Podcast Episode
TSMC Tells Nvidia and Broadcom It Cannot Meet Full AI Chip Demand
January 19, 2026
Audio archived. Episodes older than 60 days are removed to save server storage. Story details remain below.
Taiwan Semiconductor Manufacturing Company has informed its largest customers, including Nvidia and Broadcom, that it cannot fulfill their complete production capacity requests, revealing unprecedented strain on global semiconductor supply chains as artificial intelligence demand continues to surge.
The world's dominant manufacturer of advanced processors faces a capacity shortfall that extends across both chip fabrication and critical advanced packaging technologies. The situation highlights the scale of the AI infrastructure buildout underway across the technology industry.
The company manufactures chips for nearly every major technology firm, including Apple, AMD, Nvidia, and Broadcom. TSMC holds approximately ninety percent of the market for the world's most advanced semiconductors, making it an irreplaceable link in the global AI supply chain.
Broadcom is specifically seeking additional capacity for tensor processing units designed for Google. Despite TSMC's aggressive expansion efforts, including a one hundred and sixty five billion dollar commitment to facilities in Arizona, production capabilities remain overwhelmed by customer requests.
Nvidia has secured more than half of TSMC's available CoWoS capacity for 2026 and 2027, reportedly booking between eight hundred thousand and eight hundred and fifty thousand wafers for 2026 alone. This dominant allocation leaves competitors scrambling for alternatives.
The shortage has forced major customers to adjust their production targets. Google has reduced its 2026 production goal for Tensor Processing Units from four million to three million units due to limited access to CoWoS capacity. The search giant is now exploring Intel and Samsung as potential packaging alternatives.
By the end of 2025, TSMC nearly doubled its CoWoS output to seventy five thousand wafers per month. The company projects this will reach one hundred and thirty thousand wafers per month by the end of 2026, effectively doubling capacity again within a single year.
The second Arizona fabrication facility is now expected to begin three nanometre chip production in 2027, ahead of the original 2028 timeline. TSMC's first Arizona facility is already operational, producing chips on four nanometre technology.
As part of a pending United States-Taiwan trade agreement, TSMC may add at least five more fabrication facilities in Arizona beyond its current six-plant commitment. The company has repeatedly emphasized its commitment to accelerating the Arizona buildout.
We continue to speed up our capacity expansion in Arizona, Wei said during an October earnings call. We are making tangible progress and executing well to our plan.
TSMC reported fourth-quarter 2025 earnings in January 2026, delivering results that exceeded analyst expectations. The company posted quarterly revenue of thirty three point seven three billion dollars, up twenty five point five percent year-over-year, and full-year 2025 revenue of one hundred and twenty two point four two billion dollars.
Net income for the fourth quarter reached sixteen point three one billion dollars, a thirty five percent year-over-year increase, with gross margin of sixty two point three percent. Analysts had expected a twenty seven percent increase in net profit to record levels.
The company expects AI accelerator revenue to grow by at least fifty percent annually through 2029. High-Performance Computing now represents fifty eight percent of TSMC's total revenue, underscoring the company's deep dependence on AI-related chip production.
Nvidia's dominance of TSMC's advanced packaging capacity gives it priority over competitors in a market where demand far outstrips supply. The limited availability of Nvidia's GPUs and Broadcom's ASICs, both in high demand for AI applications, continues to constrain infrastructure buildouts at major cloud providers and enterprises.
The situation at TSMC confirms the strength and sustainability of the AI boom rather than signaling weakness. When the world's most important chipmaker says capacity needs to triple just to meet existing customer demand, it provides concrete validation of the scale and longevity of AI infrastructure investments underway across the technology sector.
The Three-Fold Shortage
TSMC Chairman and CEO C.C. Wei delivered a blunt assessment of the situation at the Semiconductor Industry Association Awards in November. When asked about capacity, Wei responded, not enough, not enough, still not enough. He estimated that TSMC's existing advanced-node capacity falls roughly three times short of what major customers plan to consume.The company manufactures chips for nearly every major technology firm, including Apple, AMD, Nvidia, and Broadcom. TSMC holds approximately ninety percent of the market for the world's most advanced semiconductors, making it an irreplaceable link in the global AI supply chain.
