Google’s $12.2B Chip Power Play: Option to Buy Marvell Stake
Google and Marvell Team Up: A $12 Billion Bet on Custom AI Chips
What Happened?
Aug 19 (Reuters) — Marvell Technology, a company that designs computer chips, just announced a major partnership with Google. The deal has two big parts:
- Marvell will help Google build its own custom AI chips (called TPUs, or Tensor Processing Units).
- Google gets the option to buy a huge stake in Marvell — up to $12.2 billion worth — making it one of Marvell’s top five investors.
Important Point: This isn’t a guaranteed investment. Google only gets the shares if it keeps buying chips from Marvell over the next decade.
Why Does This Matter?
The AI Chip Gold Rush
- Nvidia currently dominates the AI chip market with its expensive graphics processors (GPUs).
- Big tech companies like Google want cheaper, more efficient alternatives tailored to their specific needs.
- Custom chips (like Google’s TPUs) are better for "inference" — the process of running trained AI models (not training them).
Google’s Strategy
- Google has been building its own chips for years.
- Now it’s bringing in Marvell as a new partner to expand its "TPU ecosystem."
- This puts Marvell in direct competition with Broadcom, which has been Google’s main custom chip partner.
The Deal Broken Down
| Detail | What It Means |
|---|---|
| Warrant | Google gets the right (but not obligation) to buy Marvell shares. |
| Shares | Up to 58.97 million shares. |
| Price | $206.58 per share (set in advance). |
| Total Value | ~$12.18 billion if fully exercised. |
| Condition | Most shares unlock only if Google hits purchasing targets through 2033. |
| Result | Google would become Marvell’s 5th largest shareholder. |
Key Takeaway: Google’s stake grows only as it buys more chips. This aligns both companies’ incentives.
Market Reaction
| Company | Premarket Move | Why? |
|---|---|---|
| Marvell (MRVL) | +11% | Investors love the new revenue + Google’s commitment. |
| Broadcom (AVGO) | -2% | Fear of losing Google business to Marvell. |
The Bigger Picture: Big Tech’s $700 Billion AI Spend
- 2024 forecast: Over $700 billion on AI infrastructure (data centers, chips, networking).
- 2023 actual: ~$400 billion.
- This deal is one piece of a massive wave of investment.
Marvell vs. Broadcom: The Rivalry Heats Up
| Partner | Deal With Google | Status |
|---|---|---|
| Broadcom | Long-term agreement to develop & supply custom AI chips & components for next-gen AI racks. | Active through 2031 |
| Marvell | New broad partnership covering "a range of chips & technologies" for TPU ecosystem. | Just announced |
Both can win. Google needs massive scale and likely wants multiple suppliers for security and innovation.
Summary
- Google & Marvell are deepening ties on custom AI chips.
- Google gets a warrant to buy up to $12.2B of Marvell stock — if it keeps buying chips through 2033.
- Marvell stock surged 11%; Broadcom dipped 2% on competition fears.
- Big Tech is spending $700B+ on AI infra in 2024 — custom chips are a huge part of that.
- This deal reduces reliance on Nvidia and strengthens Google’s TPU roadmap.
FAQ
1. What is a "warrant" in this context?
A warrant gives Google the right to buy Marvell shares at a fixed price ($206.58) within a certain timeframe. It’s like a coupon — Google doesn’t have to use it, but it can if Marvell’s stock price goes higher.
2. What are TPUs and why does Google use them?
TPUs (Tensor Processing Units) are Google’s custom-built chips designed specifically for AI workloads. They’re more efficient than general-purpose GPUs for running Google’s models (like Search, Translate, Gemini).
3. Why would Google want to own part of Marvell?
It aligns incentives. If Google owns Marvell stock, it benefits when Marvell does well — and Marvell is motivated to prioritize Google’s needs. Plus, it’s a potential financial win if Marvell’s stock rises.
4. Is Broadcom losing Google as a customer?
Not entirely. Broadcom still has a long-term deal through 2031. But Google is clearly diversifying — adding Marvell as a second major partner for custom silicon.
5. How does this affect Nvidia?
It’s a long-term threat. The more big tech builds its own chips, the less they need to buy Nvidia’s expensive GPUs. But Nvidia still leads in training and general-purpose AI — for now.
Source: Reuters reporting by Anhata Rooprai; Editing by Shilpi Majumdar