This year, the Spring Festival has seen an extraordinary surge in the adoption of artificial intelligence (AI).

Masayoshi Son, known as a leading venture capitalist in Asia and often humorously dubbed a “bald con artist,” has recently made headlines with stunning newsBloomberg cited insider sources claiming the founder of SoftBank Group is aiming to raise a monumental $100 billion to establish a joint venture that would manufacture semiconductors to compete with industry giant Nvidia (NVDA-US), which is critical for powering AI technology

This past Spring Festival, Son’s fortunes appeared to be at a peak

Advertisements

Benefiting from Nvidia’s soaring success and its own impressive financial performance, Arm—a company in which SoftBank owns roughly 90% of the shares—saw its market value double in just three trading days, while SoftBank’s stock price skyrocketed nearly 20%. It’s safe to say that Son has turned a corner in his career

Moreover, the conversation around AI reached fever pitch during the Spring FestivalAround 2 AM Beijing time on February 16, American company OpenAI officially launched its first text-to-video generation model named SoraThe model was an instant hit, and some industry leaders have suggested that the timeline for achieving Artificial General Intelligence (AGI) has now shrunk from ten years to just one yearSurprisingly, even before trading began on the A-shares, the speculative interest in AI-related stocks had already surged

Advertisements

Let’s take a deeper look at Masayoshi Son's latest endeavors

According to insider information cited by Bloomberg, Son is striving to secure up to $100 billion in funding to create a chip joint venture aimed at directly challenging Nvidia in the semiconductor market crucial for AI technologies

This ambitious project is reportedly named "Izanagi." Son envisions a complementary chip company to Arm that would emerge as a formidable player in the AI chip landscape

An insider mentioned that SoftBank might contribute $30 billion, with the remaining $70 billion expected to come from institutional investors based in the Middle East

Advertisements

If successful, this initiative could represent one of the largest investments in AI infrastructure since the advent of ChatGPT

However, the specifics about how the funds will be secured and allocated remain uncertainIt has been noted that Son has been exploring various investment ideas and strategies to expand Arm's influence in the AI marketplace while experimenting with different types of next-generation chips

In response to these reports, both SoftBank and Arm have refrained from making any official comments

Currently, Son is placing significant focus on Arm

• U.S. Interest Rates Cut Again
• Avita Secures Over 11 Billion in Funding!
• Pinduoduo's Staggering Loss of Over 450 Billion!
• South Korean Stock Market Suffers Major Crash
• Euro Experiences Continuous Decline

Insiders believe that he sees an opportunity to create a tech giant that rivals the likes of the seven largest technology corporations globally

As of December 31, 2023, SoftBank capitalized on a global stock market rebound, holding approximately $62 trillion (around $410 billion) in cash and equivalents, alongside nearly $8 billion worth of T-Mobile (TMUS-US) shares and a 90% stake in ARM-US, resulting in a significant boost for the companyArm's market value alone increased by about $50 billion during the Spring Festival

While discussions continue between Son and OpenAI’s CEO Sam Altman concerning collaborations and financing for chip production, insiders reveal stark contrasts between the nature of the "Izanagi" project and Altman's ambitions

It has been reported that Son had previously sought to invest in another company developing foundational AI models and requested their assistance in his chip ventures, but the proposal was declinedSoftBank is consistently exploring the application methods of Arm's chip design technology as well

The AI frenzy is surging once more

Since October 2022, as the global AI wave has swept over the industry, Nvidia’s stock price has skyrocketed by over 500%. Firms like Goldman Sachs, Bank of America, and Morgan Stanley have all raised their target prices for Nvidia, with Goldman predicting that investments from national governments and tech giants into AI infrastructure will further enhance Nvidia's revenue

They have adjusted their earnings projections upwards by an average of 22% for Nvidia covering 2025 to 2026.

In this rapidly evolving context, Nvidia has overtaken Amazon and Google, becoming the third-largest company by market capitalization in the US stock marketUnder Nvidia's aegis, other AI chip companies, including Son’s Arm, have also become highly sought after

Furthermore, recent reports from Bloomberg revealed that the U.Sgovernment is currently negotiating to provide over $10 billion in subsidies to IntelThe discussions involve loan arrangements and direct grants, continuing from previous smaller allocations announced by the U.S

Department of CommerceCommerce Secretary Gina Raimondo stated earlier in the month that the department plans to distribute some funds from a $39 billion pool designated for the semiconductor manufacturing industry within the next two months

Additionally, on February 17, it was reported that OpenAI has successfully completed a transaction, raising its valuation to $80 billion—a nearly twofold increase in value in less than ten months

According to Guosheng Securities, the increase in the supply of top-tier global technology, alongside OpenAI's evolving business model and ecosystem, is igniting a trend toward global AI industrialization

The launch of Google's Gemini and the pressing competition in advanced technologies mark a significant turning pointWith OpenAI's GPTs ushering in a new era of light applications, the eco-development is thriving both on the B-to-B and B-to-C fronts, fostering a wealth of multimodal business models, while China excels in AI vision sectorsThe robust performance of American AI technology giants indicates that AI business has entered a preliminary commercialization stage, with a promising outlook for realization in 2024.

Unveiling Sora

In the early hours of February 16, around 2 AM Beijing time, OpenAI officially introduced Sora, its first text-to-video generation model

Sora can create movie-like realistic scenes using brief or detailed prompts or even a single static image, capable of producing up to one minute of 1080p video with multiple characters, various actions, and detailed backgroundsOpenAI has emphasized that “Sora serves as the foundation for a model capable of understanding and simulating the real world, and this function is believed to be a significant milestone toward achieving AGI.”

Following its launch, Sora quickly gained popularityOn February 16, Zhou Hongyi, founder of 360, took to social media to express his views on Sora, suggesting that its arrival indicates AGI could be realized within a mere one to two years

Huajin Securities noted that compared to previous video generation models, Sora innovates significantly in three main areas regarding its underlying model and algorithms:

1. Adoption of a diffusion model based on the Transformer architecture: Unlike models like Runway Gen1 and Stable VideoDiffusion which traditionally rely on classic U-Net architecture, Sora replaces U-Net with a Transformer architecture, greatly enhancing the model’s scalability

2. Training video data at original sizes: Unlike other video models that often resize or crop videos to standard dimensions, like compressing to a 256×256 resolution over 4 seconds, Sora trains on data in its original size

This approach allows for more flexible sampling from widescreen 1920×1080p videos, vertical 1080×1920 videos, and everything in between, thus enabling Sora to create content with original aspect ratios for various devicesAdditionally, this improves composition and framing, as a model trained on cropped footage sometimes results in partial visibility of subjectsWith Sora, the video framing sees significant enhancement

3. Generating detailed text descriptions for training videos: Unlike other text-to-video models that rely on large datasets, Sora enhances training effectiveness through the integration of DALL·E 3 and GPT research findings to create or supplement detailed subtitle descriptions for video training data, improving accuracy in predictions