As the AI boom reshapes the global semiconductor supply chain, Apple's long-standing super-VIP status as Taiwan Semiconductor Manufacturing's (TSM) "top client" is facing unprecedented challenges. With soaring demand for advanced packaging capacity from chip giants like NVIDIA (NVDA), Apple is not only confronting its most significant foundry price hikes in years but is also forced into fierce competition within a capacity allocation system it no longer dominates.

On January 15, Culpium, a news platform long focused on the Taiwanese semiconductor industry, reported, citing informed sources, that beyond price increases, a more severe reality is emerging. As GPUs from clients like NVIDIA and AMD occupy increasingly larger areas on wafers, the chip designs from this iPhone maker can no longer automatically secure priority scheduling across TSMC's nearly twenty fabs as they once did.

This shift is underscored by a notable financial divergence. According to analysis by Culpium and supply chain information, NVIDIA likely replaced Apple as TSMC's largest revenue source for at least one or two quarters last year. On an annual basis, even if it hasn't completely overtaken Apple by 2025, this shift is almost certain to occur by 2026.

This also highlights TSMC's increasingly strong pricing power and bargaining leverage. As mentioned in a Wall Street News article, TSMC's latest financial report shows its gross margin has climbed to a staggering 62.3%, approaching levels typical of software companies. Analysis indicates that while Apple still provides indispensable order volume and stability, NVIDIA is becoming the new core driver for cutting-edge process nodes that pursue ultimate performance and profitability.

The AI boom is reshaping the capacity allocation landscape. Although TSMC's Chief Financial Officer, Wendell Huang, stated they "do not discuss specific customer rankings," public data clearly outlines this trend.

According to Culpium estimates, TSMC's revenue grew 36% last year to $122 billion, while NVIDIA's sales are projected to surge 62% in the fiscal year ending January 2026. In contrast, Apple's product revenue (excluding services) is forecast to grow only 3.6% over the 12 months ending December 2025.

In fact, Apple's role as the primary engine of TSMC's revenue growth ended as early as five years ago. Back in 2018, were it not for Apple's incremental purchases, TSMC's sales might have even declined that year. Today, with smartphone revenue growth slowing to 11%, the High-Performance Computing (HPC) segment, which includes AI chips, has become TSMC's new growth engine, with the segment's revenue surging 48% last year, continuing the 58% high growth from the previous year.

TSMC stated on Thursday that revenue is expected to grow nearly 30% in 2026, but capital expenditure is projected to increase by about 32%, reaching a record $52 to $56 billion. Long-term, growth is expected to average 25% over the five years through 2029, but the AI business is forecast to grow by an average of 55% or more during the same period. This projection is higher than the previously estimated 40%-plus figure.

Notably, the ultimate strength showcased in TSMC's latest earnings report is reflected not only in record revenue and net profit but also in gross margins nearing those of software makers and fabless chip designers. The figure reached an impressive 62.3% in the December quarter, 280 basis points higher than the previous quarter. Were it not for overseas fabs (Arizona and Japan), the gross margin would have been even higher.

As the AI frenzy continues, with GPUs from clients like NVIDIA and AMD occupying larger areas per wafer, Apple's chip designs can no longer guarantee a spot in TSMC's nearly twenty fabs.

The evolution of the technology roadmap appears to favor players like NVIDIA and AMD in the short term, meaning Apple will likely still need to fight for capacity over the next year or two.

While TSMC's most advanced 2-nanometer (N2) process is already in mass production with Apple as a primary buyer, the N2P variant scheduled for mass production in the second half of this year and the new A16 node will focus more on high-performance computing demands.

According to TSMC CEO C.C. Wei, the A16 node utilizes "Super Power Rail" technology—a backside power delivery network—which is most suitable for HPC chips with complex signal routing. This technology is also a key battleground where Intel is attempting to capture foundry share from TSMC.

Although the A14 node, expected to enter mass production in 2028, is projected to rebalance support for mobile and HPC, for now, NVIDIA, with its large and singular demand for advanced capacity, is clearly a better fit for TSMC's current expansion cadence than Apple with its more diversified product lines.

Despite external criticism from analysts like Benedict Evans, who argue that TSMC's expansion is not fast enough to meet all of NVIDIA's orders, CEO C.C. Wei remained highly cautious during TSMC's recent earnings call, as mentioned in a Wall Street News article.

He candidly admitted to being "very nervous" during the investor conference, emphasizing that failure to proceed carefully could lead to a "huge disaster." Analysis suggests this caution stems from TSMC's unique business model: unlike asset-light or integrated hardware/software giants like Google or NVIDIA, TSMC bears massive depreciation risks.

Data shows that depreciation accounts for a high 45% of TSMC's cost of revenue, more than four times that of Google (10%) and nearly eight times that of fabless NVIDIA (5.7%).

As a chip designer, NVIDIA enjoys gross margins exceeding 70%, with its primary risk lying in inventory; whereas TSMC, facing an industry downturn, would be burdened by the heavy financial weight of newly built fabs.

As C.C. Wei stated, even with massive investments this year, the revenue contribution from that spending is almost zero within the same year, yet equipment depreciation is immediately booked on the accounts.

Therefore, despite the rising influence of Jensen Huang, which in some dimensions even surpasses that of Apple CEO Tim Cook, the "stability" offered by Apple remains crucial for TSMC.

Analysis suggests that while NVIDIA's products are highly sought-after, they are relatively "niche" and concentrated; whereas Apple's product lines span more than a dozen TSMC fabs, a breadth of manufacturing footprint that is difficult for a single AI chip vendor to match. When the AI bubble eventually cools and market growth flattens, the value of clients like Apple will become prominent once again.