While tech giants are still fiercely competing for wafer capacity from Taiwan Semiconductor Manufacturing, NVIDIA CEO Jensen Huang has set his sights on a longer-term goal—directly securing the land for TSMC's future fabs. This strategy highlights how NVIDIA, against the backdrop of exploding demand for AI chips, is attempting to fundamentally ensure its supply chain security.
On January 15, TF International Securities analyst Ming-Chi Kuo tweeted, "Most clients are negotiating for guaranteed capacity, while Jensen Huang is negotiating for land." Kuo pointed out that during a visit to Tainan last November, Huang expressed to TSMC his willingness to pay to ensure the reservation of the P10 and P11 plots adjacent to the Fab 18 campus, with the ultimate goal of also securing the P12 plot. At that time, the Fab 18 plan was only clearly defined up to plots P1 through P9.
This move comes as TSMC announced a record level of capital expenditure for 2026, reaching $52 to $56 billion, far exceeding previous market expectations of $45 to $52 billion. TSMC clearly stated during its earnings call that NVIDIA's strong demand is a key driver behind this increase in spending.
Simultaneously, traditional major clients like Apple are facing pressure in securing capacity, with NVIDIA potentially having surpassed Apple to become TSMC's largest customer in some quarters of 2025. The strategic layout behind the frequent visits to Taiwan
Jensen Huang visited Taiwan three times in the second half of 2025, a frequency that has drawn market attention.
According to Digitimes, following an August 22nd meeting with TSMC to discuss production plans for the next-generation Rubin GPU architecture, Huang arrived at Tainan Airport via private jet on November 7th. He toured TSMC's 3-nanometer wafer fab in the Southern Taiwan Science Park, dined with executives including TSMC Chairman C.C. Wei, and attended TSMC's annual sports day the following day.
On November 27th, during the US Thanksgiving holiday, Huang appeared again in Taipei, visiting the Siping Street and Dazhi areas. Although he publicly stated it was just to "visit friends," sources revealed that the main purpose of the trip was to engage with Quanta Computer. Given market concerns about Quanta's assembly yield and its ability to secure more orders amidst NVIDIA's tightening procurement controls, and considering Huang's close relationship with Quanta Chairman Barry Lam, this visit was seen as a discussion on ongoing collaboration strategies.
This marks Huang's fifth visit to Taiwan since the beginning of 2025. NVIDIA's plans for new offices on T17 and T18 plots in Taipei's Beitou-Shilin Technology Park have been finalized, demonstrating the company's long-term commitment to the Taiwanese supply chain. The capacity battle reshapes the client landscape
According to analysis from Culpium and supply chain sources, NVIDIA may have surpassed Apple to become TSMC's largest customer in at least one or two quarters of 2025. Although TSMC CFO Wendell Huang declined to comment on changes in client rankings, the public data speaks for itself.
TSMC's revenue grew 36% in 2025 to $122 billion, while NVIDIA's sales for the fiscal year ending January 2026 are projected to grow 62%. Apple's product revenue is estimated to have grown only 3.6% in the 12 months ending December 2025. TSMC's High-Performance Computing business grew 48% last year, following 58% growth the previous year; meanwhile, its smartphone business grew just 11%, down from 23% the year before.
According to Culpium, during a visit to Cupertino last August, C.C. Wei delivered news of the largest price hike in years to Apple executives. Worse for Apple, the company that once dominated TSMC's client list now finds itself fighting for production capacity. As the AI boom continues, the GPUs from clients like NVIDIA and AMD occupy larger areas on each wafer, meaning Apple's chip designs no longer guarantee priority across TSMC's nearly twenty fabs. The technology roadmap dictates the short-term competitive landscape
TSMC is currently mass-producing chips on its 2-nanometer node (N2), with Apple as the primary buyer. However, in the second half of this year, TSMC will simultaneously begin mass production of a new variant, N2P, and a new node called A16. Wei stated that A16, equipped with Super Power Rail technology, is "most suitable for high-performance computing with complex signal routing." This technology is TSMC's version of backside power delivery, designed to separate a chip's signals and power.
This technology roadmap favors companies like NVIDIA and AMD over the next one to two years. But according to SemiAnalysis, the A14 node, expected to enter mass production in 2028, is "designed from the ground up for both mobile and high-performance computing," at which point the balance is expected to shift back towards Apple.
TSMC's revenue for 2026 is projected to grow nearly 30%, but capital expenditure is set to increase by about 32%. Long-term, growth is expected to average 25% over the five years through 2029, but the AI business is projected to average 55% or higher growth over the same period, up from previous forecasts in the 40% range. TSMC's gross margin reached a staggering 62.3% in the December quarter, an improvement of 280 basis points from the previous quarter.
Despite the pressure to expand capacity, Wei expressed caution at an investor conference: "I am also very nervous. If we are not careful, this will definitely bring huge disaster to TSMC." TSMC's capital intensity exceeds 33%, and depreciation accounts for 45% of its cost of revenue, far higher than clients like Alphabet and NVIDIA. This means TSMC would bear the primary risk if demand falls, which also explains why the company must weigh speed against risk when expanding production.
Comments