GPUs Outshine CPUs: AI Transforms the Chip Industry

The rapid evolution of technology has brought generative artificial intelligence (AI) to the forefront, significantly influencing various markets, particularly the graphics processing unit (GPU) marketAccording to a recent report by Yole, a market research firm, the GPU market is projected to explode, reaching an astounding $190 billion by 2029, effectively doubling the market size of CPUsThis remarkable transformation predominantly arises from the surging demand for generative AI applications, illustrating the paradigm shift within the tech landscape.

As we delve into the report's findings, it becomes evident that the processor market, which includes CPUs and GPUs, is on the brink of substantial growthForecasts suggest that by 2029, the total processor market size could balloon to $480 billion, with GPUs and artificial intelligence application-specific integrated circuits (ASICs) spearheading this expansionCurrent market dynamics reveal that while APUs, CPUs, and GPUs sequentially lead the segment, GPUs are set to surpass other classifications, primarily because they are becoming indispensable for AI-driven workloadsMoreover, the forecast points to a striking growth trajectory for data center AI ASICs, propelled by the increasing focus on specialized chips for AI tasks tailored for cloud hyperscalers.

Interestingly, the telecommunications and infrastructure sectors are predicted to take a decisive lead in market performance by 2024, overtaking mobile and consumer industriesThis shift is largely attributed to significant investments directed toward AI infrastructureConcurrently, the automotive sector is emerging as a notable growth area, propelled by the industry's shift toward AI and advanced processing technologiesIn stark contrast, demand for smartphones and personal computers is expected to remain weak over the next five years, resulting in modest growth, with an estimated compound annual growth rate (CAGR) of around 5% in the mobile and consumer market segments.

Globally, Nvidia stands as the dominant player in the GPU market, with Intel experiencing fierce competition in the CPU space

Advertisements

The escalating demand for AI accelerators has positioned Nvidia to lead the generative AI revolution with its high-performance GPUs, solidifying its market positionThe processor market, fundamentally driven by CPUs and GPUs, remains under the firm grip of giants like Intel and AMD in the CPU domain, while Nvidia retains near-absolute control over the GPU sector, collectively commanding a staggering 97% of their respective markets and approximately 47% of the overall processor market.

While Intel continues to hold its ground as the CPU market leader, it is facing increasing pressure from startup competitors like Ampere Computing, along with tech behemoths like Amazon and Google, resulting in a shrinking market shareConversely, Qualcomm and MediaTek dominate the APU segment, with Apple leading in internal designsIn the laptop sector, competition is intensifying against the once-duopolistic dominance of Intel and AMDThe automotive market is also experiencing heightened competition, as numerous companies vie for opportunities within this rapidly evolving landscape.

In the realm of field-programmable gate arrays (FPGAs), Intel and AMD lead the charge, particularly within the systems-on-chip (SoC) FPGA domain, while AI ASICs are predominantly driven by data center original equipment manufacturers (OEMs) such as Google, Amazon, and Alibaba, alongside Intel's Gaudi product lineThe data processing units (DPU) market is likewise shaped by leading data center processors and hyperscale cloud providers.

In China, the government's prioritization of AI and RISC-V has fostered a competitive atmosphere for processorsOverall, the AI processor market is experiencing a rapid ascent, signaling a transformative phase within the industry.

The advent of generative AI technologies since 2023 has catalyzed processor innovation and optimized AI applications in data centers and consumer devices alikeAI personal computers are revolutionizing the laptop and desktop markets, with Apple's M series and Qualcomm's Snapdragon X-Elite/Plus leading the revolution

Advertisements

Concurrently, AMD and Intel have entered the arena with their latest 4nm and 3nm processors oriented towards AI personal computing.

Within the smartphone segment, the integration of AI has elevated performance levels, especially highlighted by MediaTek's Dimensity 9300+. Nvidia's Blackwell GPUs utilize advanced chip architecture to enhance data center capabilities, while Arm-based CPUs, such as Nvidia's Grace, are gaining traction in the server domain, especially amid increasing demands for advanced driver-assistance systems (ADAS) in the automotive sector.

Startups and hyperscalers are focusing on AI ASIC inference to challenge Nvidia's supremacy, leveraging smaller chips and high-bandwidth memory (HBM) for improved efficiencyNotably, the 4nm process is becoming a standard, while 3nm remains reserved for specific applicationsCurrently, TSMC has positioned itself as the frontrunner in advanced process technology, while Samsung faces challenges in yield productionThe widespread adoption of smaller chip architectures across the industry enables optimization for various applications.

Amid declining revenues, Intel is recalibrating its strategy to enhance the competitiveness of its foundry business, planning to harness TSMC's 3nm processes to manufacture its Lunar Lake processorsBy 2026, Intel aims to shift its focus toward in-house manufacturing for its Panther Lake processors.

Furthermore, Yole projects that advanced packaging revenues are expected to reach $10.9 billion in the second quarter of 2024, reflecting a 5% increase from the previous quarter, driven predominantly by AI investmentsHistorically, seasonal factors have influenced back-end businesses, suggesting that the first half of the year might yield the weakest performanceStill, with signs of gradual recovery in demand, revenues are anticipated to grow by 15.5% in the third quarter of 2024, reaching $12.6 billionDespite enduring soft demand and needing further inventory digestion, 2024 is expected to be a year of recovery, with robust performance anticipated in the latter half of the year.

