07/01 2024 526
Declaration: The featured image is sourced from the internet.
This is an original article from Jingzhe Research Institute. If you wish to reprint it, please leave a message for permission.
Despite the heavy sanctions from the US, NVIDIA successfully broke through the ban and sold AI GPUs to five Middle Eastern countries.
Recently, NVIDIA reached a cooperation agreement with Qatar's telecommunications giant Ooredoo. Ooredoo will now offer NVIDIA's AI and HPC GPUs in its data centers located in Qatar, Algeria, Tunisia, Oman, Kuwait, and Maldives. This marks NVIDIA's massive expansion in the Middle East.
Not long ago, the US restricted the export of advanced AI and HPC chips from companies like NVIDIA, AMD, and Intel to the Middle East, to prevent China from acquiring AI chips from the Middle East for training its large models. It is unclear which AI chips Ooredoo will provide from NVIDIA and how it will prevent Chinese entities from accessing these GPUs in the cloud.
What is certain is that this market that NVIDIA is reluctant to give up is becoming a force to be reckoned with in the global AI field and plays an important role in the AI competition between China and the US.
Middle Eastern capital begins to "take the table"
In early June, Zhipu AI, a "Tsinghua-affiliated" big model company, secured a $400 million financing round, with the investor being Prosperity7, a venture capital fund under Saudi Aramco, a Middle Eastern oil giant. This is already Zhipu AI's third financing round this year.
According to Qichacha, Zhipu AI has previously completed 8 rounds of financing, with investors including renowned institutions and companies such as Qiming Venture Partners, Legend Capital, Meituan, TAL Education, Xiaomi, Sequoia Capital, Tencent, Alibaba, Hillhouse Capital, and Shunwei Capital. As of now, Zhipu AI's valuation has reached $3 billion, on par with another hot big model unicorn, Moon's Dark Side.
In addition, MiniMax, led by Yan Junjie, an alumnus of the Chinese Academy of Sciences, has completed 4 rounds of financing, with Tencent, Alibaba, Sequoia Capital, and Hillhouse Capital as backers, with a valuation exceeding $2.5 billion.
While Chinese big model unicorns are breaking the $20 billion valuation mark, leading US AI startups are aiming for even higher goals: Anthropic, OpenAI's top competitor, has undergone 7 rounds of financing, with a valuation of $18.4 billion; Musk's xAI recently completed a $6 billion Series B financing round, with a post-investment valuation of $24 billion; according to analysts at Morgan Stanley, OpenAI is expected to hit a $100 billion valuation by the end of this year.
Faced with these increasingly robust unicorns, capital that once aimed to capture them is now showing signs of difficulty. Jingzhe Research Institute found that among the financiers of big model companies after the B round, besides top VCs like Sequoia Capital and Hillhouse Capital, almost only major internet companies like Tencent, Alibaba, and Meituan remain.
Although the popularity of big models has increased in public opinion, historical financing data shows a stepped decline in the heat in the artificial intelligence field: In 2023, the number of financing transactions in the global AI field was about 2,500, the lowest since 2017; the total financing amount was approximately $42.5 billion, a year-on-year decrease of 10%. Among them, the number of investments and financings in China's AI field decreased by 38% year-on-year, and the total financing amount decreased by 70%.
On the one hand, overly high valuations have caused venture capitalists to exclaim, "We can't afford it"; on the other hand, immature valuation systems and uncertainties in commercialization have also deterred institutions.
At this time, wealthy Middle Eastern capital has entered the stage.
It is reported that among the major financiers of xAI's huge financing round, besides Musk's old acquaintances, there are also two Middle Eastern capital figures: Vy Capital, a "mysterious" investment company in Dubai, and Kingdom Holding, a Saudi royal holding company.
Earlier this year, OpenAI CEO Sam Altman proposed a $7 trillion chip business plan aimed at achieving OpenAI's self-sufficiency in high-performance chips. To this end, Sam Altman has negotiated financing with the chairman of IHC, a UAE international holding company.
And the wealth of these tycoons is enviable - Prosperity7 manages a fund size of $3 billion and has invested in over 40 companies globally; Vy Capital has about $1 billion in disposable funds and once led a $205 million financing round for Musk's brain-computer interface company Neuralink; Kingdom Holding has a controlling asset of up to $19 billion and once invested in JD.com with a capital of $2.5 billion, holding shares in well-known companies such as PepsiCo, Amazon, and Disney; IHC is the second-largest listed company in the Gulf region after Saudi Aramco, with an investment portfolio spanning the globe and a market value far exceeding investment companies like Blackstone and Goldman Sachs.
Not content with just being a moneybag
In addition to funding leading big model companies, the Middle East is also actively developing its local AI industry. Among them, Saudi Arabia and the UAE are the two countries that are most obsessed with artificial intelligence.
Since 2016, the decision-making layers of the two governments have begun to focus on AI and have increased policy support during the recent generative AI wave.
For example, the UAE established the world's first national-level AI department in 2017 and also launched a "Generative AI Guide" to help the UAE establish a first-mover advantage in AI in the Middle East.
In 2019, the UAE also established the world's first artificial intelligence university, MBZUAI, which rose to 18th place in the global university AI department rankings in just four years.
*Image source: MBZUAI official website
In 2023, the UAE established the AIATC (Artificial Intelligence and Advanced Technologies Council) and the technology investment company MGX to promote the deployment of cutting-edge technologies. In the same year, Dubai established an AI hub to promote the widespread application of AI in government departments. A year later, the "AI Park" located in the Dubai International Financial Center was officially launched, expected to become the largest gathering place for AI and technology companies in the Middle East and North Africa.
