Nvidia CEO Jensen Huang traveled to Taiwan ahead of quarterly earnings and held meetings with TSMC executives to secure guaranteed wafer and advanced packaging capacity for upcoming AI chip generations.
Nvidia CEO Jensen Huang arrived in Taiwan on a private jet on August 23, marking his third visit to the island this year. The timing coincides with days before Nvidia’s scheduled August 26 earnings release. Market observers believe Huang utilized this pre-earnings window to hold closed-door meetings with TSMC and other core supply chain executives, finalizing key matters including next-generation AI chip capacity allocation, advanced packaging quotas, and recent server pricing adjustments. This marks the second consecutive year Huang has visited Taiwan in late August ahead of earnings to secure supply chain commitments. On August 22 of last year, he made a similarly unannounced half-day stop, meeting directly with TSMC Chairman C.C. Wei and senior executives to personally lock in production schedules and capacity for the Vera Rubin platform chips. Flying to Taiwan ahead of August earnings to shore up commitments has now become a key strategy for Huang to confirm the upper limits of supply chain capacity.
Industry analysts note that AI server demand remains red-hot, with high-bandwidth memory (HBM) and TSMC’s CoWoS advanced packaging capacity remaining tight. Recent reports indicate that next-generation servers will see price increases of more than 15% due to rising memory costs. With cloud hyperscalers showing no signs of slowing procurement, Huang’s personal visit serves not only to secure advanced process and packaging schedules from the second half of this year through the first half of next year, but also to coordinate assembly timelines in real time with contract manufacturers such as Hon Hai (2317.TW) and Quanta Computer (2382.TW).
According to Bloomberg, some of Nvidia’s major customers have been informed that servers equipped with its AI chips will see system prices rise by more than 15% in many cases for shipments early next year, with the magnitude varying by chip generation and memory configuration. Both the flagship Vera Rubin and Grace Blackwell platforms are expected to be affected. Companies contracted to manufacture servers for major data center operators such as Microsoft, Google parent Alphabet, and Oracle have recently notified customers of the price increases. Nvidia has not yet responded to requests for comment. This wave of price increases most directly reflects the continued rise in AI server memory costs. As AI servers drive surging demand for HBM, server DRAM, and high-speed storage, global memory manufacturers are shifting more capacity toward premium products, squeezing supply of conventional DRAM and niche memory in the process.
Based on market estimates that a high-end AI server equipped with Nvidia chips costs approximately $250,000 to build, a price increase of more than 15% translates to an additional procurement cost of over $37,500 per unit—a meaningful impact on cloud hyperscalers' capital expenditure budgets.
With next-generation Vera Rubin and Grace Blackwell platforms shipping next year, the beneficiary theme for Taiwan's AI supply chain is heating up again. Six stocks have emerged as market focal points: Taiwan Semiconductor Manufacturing (2330.TW), Nanya Technology (2408.TW), Winbond Electronics (2344.TW), Kinsus Interconnect Technology (3189.TW), Hon Hai, and Quanta Computer. Among these, Nanya Technology primarily benefits from tightening DRAM supply-demand dynamics and rising quotes. Winbond Electronics, with its product portfolio spanning niche DRAM and NOR Flash, stands to benefit from improving memory pricing and demand as AI server-related applications expand. TSMC, while not a direct beneficiary of the memory price increases, remains the core of Nvidia's AI chip supply chain. New platforms like Vera Rubin utilize advanced process nodes and advanced packaging. If large cloud service providers maintain procurement and capital expenditure even after AI server prices rise, it would further confirm the resilience of AI computing demand, supporting sustained demand for TSMC's advanced process and CoWoS capabilities.
According to TrendForce estimates, TSMC's CoWoS advanced packaging monthly capacity could expand to 120,000–140,000 wafers in 2026. Adding the 50,000–60,000 wafers of new capacity from outsourced semiconductor assembly and test (OSAT) partners, total industry monthly capacity could approach 200,000 wafers, reflecting an accelerating pace of advanced packaging expansion to keep up with AI chip demand. Kinsus benefits from ABF substrate demand driven by AI GPUs and high-performance computing chips. As next-generation AI chips continue to increase in package size, layer count, and transmission specifications, the unit value of high-end substrates is also expected to rise.
For AI server ODM manufacturers Hon Hai and Quanta, both stand to benefit from Nvidia's continued new platform ramp and expanding AI server shipment scale. However, server price increases primarily stem from rising component costs. For ODM manufacturers, revenue may rise alongside higher system unit prices, but actual profitability still depends on material cost pass-through and product mix. Therefore, a 15% server price increase does not translate to a simultaneous 15% increase in profits.
Taiwan's stock market rose 290.55 points on August 21, closing at 45,224.29. Institutional investors noted that Nvidia's earnings release this week and Federal Reserve Chair Kevin Warsh's speech at the global central bank symposium are the key market events to watch. U.S. stocks closed mostly higher last Friday, with the Dow Jones Industrial Average rising 517.8 points, or 0.98%; the S&P 500 gaining 33.21 points, or 0.43%; the Nasdaq Composite advancing 113.29 points, or 0.43%; while the Philadelphia Semiconductor Index fell 59.644 points, or 0.51%. TSMC's American Depositary Receipts (ADRs) rose $2.95, or approximately 0.71%. Taiwan index futures overnight trading was relatively weak, fluctuating down 64 points to close at 45,074.
Senior analyst Wang Zhaoli said that Nvidia, the leading AI chip company, is about to report earnings, making it the headline event for the market this week. With AI continuing to flourish, Nvidia's operational performance is expected to deliver impressive results. However, Nvidia's announcement of a $500 billion financing partnership with six major asset management firms, including Apollo, will be a key focus for investors. The six major financing platform partners in Nvidia's collaboration were highlighted. Market consensus expects Nvidia's revenue for the current quarter (to be reported on August 26) to land in the $93 billion to $95 billion range, representing year-over-year growth of nearly 96%, slightly above Nvidia's own guidance of approximately $91 billion. Nvidia's stock closed at $214.72 last Friday, down approximately 9% from its peak. Compared with chip stocks like Micron, Intel, and Qualcomm, which have fallen 20% to 40%, Nvidia has shown relative resilience, likely supported by capital positioning.