ASIC and optical interconnects drive high-speed growth! Bank of America strongly supports the soaring Marvell (MRVL.US), claiming there's still 55% upside potential

According to Zhitong Finance APP, on September 11, the US announced that the overall and core CPI for August rose by 0.4% and 0.3% month-on-month, respectively. Both increased faster than in July, and the core CPI exceeded expectations, but the S&P 500 and Nasdaq Composite still rose by 0.86% and 0.96% respectively that day. The pullback in oil prices and strong performance from AI-related technology companies continue to provide bullish support, highlighting that the market is still weighing inflation pressures amid geopolitical conflicts and profit growth trends. Wall Street's bullish logic surrounding Marvell Technology (MRVL.US), which focuses on AI ASIC and data center optical interconnect chips, precisely reflects the latter—namely, "the negative logic of the AI computing power theme dominating the profit trajectory."

Wall Street financial giant Bank of America maintains a $365 target price for Marvell, valuing its expansion of revenue share per AI system through custom AI accelerators (i.e., AI ASICs) and supporting optical interconnect chips for massive AI inference demand; compared to the closing price of $236.10 on September 11, this target implies about 55% potential upside. However, the $40–45 billion related revenue opportunity by 2030 is a scenario estimate, with realization depending on client projects, product share, and profitability.

The greater the demand for AI computing power, the more the value of data center AI chips and optical interconnects expands

From an engineering and business model perspective, Marvell benefits from AI expansion mainly through "custom AI accelerated computing + system-matching optical interconnect/optical communication". ASICs designed for specific workloads can improve energy efficiency and reduce large-scale deployment costs by optimizing compute units, memory access, and data flow. Meanwhile, more accelerators urgently need comprehensive supporting network, memory interface chips, and high-speed data center connection chips based on optical technology.

Marvell calls these supporting businesses "XPU supporting chips," including PCIe retimers, coprocessors, CXL controllers, high-speed optical interconnect chips, and key components of co-packaged optics (CPO). According to Bank of America's latest estimates, each custom processor combined with one or two supporting chips priced at $500–$1,500 can bring Marvell multiple revenue increments when clients expand. The complexity of high-speed analog circuits, interface intellectual property, and system verification also sets a high bar for customers to self-replicate these products.

Marvell's second growth curve—its “optical interconnect chip business” apart from working with cloud giants like Amazon to create AI ASICs—relies on the underlying driver of demand for optical interconnects: the effective AI computing power of clusters is increasingly dependent on inter-node communication efficiency. Expert parallelization in mixture-of-experts models, distributed inference, and systems using prefill and decode separation can all increase data exchange; network congestion can cause expensive accelerators to wait for data, reducing overall cluster utilization.

As transmission speed and distance increase, the signal loss, compensation, and power consumption challenges of copper connections intensify, pushing more links to adopt optical connections. Marvell participates through optical communications DSPs, high-speed serial interfaces, drivers, and related interconnect products. The mass-produced Ara platform and extended 1.6T product portfolio exemplify concrete products built for this upgrade.

Lumentum’s demand outlook provides another side of evidence for this industrial chain. Citi meeting notes recorded management's judgment: shipments of related optical module devices (i.e., optical modules for AI accelerator tightly integrated interconnection) are expected to roughly double from 2026 to 2027, with shortages of EML and CW lasers likely persisting through 2027; improved visibility for OCS and NPO businesses pushed the fiscal 2028 EPS target up to $40.

The two companies benefit in different segments: Marvell focuses on semiconductors for processing, transmission, and connecting data, while Lumentum provides laser light sources, optical components, and optical switching systems. Therefore, the growth logic of AI chips and optical interconnects is still sustainable, but continued stock price increases require continued realization in orders, production capacity certification, gross margins, and cash flow.

Bank of America sends a strong bullish signal to Marvell investors

According to Seeking Alpha’s statistics, the AI chip giant’s share price has risen more than 160% in the past six months, with demand for its data center chips fueling market optimism. In a report shared with TheStreet, Bank of America analyst Vivek Arya, after meeting with CEO Matt Murphy and CFO Dan Durn, believes there is ample reason to remain bullish on the stock.

Wall Street’s overall evaluation of Marvell is more bullish, but target prices diverge. As of September 11, StockAnalysis shows that the S&P Global summary of 45 analysts rates it as a “strong buy,” with an average one-year target price of $284.64—about 20.6% above the $236.10 closing price—but the lowest target implies a drop of about 20%. The highest target in the sample is $400 (about 69.4% potential upside); public ratings show KeyBanc analyst John Vinh maintained an “overweight” rating and a $400 target price as of August 28. Bank of America’s $365 target price is also notably above the market average. Wall Street’s bullish consensus is on the expansion of AI data center business, while differences focus on the speed of growth realization and the valuation multiples the market is willing to give.

