Friday, August 28, 2026
DarkSubscribe
AI Infrastructure · News & Analysis
HomeChips & HardwareReport
Chips & Hardware · Report

Broadcom projects its artificial intelligence revenue will reach one hundred sixteen billion dollars by two thousand twenty-seven, with Samsung benefiting from custom silicon and networking demand.

The forecast underscores explosive growth in custom application-specific integrated circuits and Tomahawk interconnect deployments, validating the shift toward disaggregated artificial intelligence infrastructure.
Trade pressSlicast · August 24, 2026 · US · Source: Google News
importance 86

Broadcom’s expansion into artificial intelligence semiconductors is projected to sustain its rapid growth trajectory through fiscal 2027. As major technology companies including Google and Meta accelerate their proprietary AI chip development, Broadcom’s AI-related revenue is expected to reach $116 billion (approximately 160.5 trillion won) by that year. In a preview report released on the 19th ahead of Broadcom’s upcoming earnings announcement, global investment bank Citi forecasted this milestone. Broadcom is scheduled to release its third-quarter results for fiscal 2026, covering May through August, on the 2nd of next month.

Citi maintains a “Buy” rating on Broadcom with a $500 price target, noting that investor attention is likely to shift toward the company’s 2027 AI business outlook rather than near-term quarterly details. The bank also projects Broadcom’s total revenue for fiscal 2027 will reach $166.5 billion (approximately 230.4 trillion won), meaning roughly $7 of every $10 in total revenue would originate from the AI segment. This growth is largely driven by application-specific integrated circuits (ASICs) tailored for Google and Meta. Citi forecasts Broadcom’s AI ASIC revenue share will rise to 44% in the October quarter and exceed 55% by the second half of 2028. Within the broader $116 billion AI revenue projection, Google and large language model (LLM) companies alone are expected to contribute $90.3 billion (approximately 125 trillion won).

Despite strong demand, execution risks remain centered on data center availability and power infrastructure. Delays in facility construction, power procurement, and regulatory approvals could push some AI investments into 2028, creating a potential mismatch between chip supply and deployment readiness.

Broadcom’s AI expansion is closely tied to a strategic partnership with South Korea’s Samsung Electronics (005930.KS). Last month, the companies signed a memorandum of understanding to explore collaboration opportunities exceeding $200 billion (approximately 276.8 trillion won) over five years through 2030, primarily in memory and foundry services. The partnership covers high-bandwidth memory (HBM), sub-2-nanometer foundry processes, and advanced packaging technologies. Citi expects HBM capacity per chip to increase alongside Broadcom’s growing ASIC complexity, meaning higher AI chip sales will directly correlate with greater HBM content per unit. Following the announcement, Reuters reported that securing Broadcom’s production volumes could improve utilization rates at Samsung’s advanced semiconductor facilities, strengthening its foundry operations and helping narrow the competitive gap with industry leader TSMC.

Optimism also extends to the memory sector. According to Barron’s, JPMorgan stated that “Samsung will retain its status as a pivotal HBM supplier for Broadcom.” The bank further projects that over the next five years, 90–95% of Broadcom’s semiconductor purchases from Samsung will be memory-based, with only 5–10% dedicated to foundry wafers. Samsung’s foundry business has shown recent signs of recovery; Reuters reported that Samsung raised prices for new orders on select advanced nodes, including 4nm and 5nm, by up to 15% amid rising AI semiconductor demand. This pricing shift indicates that as TSMC’s capacity tightens, customers are increasingly routing orders to competitors like Samsung.

Meanwhile, the competitive landscape is evolving. On the 19th, Marvell Technology (MRVL) announced an expanded partnership with Google to develop custom chips for the company’s AI frameworks. Under the agreement, Marvell will issue warrants for up to 58.97 million shares at $206.58 per share, valued at approximately $12.2 billion (around 16.9 trillion won). Notably, the warrants do not create an immediate financial obligation for Google. Per Marvell’s SEC filing, only about 1.4 million shares will be allocated in the first year, with the remainder contingent on revenue performance. Each tranche unlocks for every $500 million (approximately 690 billion won) in custom chip revenue generated by Google from fiscal Q3 2027 through fiscal 2033. Fully vesting the equity stake would require Google to purchase roughly $120 billion (approximately 166.1 trillion won) in Marvell products over that period.

Historically, Google has relied heavily on Broadcom for the design of its Tensor Processing Unit (TPU). This new arrangement establishes Marvell as a second major supplier. Marvell has previously developed custom silicon for Amazon (AMZN) and Microsoft (MSFT), positioning it across all three major U.S. cloud providers. Following the announcement, Marvell shares rose approximately 10% to close near $234, while Broadcom dipped roughly 4–5%. However, Broadcom holds a long-term contract with Google extending through 2031, indicating Marvell’s role is additive rather than substitutive. Morningstar analyst William Kerwin described the deal as a significant win for Marvell, emphasizing that Google is diversifying its supplier base rather than abandoning Broadcom.

As Broadcom’s AI business continues to scale, Samsung stands to benefit substantially from both HBM and foundry opportunities. Big Tech’s AI infrastructure investment is estimated to exceed $700 billion (approximately 968.7 trillion won) this year, driving higher HBM requirements per chip. Nevertheless, Marvell’s expanded role with Google may present a longer-term challenge to Broadcom’s dominance in the AI ASIC market. While Broadcom remains central to Google’s TPU architecture, competition is expected to intensify as Marvell expands into inference accelerators, storage controllers, networking hardware, and memory technologies.

Investors will closely monitor Broadcom’s earnings release on the 2nd and Marvell’s quarterly results on August 27. Analysts note that commentary from Marvell’s management regarding Google’s actual demand will likely carry more weight than the headline figures surrounding the warrants.

Read the original
Broadcom projects its artificial intelligence… · Slicast