Google Corporation has obtained the right to acquire up to 58.97 million shares of chip manufacturer Marvell Technology. This is a key element of the expanded partnership in the development of processors for artificial intelligence, disclosed in an 8-K form report filed with the U.S. Securities and Exchange Commission.
This move is a vivid illustration of how the financial architecture of the AI industry is becoming increasingly complex and intertwined. We are witnessing a transition from simple "seller-buyer" transactions to multi-layered agreements involving guarantees, stock options, and joint capital raising for the construction of giant computing infrastructure.
An option tied to purchases
The collaboration covers a wide range of components for TPU infrastructure — Google's proprietary AI accelerators. This includes not only inference processors but also storage controllers, networking solutions, memory interfaces, and near-memory computing technologies. Such a comprehensive approach indicates a deep integration of Marvell into Google's technology stack.
The mechanics of the option are extremely interesting. Only about 1.36 million shares will be unlocked in equal portions during the first year. The rest of the warrant is strictly tied to the volume of Google's orders: the shares are divided into 240 tranches, each of which is activated after purchases bring Marvell an additional $500 million in revenue. Full unlocking of the option will require cumulative sales of $120 billion — this demonstrates the scale of the parties' long-term commitments and plans.
Google will be able to exercise the option at a price of $206.58 per share until August 18, 2033. In the event of full exercise of the warrant, the technology giant could become the fifth-largest shareholder of Marvell. The market reacted positively to this news: Marvell shares rose, while shares of the larger competitor Broadcom lost more than 5% of their value.
Analysts, particularly William Kerwin of Morningstar, call this "a big win for Marvell." However, I would not view the deal as a direct displacement of Broadcom from Google's supply chain. Rather, we are seeing a strategic expansion of the supplier pool amid the explosive growth of the hyperscaler's AI capacity needs. This is sensible risk diversification.
Reducing dependence on Nvidia
This agreement is part of a broader trend. The largest technology corporations are actively seeking to reduce the cost of AI computing and lessen their dependence on Nvidia's dominant accelerators. Google has been developing its own TPUs, optimized for training and inference of models, for years. Marvell, in turn, plays a key role in creating the components needed to combine thousands of chips into unified computing systems.
Notably, this is happening against the backdrop of an internal restructuring of Google's AI division, where the influence of leaders associated with Google Cloud is growing. For Marvell, the agreement is not just a guaranteed major customer. The expenditure-linked warrant makes Google economically interested in the growth of the supplier's own market capitalization, creating a unique synergy of interests.
Such schemes, combining purchases with equity participation, are becoming the new norm. Recall the recent agreement between AMD and OpenAI, where the ChatGPT developer also received the opportunity to become a significant shareholder of the chip manufacturer.
Financing one's own demand and new risks
An even more illustrative example is the deal between Nvidia and OpenAI. Jensen Huang's company agreed to provide a guarantee of up to $105 billion for OpenAI's data center in Ohio, as well as invest $1.5 billion directly in SB Energy. This creates an unusual chain: Nvidia financially supports the construction of infrastructure that will then become a major buyer of its own accelerators. Huang rejects accusations of "circular financing," but such structures inevitably intensify investor questions about the interdependence of key market players.
Against this backdrop, Wall Street is actively joining the financing of AI infrastructure. Total AI spending already exceeds $730 billion per year, and Nvidia, together with Apollo, BlackRock, and Goldman Sachs, is creating platforms to attract more than $500 billion in third-party capital. Private funds are beginning to view computing infrastructure as a separate asset class, providing financing secured by equipment.
AI has finally transformed from a "software" industry into a capital-intensive one. As Professor Amit Joshi rightly notes, competitive advantage increasingly depends less on the model itself and more on the ability to finance, build, and operate infrastructure more cheaply. However, these new financial schemes, despite their effectiveness in accelerating construction, carry systemic risks. The Bank for International Settlements has already warned of macro-financial dangers associated with the growth of opaque debt financing and the potential rehypothecation of the same assets.
My view: we are witnessing a fundamental shift. Technology giants are turning into financial engineers, and chip manufacturers into hostages of their capital expenditures. This interdependence may accelerate development, but it creates a fragile system where problems of one player can instantly spread across the entire ecosystem.