As of September 11, the stock price was up nearly 12%, while large peers in the semiconductor sector such as NVIDIA (NVDA.US) fell over 5% in the past week. Specifically, Marvell participates in building the hardware that powers AI operations. The company designs custom processors as well as high-speed optical interconnect chips that transfer data between processors, enabling large-scale computing systems to work efficiently in concert.

The company’s expanded partnership with Google is particularly noteworthy, covering chips supporting Google’s AI infrastructure. Moreover, the agreement gives Google the right to purchase Marvell stock at a specific price depending on future procurement.

However, Arya’s latest insights after a recent lunch with management have already moved beyond any single client. Bank of America highlights that Marvell’s continuously expanding product portfolio could further deepen its role in AI systems. With analyst day approaching on October 6, investors may soon have a clearer view of just how large this opportunity could be.

Bank of America maintains a “buy” rating and a high $365 target price for Marvell, suggesting nearly 55% upside from its September 11 close of around $236.10.

The analyst team led by Vivek Arya bases their confidence on Marvell’s potential to sell more technology products per AI system.

Arya and his team believe Marvell’s biggest opportunity lies in supporting chips for connecting processors, managing memory, and transferring data. Compared with processors themselves, it is much harder for customers to replicate this business.

Marvell has already delivered these products to the four major US hyperscale cloud providers, and each custom processor requires one or two supporting chips, priced at $500–$1,500 apiece. This creates new opportunities for Marvell to increase revenue as customers expand computing power.

Bank of America estimates that by 2030, this supporting chip market alone could exceed $60–65 billion. Assuming a 40%–50% market share, Marvell’s annual sales opportunity could reach $30 billion, while management’s projection for 2028 is $3–4+ billion.

In addition, custom processors offer another important growth driver. Bank of America believes Marvell’s potential sales in this area could reach $15 billion by 2030, bringing the combined opportunity to $40–45 billion. However, it’s important to note that these are scenario-based model estimates, not confirmed orders yet.

The profit potential also explains this optimism. BofA analyst Arya believes Marvell’s per-share earnings potential for calendar year 2028 is close to $14, higher than the baseline of $11; for every $1 billion increase in sales, EPS could rise by 30–35 cents.

His target price is based on a 33x multiple of core earnings excluding equity compensation. Therefore, this bullish scenario depends largely on strong growth and investors continuing to award Marvell a premium valuation as it executes its strategy.

For context, according to Seeking Alpha data, the stock’s non-GAAP P/E was 42x over the past five years, so a 33x baseline valuation is not especially high.

What could interrupt Marvell’s rally?

The biggest risk to Marvell’s stock price is the gap between its huge market opportunity and the actual profit realization. BofA’s envisioned $40–45 billion sales scenario depends largely on market expansion, customer expenditure, and substantial stock price appreciation. This revenue still needs to be realized in the future.

Execution of customer projects is arguably the first test. BofA noted uncertainties in next-gen chip projects for Amazon and Microsoft. Delays in product launches may postpone sales, while R&D costs will continue, thus weakening the profit growth investors anticipate.

Competition is also likely to exacerbate this risk. In the AI ASIC sector, Broadcom (AVGO.US) may remain the dominant competitor in the custom chip space, while standardized AI accelerators led by NVIDIA and AMD are also vying for the same budget. Marvell’s broad interconnect portfolio provides multiple growth paths, but winning business doesn’t guarantee pricing power or the margins needed to meet BofA’s profit assumptions.

In addition, the Google agreement deserves scrutiny. The much-discussed $12 billion figure refers to a framework tied to procurement, not guaranteed orders. As Peace Longe pointed out, nearly 59 million shares could be issued under these warrants, which creates potential dilution while presenting commercial opportunity.

Valuation is another pressure point. BofA’s target price is equivalent to about 33 times 2028 calendar-year earnings, noticeably above the 26x median historical multiple cited. Therefore, this lofty target implies investors must continue to give Marvell a premium. Slower growth could simultaneously undermine profit expectations and the multiples applied to them.

Thus, when assessing analyst day on October 6, delivery timelines, margin expectations, and customer project ramp-up progress deserve major attention. After such a substantial stock run, simply raising market size forecasts provides limited confidence support.

At that point, the strongest confirmation may be clearer evidence that sales growth can translate into sustainable actual EPS and free cash flow per share.