While Apple is expected to buy chips from TSMC’s upcoming fab in Arizona to reduce its reliance on manufacturing in Asia, it seems unlikely the company will ever truly rely on Taiwanese chipmaking itself. As such, our colleagues at Bloomberg believe that the reported decision, which has yet to be officially announced, is more of a marketing gimmick rather than a decision based on necessity.
But while Apple has yet to confirm plans to use TSMC’s fab in Arizona, which is due to come online in 2024, let us analyze what Apple needs two years from now and TSMC in a couple of powerful company is going to offer. years. What you have to keep in mind is that we’ve known about Apple’s alleged intention to use the fab in Arizona since Bloomberg reported. We believe the information is accurate as Bloomberg is a highly respected source, but we cannot verify this.
apple needs
TSMC’s fab in Arizona will produce chips using process technology from the company’s N5 (5nm class) family, which includes the N5, N5P, N4, N4P and N4X. Apple currently uses the names N5, N5P, and N4 for various system-on-chips used for smartphones, PCs, and consumer electronics. Meanwhile, from reports, we know that Apple, being TSMC’s alpha customer for state-of-the-art process technologies, is the first company to use TSMC’s N3 (3nm class) node for at least one of its SoCs. .
But designing chips for a leading-edge process technology (we’re looking at ~$600 million just for the physical implementation and embedded software) and producing them on a leading-edge node is expensive. Apple uses its N4-based A16 Bionic SoC only for its premium iPhone 14 Pro, which sells from $999. Meanwhile, the more mainstream vanilla iPhone 14 (from $799) still uses the N5P-based A15 Bionic SoC from 2021.
Perhaps a more colorful example would be Nvidia’s launch of its 4N (a customized version of the N4 with significant performance enhancements) graphics processor based on the Ada Lovelace architecture. The AD102 and AD103 GPUs are used exclusively for the company’s flagship GeForce RTX 4090 and RTX 4080 graphics boards, which carry $1,599 and $1,199 price tags, respectively (and are the best graphics cards available today, so to speak). ). In contrast, AMD’s range-topping Radeon RX 7900 XT and RX 7900 XTX (based on the Navi 31 GPU) use TSMC’s N5 technology and will retail from December 13 for $999 and $899, respectively.
But even though the new process is expensive, Apple needs leading production nodes not only for flagship smartphones, but also for its PCs. The company’s M1-based Macs broke some performance records and clearly exceeded expectations. Meanwhile, Apple’s M1 Max with its 57 billion transistors – one of the most complex chips ever designed – is implemented on TSMC’s N5P process which shows that the company’s ambitions for performance are, to put it mildly, are high for
TSMC Capabilities
While we think we know some vectors of Apple’s SoC development, we need to analyze them in the context of TSMC’s plans and capabilities. And they may offer some additional food for thought.
Back in October, TSMC cut its 2022 CapEx by $4 billion for the second time this year, citing slower demand for chips as well as availability of wafer fab equipment (WFE) in the medium-term future. The reduced CapEx won’t have an immediate impact on the company’s business, but lower spending and longer runtimes for fab tools could have an impact on TSMC over the next few years. In fact, WFE availability appears to be a major problem for the semiconductor industry in general.
TSMC’s move to reduce its CapEx from $40 billion to $36 billion was a bit surprising as the company’s revenue and profits hit all-time records in the third quarter of 2022. YoY growth and 11.4% quarter-on-quarter growth. The company’s net income increased to $8.717 billion in Q3 2022 from $4.849 billion in Q3 2021.
TSMC’s main profitability driver these days is the aggressive adoption of its advanced manufacturing technologies, the N7 and N5 (5 nm and 7 nm-class, respectively), by its customers, and especially Apple – TSMC’s biggest customer. TSMC’s N5 family of production nodes became the largest revenue contributor in the third quarter as it accounted for 28% of the company’s revenue (up from 18% in Q3 2021), which isn’t particularly surprising as the company has launched several high-end devices. production is produced. Profile products from high-profile clients such as Apple (M2, A16), AMD (Zen 4 CCDs), Nvidia (H100, AD102, AD103, AD104), and Qualcomm (Snapdragon 8+ Gen 1). Meanwhile, N7 manufacturing technologies (and Apple still uses it to man its A16 Bionic SoCs) accounted for 26% of TSMC’s earnings (down from 34% in Q3 2021). As it turns out, 54% of the world’s No. 1 foundry’s revenue was contributed by its advanced nodes (up from 52% in Q3 2021).
Nonetheless, TSMC cut its capital expenditure from $40 billion to $36 billion, citing the mid-term outlook and WFE availability, but not citing US sanctions against the Chinese supercomputer and semiconductor industries.
In fact, demand for PCs and smartphones is falling and some say the bottom is yet to be reached. Meanwhile, high-performance computing products (i.e., a vague term that TSMC uses for PCs, HPC, and other performance-hungry applications) now account for 39% of TSMC’s revenue, showing slow but steady growth. represents. Meanwhile, as CPU sales for client PCs are falling, HPC’s contribution to TSMC’s revenue is also expected to decline, according to analysts at China Renaissance, which could push the foundry’s N7 capacity utilization to around 70% in Q4 2022 and 1H 2023. Will reduce by %.
Slow Capacity Expansion and How It Affects the Mac
Analysts believe TSMC’s CapEx cuts will slow capacity expansion for TSMC’s N6/N7 production nodes, one of the company’s most important revenue contributors today. Meanwhile, as the foundry will have additional N7-capable capacities in the coming quarters, it will redeploy some of these production lines to make advanced chips on its N4/N5 nodes.
“TSMC will redeploy some N6/N7 idle capacity to N4/N5 and slow Fab 22 expansion (built for N7/N28),” wrote analyst Szeho Ng in a note to clients.
While the company did not say anything about slowing capacity expansion for the N5 family of manufacturing processes, it said that it expects demand for the N6/N7 to increase in the second half of 2023, when the utilization rate of the N7-capable product lines will increase. increases.
Meanwhile, it’s unclear how the CapEx cut will affect the company’s ramp up of its N3 (3 nm-class) family of manufacturing technologies. And this is where it concerns Apple: we have no idea what it produces on the N3. Apple may use the N3 for its PC-bound processor. Those Macs may not end up well in popular things like the Mac Mini, MBA, or MBP 13 Pro, but the expensive Mac Studio and MacBook 14/16 are machines that can absorb the cost. Yet the thing is, no one knows what Apple is planning to do. All we know is that China Renaissance is based on comments from a financial analyst firm that is poised to be accurate, but it is not an official source.
“We expect 2023 CapEx to be flat,” wrote the analyst. “We expect the focus of 2023 CapEx to be ready for future N4/N5 demand waves with Qualcomm/Nvidia planning moves. [Samsung Foundry], The pace of TSMC’s N3 buildout plans is more muted, given the initial focused customer and cost concerns behind the N5/N7 expansion pace in similar phases of ramp-up.
summary
While we don’t know exactly what Apple is planning with its future chips, we do know for certain that the Arizona fab may not be large enough for its ambitious Apple silicon plans. But will the company need chips produced there? Absolutely! Will it be a marketing gimmick? We can only wonder at this point.
