Marvell FY31 Targets, AMD Supply Crunch, EmbeddingGemma 2: Oct. 6

Three separate chip and AI stories broke Tuesday with meaningful implications for investors: Marvell put numbers on its long-term ambitions, AMD’s CEO flagged a tightening supply picture for 2027, and Google DeepMind shipped a new on-device embedding model that quietly expands Alphabet’s AI product surface.

Key points

• Marvell laid out FY31 revenue targets at its investor day, projecting a multi-year growth path that carries a significant valuation premium relative to current earnings.
• AMD CEO Lisa Su said AI processor demand is running ahead of available supply and AMD is preparing a larger supply push for 2027.
• Google DeepMind released EmbeddingGemma 2, a model built for on-device search and retrieval tasks.
• Nvidia-backed Reflection AI debuted Beam, an open-weight model for coding, reasoning and agentic tasks, positioned as a US answer to Chinese open-weight leaders.
• Franklin Templeton argues the AI investment cycle is rotating away from chip makers and hyperscalers toward companies using AI to lift margins and productivity.

What did Marvell promise investors for FY31?

Marvell’s investor day on October 6 was the headline event for chip investors today. The company projected revenues in a range that would represent substantial growth from current levels by fiscal year 2031, driven primarily by custom AI silicon and data center connectivity. The targets are ambitious, and the GuruFocus analysis accompanying the presentation flagged that the stock already trades at a significant valuation premium, meaning execution risk is baked in from day one. For investors, the key question is how much of that FY31 upside is already priced into MRVL today. Custom ASIC demand from hyperscalers is the core growth driver Marvell is betting on, a theme that has also underpinned Broadcom’s recent bullish guidance. Whether both can win large enough slices of that market to justify their multiples remains the central tension in the custom silicon trade.

Is AMD’s supply crunch an opportunity or a warning sign?

Lisa Su’s public comments Tuesday carried a dual message. On one hand, demand for AMD’s AI processors remains ahead of what the company can currently ship, which is a strong demand signal. On the other, Su indicated AMD is gearing up for a larger supply push in 2027, suggesting meaningful capacity relief is still over a year away. A separate report from GuruFocus noted Su projected semiconductor demand will remain elevated on the back of continued AI growth. For investors, persistent supply constraints cut two ways: they support pricing power in the near term, but they also mean AMD cannot fully capture revenue it could otherwise book. The 2027 supply ramp is worth watching closely as a potential inflection point for AMD’s market share against Nvidia.

What is EmbeddingGemma 2 and why does it matter for Alphabet?

Google DeepMind’s EmbeddingGemma 2 is designed to run search and retrieval tasks directly on a device rather than routing queries to the cloud. That distinction matters commercially. On-device inference reduces latency, cuts cloud compute costs for Alphabet, and positions Google more competitively against Apple’s own on-device AI push. For investors in GOOGL, the model is less a revenue event on its own and more a platform move: embedding strong retrieval capability into devices deepens Google’s search moat at the hardware level, which becomes increasingly important as AI-native interfaces chip away at traditional browser-based search.

Franklin Templeton’s rotation thesis: who benefits next?

The broader investment framing worth flagging comes from Franklin Templeton portfolio manager Grant Bowers, who argued Tuesday that the next leg of the AI cycle will be driven by companies deploying AI to improve productivity, margins and capital efficiency, rather than by chip makers and hyperscalers alone. The note also referenced rising US interest rates as a factor complicating valuations for high-multiple AI names. This is a view that has been gaining traction among institutional managers, and it has direct implications for how to position across the AI stack. Pure chip plays and cloud infrastructure names have led the cycle so far. If Bowers is right, the next wave of outperformance may come from software companies and industrial adopters that can show AI-driven earnings improvement. Meanwhile, Nvidia-backed Reflection AI’s Beam model is a reminder that the open-weight model race is intensifying, with early testing suggesting Beam could be among the more token-efficient models in its class. Nvidia’s backing is notable: it keeps the company embedded in the software layer of AI even as hardware competition grows.

Nothing in this article is investment advice. This site is independent and not affiliated with any company or broker mentioned.