At the same time, Saudi Arabia and the UAE have also joined the AI arms race, purchasing thousands of high-performance NVIDIA AI chips, which would cost a total of $240 million based on the then-market price of H100. The two countries have contacted Western AI startups, hoping to exchange computing power resources for code and expertise in large language models.
The Middle East has also achieved impressive results in the output of big models.
In February last year, the UAE's national research institution TII released the Falcon 40B big model, whose performance surpassed Meta's LLaMA, making it one of the "best models in the open-source world" at the time; the subsequently launched Falcon 180B matched the performance of Google's PaLM 2 Large, topping the Hugging Face rankings in the open-source community; the recently revived Falcon 2 11B surpassed Llama 3 8B again in performance, comparable to the performance of the current leading open-source big model, Gemma 7B.
The UAE AI giant G42 and MBZUAI University jointly launched the open-source large language model Jais, which performs comparably to the English commercial model ChatGPT on Arabic language tasks.
In March this year, Saudi Aramco released the Metabrain AI, an AI industrial big model with 250 billion parameters. The model was trained using 7 trillion data points accumulated by Saudi Aramco over 90 years and can analyze drilling plans, geological data, historical drilling times, and costs to recommend the most ideal drilling solutions. This is just the beginning, as Saudi Aramco plans to develop a trillion-parameter AI model this year, racing towards the goal of "oil industry GPT-4".
*Image source: Saudi Aramco official website
In terms of big models, which are money-consuming beasts, ample financial resources, resources, and computing power are the core advantages of the Middle East. For example, Saudi Arabia plans to establish a national AI fund of up to $40 billion, making it the world's largest AI investor; the UAE has the 15th most powerful supercomputer in the world, capable of performing 2 billion floating-point operations per second; the UAE has also retained AI talents from around the world through measures such as golden visas.
At the same time, funding from governments and royal families has also allowed countries like the UAE to avoid being hampered by commercialization issues in research on "technology preceding application" such as big models. The government's open and forward-thinking approach has also paved the way for research institutions to have a flexible organizational structure, enabling them to allocate and utilize resources more efficiently.
Overall, the big models in the Middle East have not yet revealed a very clear commercial direction.
In terms of general big models, the big models in the Middle East strive to maintain a leading position in the open-source field; in terms of vertical big models, the current application scenarios are concentrated in the oil industry and are expected to be promoted to countries rich in oil and natural gas resources such as the US, Canada, China, and Africa. IDC predicts that by 2025, the global oil and gas engineering software market will be close to $12 billion.
Why is the oil treasure trove targeting AI?
The Middle East's eagerness for artificial intelligence is closely related to its economic transformation needs. As the pillar industry in the Middle East, the price of oil is highly linked to the wealth of the Middle Eastern tycoons. According to previous data from the Institute of International Finance, every $10 increase in the price of a barrel of oil adds $40 billion annually to Saudi Arabia's earnings.
Walking on one leg has brought deep anxiety to Middle Eastern countries.
As early as the end of the 20th century, Dubai planned to transform its economy from an oil-based economy to a knowledge-based and service-oriented economy.
In 2016, Saudi Arabia launched the "Vision 2030," aimed at diversifying the country's economy and reducing its reliance on oil, with AI being a significant branch. More than two-thirds of the vision's goals are related to AI.
In 2019, the UAE established a national strategy to build a global AI hub, planning to make the digital economy account for 20% of the UAE's non-oil GDP by 2031.
In recent years, international oil prices have fallen significantly, hitting a low since 2020. To maintain oil prices, Saudi Arabia has significantly reduced production, but the simultaneous decline in crude oil export prices and quantities has significantly impacted Saudi Aramco's performance. Against this backdrop, the demand for economic transformation in countries like Saudi Arabia is more urgent than ever, and AI is the lifeline they have grasped.
A PwC report predicts that by 2030, AI will account for nearly 14% of the UAE's GDP and 12.4% of Saudi Arabia's GDP. AI will contribute $135 billion to Saudi Arabia's economy, making it the largest beneficiary of this technology in the Middle East.
Years of ahead-of-the-curve布局have made the Middle East a key player influencing global AI development.
Previously, G42's cooperation with Chinese companies caused concern in the US, fearing that its "advanced technology" would be leaked to China by G42. Ultimately, the US government negotiated behind the scenes to make G42 withdraw its investment from China and switch to US technology. After restricting chip trade with China, the US also worried that China would purchase from the Middle East, so it slowed down the chip export licenses of companies like NVIDIA to the Middle East.
Due to a lack of technological differentiation, price wars in the domestic big model field have been fierce. From late May to early June, multiple big model manufacturers, including Zhipu AI, announced price reductions for their entire line of big model products.
Since September last year, Zhipu AI has continuously invested its financing in the research and development of base big models to support a rich industrial ecosystem. At the same time, Zhipu AI has also invested in the multimodal big model company Shengshu Technology and is stepping up its layout in Sora-like text-to-video technology.
This recent financing from oil tycoons may allow Zhipu AI to be more at ease in the low-price competition and focus on building differentiated competitive advantages.
From a broader perspective, the involvement of Middle Eastern capital is expected to enrich the computing power resource reserves of domestic big model startups, increasing their chips in future technological research and development and market competition. For startups, the surging demand for digital transformation and generous incentive policies in Middle Eastern countries also make it a shining new market.