Everyone wants to learn from Nvidia's "chip-for-financing" model; Google and AMD are both supporting "AI cloud."

Everyone wants to learn from Nvidia's "chip-for-financing" model; Google and AMD are both supporting "AI cloud."

Nvidia's "compute-finance" closed loop created by supporting CoreWeave is so alluring that Google and AMD are forced to "cross the river by feeling the stones", trying to use real money to build their own AI chip ecosystem.

On February 20th, according to the Wall Street Journal, citing sources familiar with the matter, Google is exploring ways to leverage its ample financial resources to build a broader AI ecosystem. The core strategy is to provide funding to data center partners on the condition that they must use Google's TPU chips. Meanwhile, Wallstreetcn wrote, AMD is aggressively promoting its AI chips by guaranteeing loans for customers.

This strategy is seen by the market as a direct replication of Nvidia's "CoreWeave model"—that is, by supporting "Neocloud" vendors, bypassing traditional cloud giants already occupied by Nvidia or with self-developed chips, to build their own "trusted troops".

Emulating Nvidia's Path: Investing in "New Cloud" and Mining Companies

Sources disclosed that Google is negotiating to invest about $100 million in cloud computing startup Fluidstack. This deal values Fluidstack at around $7.5 billion.

This is not only a financial investment, but also a strategic bond. Companies like Fluidstack, these "emerging clouds," are important variables in today's AI compute market, specializing in providing compute services for AI companies. Previously, Nvidia did the same by supporting CoreWeave, enabling it to accumulate large amounts of GPUs and break into the market.

Now Google wants to "copy the formula." According to sources, Google hopes to amplify Fluidstack's growth potential through funding and encourage more compute providers to use its AI chips.

In addition to direct investment, Google has also reached into former crypto mining companies that are undergoing transformation.

The report points out that Google has already provided backstopped financing for projects involving companies like Hut 8, Cipher Mining, and TeraWulf. These companies possess ready data center infrastructure and are eager to transform into AI compute factories, and Google's funding support will secure their adoption of TPUs.

AMD's Aggressive "Bet": If It Doesn't Sell, I'll Rent

Compared to Google's direct investments, AMD's approach is even more aggressive and risk-prone.

Reports say AMD will provide substantial guarantees for a $300 million loan to data center startup Crusoe. This Goldman Sachs-funded loan will be used to purchase AMD's AI chips.

The most notable is the “backstop clause.” Sources reveal that if Crusoe cannot find customers to use these chips, AMD has agreed to rent these chips from Crusoe. This makes AMD the "last lessee," alleviating the demand-side concern for customers.

While this approach may boost sales in the short term, it raises huge risk exposure for chip manufacturers if AI demand slows. Should the market cool, AMD's balance sheet will take a direct hit.

Why Take a Detour?

The reason Google and AMD choose this winding path is that the main road is blocked.

For Google, although star startups like Anthropic are expanding use of TPUs, traditional cloud competitors (such as Amazon AWS, Microsoft Azure) show little interest in TPUs.

Industry insiders point out: "Major cloud service providers seem lukewarm, partly because they view Google as a competitor." Additionally, Amazon AWS is actively developing its own AI chips.

Therefore, supporting neutral third-party "emerging cloud" firms has become the best choice for Google and AMD to break the blockade.

Internal Game and Capacity Bottlenecks

To accelerate commercialization of TPUs, Google has even considered organizational restructuring internally.

Sources say some Google Cloud managers have recently revisited a longstanding internal debate: "Whether to reorganize the TPU team as an independent department." Such a split might allow Google to bring in external capital and expand investment opportunities.

However, Google officially denies this proposal. A Google representative stated explicitly that "there is no plan to reorganize the TPU department," and emphasized that "maintaining close integration between the chip team and other company units (such as the Gemini model development team) is advantageous."

Beyond organizational issues, the more practical obstacle is capacity.

Although Alphabet now co-designs TPUs with Broadcom and outsources manufacturing to TSMC, with global AI demand surging, TSMC's advanced capacity is stretched thin.

Sources say "TSMC may prioritize its largest customer Nvidia over Google". In addition, the worldwide shortage of essential HBM memory chips for AI chips is a hard barrier for Google to ramp up TPU shipments.

Risk Warning and Disclaimer ClauseThe market carries risk, investment requires caution. This article does not constitute personal investment advice, nor does it take into account individual users' specific investment objectives, financial status, or needs. Users should consider whether any opinions, views, or conclusions herein suit their particular circumstances. Investment based on this is at your own risk.