipped 0.8% and Meta fell 0.9%. Chip-linked peers also eased, with AMD down 0.5%, Broadcom down 1.1%, Micron down 2.5% and Marvell down 2.1%; Amazon and Alphabet edged lower. TSMC rose about 1.4%.
Sources cited by Investing.com said Nvidia has asked TSMC to begin producing additional H200 chips, with work expected to start in the second quarter of 2026, and plans to price the chip at about $27,000, with final pricing varying by volume and customer terms. The report also said Nvidia expects initial shipments for China before the Lunar New Year holiday in mid-February.
The China angle matters because Beijing has been pushing its chip industry to rely less on foreign technology. Chinese authorities have told some domestic chipmakers to use locally produced equipment for at least 50% of their purchases in 2026, Reuters reported, a move that could reshape supply chains and pricing for global vendors.
Data-center buildouts remain a near-term demand signal for the sector. Elon Musk said xAI bought a third building to expand infrastructure, aiming to lift training capacity to nearly 2 gigawatts and, longer term, expand its Memphis “Colossus” cluster to at least 1 million GPUs, Reuters reported. Reuters
Deal activity is also feeding the AI narrative on Wall Street. Meta said it would acquire Chinese-founded AI startup Manus and integrate the service into its products, including Meta AI; a source said the deal values the Singapore-based firm at $2 billion to $3 billion, Reuters reported.
In cloud and infrastructure, Brookfield said it is launching a cloud business called Radiant to let companies lease AI chips housed in its data centers, Reuters reported, adding another entrant to a market dominated by big hyperscalers.
Nvidia is also exploring software and model assets. The company is in advanced talks to buy Israeli AI startup AI21 Labs for up to $3 billion, Reuters reported, in a move that would deepen its footprint beyond silicon.
U.S.-China tech controls stayed a live factor for chip stocks late in the year. Washington granted TSMC an annual export license allowing it to import U.S.-made chipmaking equipment for its China operations, Reuters reported, as prior authorizations were set to expire at year-end.
China’s domestic buildout is also moving into key AI components beyond processors. Chinese memory chipmaker ChangXin Memory Technologies has tapped advisers for a Shanghai IPO and is investing in high-bandwidth memory (HBM) — the stacked memory used alongside AI accelerators — Reuters reported.
With U.S. markets shut on Thursday, traders head into Friday watching two pressure points: whether China approvals and U.S. export rules keep shifting, and whether AI “capex” — capital spending on chips, servers and power — holds up as companies set 2026 budgets. Any hint of slowing orders or supply bottlenecks can quickly spill across the AI supply chain, from chipmakers to cloud and data-center builders.