Broadcom is specifically seeking additional capacity for tensor processing units designed for Google. Despite TSMC's aggressive expansion efforts, including a one hundred and sixty five billion dollar commitment to facilities in Arizona, production capabilities remain overwhelmed by customer requests.
The Packaging Bottleneck
Beyond raw wafer production, TSMC faces severe constraints in advanced packaging technology. The company's CoWoS packaging capacity, essential for high-performance AI accelerators, is completely sold out through 2026. CoWoS stands for Chip-on-Wafer-on-Substrate, a sophisticated packaging technique that enables the performance levels required by modern AI chips.Nvidia has secured more than half of TSMC's available CoWoS capacity for 2026 and 2027, reportedly booking between eight hundred thousand and eight hundred and fifty thousand wafers for 2026 alone. This dominant allocation leaves competitors scrambling for alternatives.
The shortage has forced major customers to adjust their production targets. Google has reduced its 2026 production goal for Tensor Processing Units from four million to three million units due to limited access to CoWoS capacity. The search giant is now exploring Intel and Samsung as potential packaging alternatives.
By the end of 2025, TSMC nearly doubled its CoWoS output to seventy five thousand wafers per month. The company projects this will reach one hundred and thirty thousand wafers per month by the end of 2026, effectively doubling capacity again within a single year.
Accelerated Arizona Expansion
TSMC is significantly expanding its United States presence despite the ongoing capacity constraints. Last week, the company purchased nine hundred acres of Arizona state trust land for approximately one hundred and ninety seven million dollars, expanding its footprint near Phoenix for future fabrication plants and packaging facilities.The second Arizona fabrication facility is now expected to begin three nanometre chip production in 2027, ahead of the original 2028 timeline. TSMC's first Arizona facility is already operational, producing chips on four nanometre technology.
As part of a pending United States-Taiwan trade agreement, TSMC may add at least five more fabrication facilities in Arizona beyond its current six-plant commitment. The company has repeatedly emphasized its commitment to accelerating the Arizona buildout.
We continue to speed up our capacity expansion in Arizona, Wei said during an October earnings call. We are making tangible progress and executing well to our plan.
Pricing Power and Record Earnings
The supply crunch has given TSMC significant leverage to implement price increases ranging from three percent to ten percent on advanced nodes, effective January 2026. Analysts at Goldman Sachs now expect TSMC's gross margins to reach and hold above sixty percent through 2028. J.P. Morgan projects margins could rise further due to hot-runs triggered by supply tensions.TSMC reported fourth-quarter 2025 earnings in January 2026, delivering results that exceeded analyst expectations. The company posted quarterly revenue of thirty three point seven three billion dollars, up twenty five point five percent year-over-year, and full-year 2025 revenue of one hundred and twenty two point four two billion dollars.
Net income for the fourth quarter reached sixteen point three one billion dollars, a thirty five percent year-over-year increase, with gross margin of sixty two point three percent. Analysts had expected a twenty seven percent increase in net profit to record levels.
Capital Expenditure Surge
In response to overwhelming demand, TSMC plans to spend between fifty two billion and fifty six billion dollars in 2026 on capacity expansion. This figure substantially exceeds Wall Street expectations of forty one billion dollars and marks a significant increase from the forty one billion dollars spent in 2025.The company expects AI accelerator revenue to grow by at least fifty percent annually through 2029. High-Performance Computing now represents fifty eight percent of TSMC's total revenue, underscoring the company's deep dependence on AI-related chip production.
Market Implications
The capacity shortage at TSMC has ripple effects throughout the technology industry. Several high-profile customers, including Nvidia and Broadcom, have requested production increases in recent months as their supply remains capacity-constrained. Industry analysts expect the shortage of AI-capable chips to persist well into next year.Nvidia's dominance of TSMC's advanced packaging capacity gives it priority over competitors in a market where demand far outstrips supply. The limited availability of Nvidia's GPUs and Broadcom's ASICs, both in high demand for AI applications, continues to constrain infrastructure buildouts at major cloud providers and enterprises.
The situation at TSMC confirms the strength and sustainability of the AI boom rather than signaling weakness. When the world's most important chipmaker says capacity needs to triple just to meet existing customer demand, it provides concrete validation of the scale and longevity of AI infrastructure investments underway across the technology sector.
Published January 19, 2026 at 1:09am
