Google is negotiating with semiconductor company Marvell Technology regarding custom Tensor Processing Unit (TPU) development and exploring the possibility of developing dedicated chips optimized for large language model inference.
This move, combined with Marvell's earlier $2 billion strategic partnership agreement with NVIDIA, further clarifies Marvell's positioning in the AI data center custom chip market.
According to FundaAI, Google is actively discussing TPU development projects with Marvell, which will participate as a design services provider; discussions also extend to a dedicated chip specifically optimized for large language model (LLM) inference workloads.
These negotiations come just days before Google renews its long-term TPU and networking agreement with Broadcom until 2031, demonstrating Google's proactive strategy to diversify custom chip suppliers while leveraging Marvell's technical expertise in high-speed interconnects to optimize costs and performance.
<h2>Google Negotiations: Strategic Layout Behind Supply Diversification</h2>
FundaAI reports that Google's discussions with Marvell involve two levels: first, TPU custom development where Marvell participates as a design services provider; second, a dedicated chip optimized for LLM inference scenarios, known externally as the Language Processing Unit (LPU) architecture.
This development occurs shortly after Google extended its cooperation agreement with Broadcom until 2031.
For Google, introducing multiple custom silicon suppliers amid rapidly expanding AI infrastructure investments is a standard practice to diversify technology risks and maintain bargaining leverage. Marvell's accumulation in high-speed interconnect technology precisely fills the gap in Google's needs for specific workload optimization.
For Marvell, even though it remains in the negotiation phase, the potential significance cannot be overlooked. The company's custom ASIC business currently generates approximately $1.5 billion in annual revenue; once the Google contract is finalized, it will add another high-value revenue stream on top of existing design orders.
<h2>NVIDIA's $2 Billion Investment Brings Marvell to Core of AI Ecosystem</h2>
The backdrop to Google's negotiations is a key agreement announced by Marvell last month—NVIDIA's $2 billion strategic investment and the establishment of deep collaboration through the NVLink Fusion framework.
Under the agreement, Marvell will design custom XPU and scaling networks compatible with NVLink, directly embedded within NVIDIA's rack-level AI architecture to operate alongside GPUs, Vera CPUs, network interface cards, and switches. Both parties will also collaborate on silicon photonics technology for optical interconnects.
NVIDIA characterized this cooperation as an expansion of the AI ecosystem, stating that it maintains its core position in interconnect architecture while providing customers with greater flexibility. For Marvell, the $2 billion cash injection directly strengthened its balance sheet and accelerated its business expansion pace in AI communications and AI-RAN markets.

GOOG

AVGO

GOGL

Google is in negotiations with Marvell regarding the development of custom TPU and large model inference chips, just days before renewing its cooperation agreement with Broadcom, clearly signaling an intent to diversify its supply chain. Last month, NVIDIA made a $2 billion strategic investment in Marvell and established deep collaboration through the NVLink Fusion framework, bringing Marvell into the core of the AI ecosystem

- Google is negotiating with semiconductor company Marvell Technology to develop custom Tensor Processing Units (TPUs) and specialized chips for large model inference.  
- This discussion follows Marvell's recent $2 billion strategic agreement with NVIDIA, enhancing its position in the AI data center custom chip market, while Google aims to diversify its chip suppliers to mitigate technical risks.  
- Marvell’s ongoing negotiations could significantly boost its annual custom ASIC revenue, which currently stands at approximately $1.5 billion, by adding valuable contracts from Google.

Wallstreetcn

Alphabet - C

Broadcom

Golden Ocean

GraniteShares 2x Long NVDA Daily ETF

Spdr Select Tech

Direxion Semicon Bull 3X

YieldMax NVDA Option Income Strategy ETF

Global X Cloud Computing ETF

iShares Expanded Tech Software Sector ETF

First Trust Nasdaq Food & Semicon

Direxion Daily AVGO Bull 2X Shares

T-Rex 2X Long NVIDIA Daily Target ETF

ISHRS S&P Glb It

GraniteShares 2x Long MRVL Daily ETF

First Trust Index NextG ETF

Leverage Shares 2X Long AVGO Daily ETF

Defiance Daily Target 2X Long AVGO ETF

Direxion Daily NVDA Bull 2X Shares

Invesco Semiconductors ETF

Direxion Daily GOOGL Bull 2X Shares

GraniteShares 2x Long AVGO Daily ETF

Invesco PHLX Semiconductor ETF

Global X Data Center & Dgtl Infrs ETF

Alphabet

Marvell Tech

NVIDIA

NVIDIA Corporation

Google Advances Chip Supplier Diversification, Enters Negotiations with Marvell for Custom TPU and Inference Chip Development