Marvell vs. Broadcom: Same AI Stack, 20x Valuation Disparity

December 9th, 2025 by Trefis Team
MRVL: Marvell Technology logo
MRVL
Marvell Technology

The investment narrative around the Artificial Intelligence boom thus far has largely centered on who supplies the foundational hardware.  In this high-stakes contest, two semiconductor giants—Broadcom (AVGO) and Marvell Technology (MRVL)—are fierce competitors in the most critical areas of the AI infrastructure build-out: logic chips and networking.  On paper, they compete in the same markets. In the stock market, they live in different orbits Since the launch of ChatGPT in November 2022:

Image by Cristian Ibarra from Pixabay

Today, Broadcom commands a $1.8 trillion market cap. Marvell sits near $80 billion. The multiples and scale explain part of the story.

  • Broadcom: 41x forward 2026 earnings, Marvell: 24x forward earnings

  • Broadcom Revenue (FY): >$63B, Marvell Revenue (FY): about $8B

Where The Hyperscalers Real Money Is Flowing

Hyperscalers are spending tens of billions on custom silicon, moving beyond general-purpose GPUs to better optimize their costs and reduce reliance on Nvidia, which effectively dominates the AI compute space.

Two layers matter most, namely Custom AI compute (ASICs) and High-speed networking. Both Broadcom and Marvell Technology play here, but Broadcom does it with far bigger customer scale, deeper hyperscaler reach, and broader product coverage. The revenue and margin gap explains the power dynamic.

  • Broadcom’s chip business brought in $9.2B last quarter, 26% YoY growth, and 67% EBITDA margins (boosted by software).
  • Marvell: about $2B in total revenue, 37% YoY growth, but only 15% operating margins.

Marvell is growing faster. Broadcom is growing at scale.

Why Broadcom’s Scale Is Structural

This scale advantage is structural for Broadcom. Why?

  • It is one of the largest buyers of advanced wafer capacity and has pricing leverage Marvell simply does not have.
  • Has preferred access at TSMC via long-term volume deals.
  • Scale at this level turns into permanent margin power.

Custom AI Chips (ASICs): Similar Tech, Different Tier

Both design high quality custom AI chips, but Broadcom owns the top end.
Both build specialized AI ASICs. Technically, Marvell is very strong. But Broadcom is the clear market leader. Broadcom’s customer list is the real moat.

  • Broadcom has diversified wins at Google with its Tensor Processing Units, which are slowly posing a threat to Nvidia. Besides this, other big players like Meta and China’s ByteDance are also customers.
  • Separately, The Information reported that Microsoft is exploring Broadcom as a potential future chip design partner.
  • That translates higher volumes and deep integration into multiple AI roadmaps at once.
  • Marvell’s custom silicon story is tightly linked to Amazon Web Services-more exposed to a single customer’s capex cycles and design choices.

On Monday, Benchmark analyst Cody Acree moved Marvell to Hold, citing that Amazon was likely to shift future Trainium 3 and 4 designs to Alchip. Neither Amazon nor Marvell has confirmed this. For now, it remains speculation, but it highlights the inherent volatility of customer-concentrated custom silicon revenue.

High-Speed Networking: Control vs Participation

Both Broadcom and Marvell Technology supply the networking silicon that binds tens of thousands of accelerators into a single AI training fabric.

  • Broadcom’s Tomahawk and Jericho platforms are the de-facto standard for hyperscale Ethernet switching. Marvell competes in Ethernet physical layer devices, DSPs, and optical interconnects.
  • That said, Broadcom has architectural control over how AI clusters are physically built, not just revenue scale.

Broadcom’s Software Upside

The central reason for the valuation disparity is Broadcom’s massive, high-margin Infrastructure Software division, anchored by its recent acquisition of VMware. Post-deal, Broadcom eliminated perpetual licenses, forced a subscription-only model, and pushed through product consolidation and price increases.  The result?

Exceptionally high software margins and 67% adjusted EBITDA margins at the company level and roughly 40% operating margins.  That predictable software cash flow acts like a financial floor, letting the market award Broadcom a premium multiple even in weak chip cycles. The software is becoming more relevant to AI as well. Broadcom is adapting VMware to build “AI factories” using GPU virtualization and workload orchestration.

Marvell Technology sits on the opposite end of the spectrum. Marvell is a pure-play semiconductor company, with revenue tied almost entirely to hardware shipments. That means what in practice? Lower structural margins and full exposure to semiconductor cyclicality, even during strong growth phases.

Marvell’s Blueprint to Close the Gap

To justify a valuation multiple closer to its rival, Marvell must execute an aggressive strategy centered on growth, technical differentiation, and financial discipline.

1. Owning the Optical Future

For interconnections, copper is hitting hard physical limits on power dissipation and speed. The future of AI interconnects is light, not metal. Marvell is betting on Co-Packaged Optics (CPO) and photonic fabric as the next AI bottleneck.

  • The acquisition of Celestial AI is a direct attempt to lock in tech and talent for photonics.  Why Marvell Wants To Buy Celestial AI
  • There’s a possibility that this could be a potential platform-defining shift in AI infrastructure. If it works, Marvell could leapfrog incumbents.

2. Break the AWS Dependence

Marvell’s custom ASIC story is tightly tied to Amazon Web Services. A major customer = high revenue concentration risk.

  • To re-rate, Marvell needs another Tier-1 hyperscaler win.
  • A second major design win would validate its competitiveness, smooth revenue volatility and immediately change how the market prices its risk. Even if recent Trainium reports prove unfounded, the episode reinforces why Marvell ultimately needs a second Tier-1 hyperscaler win to stabilize long-term growth confidence.

3. Win on Trust, Not Just Technology

  • Broadcom’s post-acquisition pricing actions at VMware serve as a quiet reminder to hyperscalers of vendor concentration risk.
  • That dynamic creates an opening for Marvell to position itself as a strategic counterweight to Broadcom, not just on performance, but on long-term partnership credibility.

4. Prove It Can Scale Profitably

Growth alone will not close the valuation gap. Margins must also follow. Marvell’s reported operating margins for the first nine months of the year were roughly 15% compared to 40% for Broadcom.

Marvell has to keep scaling high-margin data center revenue enforcing strict cost discipline. The market is waiting for visible operating leverage gains, and a narrowing of the profitability gap vs Broadcom. Without margin expansion, the re-rating simply does not happen.

The asset allocation framework of Trefis’ Boston-based, wealth management partner yielded positive returns during the 2008-09 period when the S&P lost more than 40%. Our partner’s strategy now includes Trefis High Quality Portfolio, which has a track record of comfortably outperforming its benchmark that includes all three – the S&P 500, S&P mid-cap, and Russell 2000 indices.

 

Invest with Trefis Market-Beating Portfolios
See all Trefis Price Estimates