In terms of capital expenditures, the second quarter of 2024 saw a slight increase compared to the previous quarter

Advertisements

Leading players in the advanced packaging sector reported around $10.3 billion in capital expenditures in 2023, a 19% decline from the prior year, yet projections indicate a 7% growth in 2024. New analyses within the monitoring framework have introduced two distinct scenarios regarding the adoption of fan-out panel-level packaging (FOPLP). The most optimistic perspective anticipates significant transitions in CoWoS and InFO production moving toward panels by TSMC starting in 2027, further enhancing UHD FO towards PLP shifts.

Looking ahead, Yole predicts that advanced packaging revenues will continue to grow at a robust CAGR of 12.7%, propelling the entire market size from $39 billion in 2023 to $80 billion by 2029. However, the entirety of the semiconductor sector has faced struggles in 2023, impacting the advanced packaging marketUltimately, as demand rises and advanced packaging adoption escalates, the sector is set to rebound in 2024.

Diving deeper into the advanced packaging market drivers, Yole identifies mobile and consumer, telecommunications and infrastructure, and automotive markets as key contributors, heavily influenced by trends such as high-performance computing (HPC) and generative AIAmong the various packaging platforms, 2.5D/3D packaging is expected to experience the fastest growth over the next five yearsIndustry leaders, including TSMC, Intel, and Samsung, alongside top outsourced semiconductor assembly and test (OSAT) companies such as ASE, Amkor, and JCET, are poised to invest significantly, with an estimated $10.7 billion earmarked for advanced packaging initiatives in 2024.

The pervasive influence of AI chips continues to expand, marking AI as the dominant driving force within the chip industryChristophe Fouquet, CEO of ASML, elucidated this reality during a recent earnings call, candidly stating, “Without AI, the semiconductor industry would suffer tremendously.” The profound impact of AI is evident not only in ASML's and TSMC's financials but also in the broader semiconductor ecosystem.

Historically, the semiconductor industry has served as a barometer for the global economy, as chips are critical components across a range of products, from data center servers to household appliances like dishwashers

Companies providing the machinery necessary to manufacture these chips are on the industry’s frontlineBefore semiconductor companies can begin production, they must navigate a lengthy process of constructing, installing, and testing the equipment involved in chip manufacturingTherefore, firms like ASML maintain a unique, long-term perspective concerning customer sentiment, particularly in the context of the AI revolution.

Nonetheless, there are warning signs in other fields beyond AIAs inventory levels rise, demand is declining among automotive and industrial suppliersIntel is taking measures to reduce costs while postponing plans for new factories to address declining revenues and increasing lossesFurthermore, Samsung Electronics issued an apology to investors this month regarding disappointing financial results attributed to delays in high-bandwidth memory productionInvestors are now closely monitoring Texas Instruments, which is set to announce earnings due to its widespread analog chip usage across various customers.

Overall, the future landscape for equipment manufacturers appears grim, with numerous companies’ stock prices reaching historic highs earlier this yearSome traders have already begun offloading shares, reacting swiftly to uncertain market conditionsFor instance, ASML encountered its worst week since early September, witnessing a 14% decline in its US-listed sharesOther prominent players, such as Applied Materials, experienced a 9.1% drop, while KLA and Lam Research fell by more than 12% each.

In tandem with these trends, Israeli chip companies also faced declines on the Tel Aviv Stock Exchange, mirroring the decreasing fortunes in U.S. marketsNova led the downturn with a significant drop of approximately 14.8%, erasing roughly $3.6 billion in market value, leaving it with a valuation around $5.4 billionTower Semiconductor fell by 3.6%, losing around $720 million, while Camtek also saw a decrease of 3.5%, resulting in a valuation of approximately $1.4 billion.

Following this trend of caution, Cantor Fitzgerald analyst C.J

Muse expressed a similar sentiment in a recent report, stating, “We have always maintained a cautious stance regarding other semiconductor equipment companiesHowever, we initially believed that companies like ASML, with longer delivery times, would perform exceptionally wellClearly, that assumption is proving erroneous.”

As the costs associated with AI chips remain high, the landscape of competition in chip development becomes narrowerMarvell's Executive Vice President and Chief Technology Officer, Noam Mizrahi, conveyed in a recent Nikkei interview that maintaining competitiveness requires substantial funding, leading only a handful of top chip developers to sustain investments in semiconductor technology designed for AI computation.

Mizrahi further emphasized, “Even if there are limited participants in this game, the market for customized AI chips is set to surge dramatically.” Insights from Mizrahi reflect a vision that the market is vast and open for those capable of making long-term investments. “Investment in packaging technology and 5nm, 3nm, or 2nm chip fabrication processes is criticalThere is a need for investments across the board, including intellectual property, interface technology, and memory technologyTherefore, I believe this field will not become overly crowded,” he asserts.

According to the International Business Strategies (IBS), the cost of manufacturing a single wafer for 7nm chips reaches nearly $10,000, surpassing $14,000 for 5nm wafers, and nearing $20,000 for 2nm productionChip developers allocate hundreds of millions of dollars to each generation of chips to sustain relevance in the marketplace.

Despite Nvidia and AMD's consumer-grade offerings remaining popular, industry titans like Google, Microsoft, Amazon, and Meta are investing heavily in developing their own semiconductors for data centers, aiming to optimize performance and achieve product differentiationThese tech giants are not only building their dedicated chip design teams but also seeking partners to facilitate the development and production of these chips.

Consequently, Marvell forecasts an explosive 45% compound annual growth rate for the custom computing chip market between 2023 and 2028, ballooning from a value of $6.6 billion to a staggering $42.9 billion

Advertisements

Advertisements

Your email address will not be published.Required fields are marked *