Underappreciated AI chip stocks are drawing new attention from investors as dealmaking accelerates across the artificial intelligence landscape. According to Bloomberg Intelligence Senior Semiconductor Analyst Kunjan Sobhani, companies such as Marvell Technology (MRVL) and Broadcom (AVGO) may be positioned for significant upside as the next wave of AI infrastructure demand develops. While Nvidia continues to dominate the industry narrative, Sobhani argues that investors may be overlooking two critical beneficiaries of expanding hyperscaler activity.
Key Points
Speaking in a recent interview, Sobhani emphasized that the second half of 2026 could mark a major turning point for AI chip demand. Recent partnerships — including Google’s deal with Anthropic and AWS’s multiyear commitments — reinforce a broader shift in how hyperscalers are approaching custom silicon. As these companies begin supplying their in-house AI accelerators to external customers, competition with Nvidia is intensifying, opening the door for new winners.
Broadcom and Marvell, he said, are particularly well-positioned. Their valuations remain attractive relative to future growth expectations, and the market has not fully priced in their potential contribution to next-generation AI data center hardware.
Growing Deal Momentum Highlights Emerging Winners
The AI industry is experiencing one of its fastest deal cycles to date, with hyperscalers racing to secure compute capacity and diversify chip supply. Sobhani points to recent transactions as a signal of a larger trend: companies once focused solely on internal AI development are now commercializing their custom chips.
This shift could relieve pressure on Nvidia’s supply constraints while creating new business channels for chip designers working behind the scenes. Broadcom and Marvell fall squarely into that category. Both companies manufacture custom and semi-custom silicon used in networking, accelerators, and system-level architectures essential to large-scale AI training and inference.
According to Sobhani, investors may be underestimating the magnitude of this shift. Despite strong fundamentals, these two names trade at valuations that do not yet reflect long-term AI infrastructure demand.
What the Market Has Priced In — And What It Hasn’t
While enthusiasm for AI-linked equities remains high, Sobhani noted that today’s valuations across mega-cap chipmakers appear reasonable relative to expected growth. The market has already priced in strong performance through 2026 and part of the first half of 2027. The next phase, he said, depends heavily on execution.
The biggest risks lie not in demand, but in supply. With new AI server racks coming from AMD and custom ASIC programs rolling out from multiple vendors, the question becomes whether manufacturers can scale without disruption. Data center construction may also face limitations related to power availability and infrastructure readiness.
If supply ramps smoothly, several companies could exceed expectations in 2026. But even minor delays in large-scale deployments could spark short-term volatility, given the high expectations investors now carry into each quarterly earnings cycle.
Is the AI Market in a Bubble? Analysts See Something Else
In response to growing conversations about an AI valuation bubble, Sobhani referenced a recent Bank of America analysis characterizing the current environment as more of an “air pocket” — a temporary cooling period supported by earnings strength.
He agrees with that view. Interest rate volatility, year-end market positioning, and exceptional performance across tech stocks create natural pauses. But fundamentals remain intact. As long as companies deliver on projected output for 2026 and early 2027, Sobhani sees limited downside risk.
However, one dynamic he warns investors to watch closely is the “beat-and-raise” trend that has defined AI earnings in recent quarters. With expectations now elevated, even slight timing shifts in chip production or deployment schedules could lead to notable market reaction.
A Key Inflection Approaches as Hyperscalers Scale Outward
The most influential change in the AI chip industry, Sobhani argues, may be the commercialization of hyperscaler-designed silicon. Companies like Google and Amazon have historically designed accelerators primarily for internal workloads. Over the next two years, they are increasingly expected to sell these chips externally, fueling competition in a market long dominated by Nvidia.
This shift benefits companies like Broadcom and Marvell, which supply foundational components that power these next-generation AI chips and systems. As more hyperscalers open their architectures, demand for supporting silicon — networking hardware, custom accelerators, controllers, and connectivity systems — could grow rapidly.
That is where underappreciated AI chip stocks come into focus.
Sobhani highlights:
- Marvell (MRVL): A leader in data center networking, custom accelerators, and high-speed connectivity required for AI clusters.
- Broadcom (AVGO): A key supplier of ASICs and networking technologies that underpin hyperscale AI training environments.
With the AI build-out still in its early innings, both companies stand to benefit from diversification across cloud providers and the rise of open AI chip ecosystems.
Conclusion
The race to expand AI infrastructure is accelerating, and the market is evolving beyond a single dominant chip supplier. As hyperscalers commercialize their internal designs and pursue external partnerships, opportunities are emerging for companies that supply the essential hardware beneath the surface.
For analysts like Kunjan Sobhani, the next major catalyst will arrive in late 2026, when demand inflection, supply expansion, and broader architectural shifts converge. Amid this landscape, underappreciated AI chip stocks such as Marvell and Broadcom may be positioned for outsized gains if execution and infrastructure readiness remain on track.
As the AI cycle matures, the story is becoming less about one company — and more about the ecosystem that powers the next era of intelligence.
