GDSC Hult Weekly Update: May 27th - June 3rd Newsletter
Elon Musk’s xAI Valued at $24 Billion After Latest Fundraising Round
Wall Street Journal (May 27, 2024)
Despite the steep decline in Tesla’s shares and the devaluation of X since Musk's acquisition in late 2022, his ability to attract investment remains strong. Musk announced that xAI has raised $6 billion, bringing its valuation to $24 billion. This funding, exceeding initial expectations of $3 billion at an $18 billion valuation, will be used to launch xAI’s first products, build advanced infrastructure, and accelerate research and development.
Investors in this round include prominent firms like Sequoia Capital, Andreessen Horowitz, Valor Equity Partners, Vy Capital, Fidelity Management & Research, and Saudi Prince al-Waleed bin Talal. Additionally, Musk’s personal funds and contributions from X amounting to $750 million and $250 million respectively will secure X investors a 25% equity stake in xAI. Now, xAI ranks as the second most valuable AI startup globally, behind OpenAI ($86 billion) and ahead of Anthropic ($18 billion).
xAI’s chatbot Grok, trained on Oracle’s computing chips, debuted on X in November. The latest version, Grok 1.5, was introduced earlier this year. A significant draw for investors is the extensive data sources available through X and Tesla. X data is already utilized to train Grok, providing new summaries for Stories on the platform. Investors are also keen on the potential use of Tesla’s visual data to train xAI models and integrate the technology into Tesla’s Optimus humanoid robot. Musk highlighted Tesla's advantage in real-world video data, which is crucial for AI development.
China’s $47B semiconductor fund puts chip sovereignty front and center
TechCrunch (May 28, 2024)
The latest round of China's National Investment in the Integrated Circuit Industry, known as the Big Fund, is nearing $47.5 billion. This major investment highlights China's drive for semiconductor independence from international suppliers and the ongoing chip war with the West. China produces 60% of the world’s legacy chips, essential for cars and appliances. The funds will mainly support wafer manufacturing and High Bandwidth Memory (HBM) chips, vital for AI, 5G, and IoT technologies.
Taiwan is a focal point in this competition, given its dominance in semiconductor production, accounting for 90% of the global output. China aims to control Taiwan's production capabilities, while the U.S. and its allies seek to prevent such a shift to avoid strategic disadvantages. Companies like ASML and TSMC, with the ability to disable chip-making machines, could mitigate the impact of a potential Chinese invasion.
The geopolitical tension affects U.S. companies like Nvidia, which must balance relations with China and the U.S. while staying competitive. If Chinese AI firms lose access to Nvidia products, their goal of self-reliance could be hindered. Although the Big Fund III surpasses the $39 billion in direct incentives from the U.S. CHIPS Act, it is not as extensive as the entire CHIPS Act federal funding. This investment is causing a surge in the Chinese stock market, as investors eagerly await to see which companies will secure the most benefits.
The next wave of AI hype will be geopolitical. You’re paying
Financial Times (May 29, 2024)
The private and public sectors are pouring billions into AI, with Amazon, Meta, Google, and Microsoft collectively spending $200 billion in capex this year, according to Bernstein Research. Nations such as Saudi Arabia, Singapore, Germany, the UK, and India are also making substantial investments, with the US alone hitting $4 billion in AI spending. The US Senate's Roadmap for Artificial Intelligence Policy suggests a $32 billion R&D budget for AI, reflecting the massive scale of investment.
Hardware costs are escalating, with public spending potentially exceeding $25 billion annually. The Senate’s budget excludes defense AI spending, but a Brookings Institution study notes the US Department of Defense spent over $4 billion on AI procurement last year. This spending will likely rise as other nations try to stay competitive, beginning with last year's technology as a baseline.
According to Barclays, if hardware cost inflation continues, the expense for a top-tier AI computing cluster could surpass $5 billion, limiting AI investment to about 15 nations. As investing in AI becomes a matter of national security, countries like Saudi Arabia, Singapore, Germany, and India, which have established significant AI investment funds, may motivate others to solidify their AI strategies earlier. Nvidia stands to benefit significantly, as governments rely on its comprehensive solutions due to a lack of custom engineering skills. This urgency to invest is driven by the need to stay competitive, even when the exact benefits are not fully understood.
The private sector is also grappling with the challenge of sustaining AI investments amid high costs and uncertain returns. With the rapid advancement of AI technologies, companies must continually invest in cutting-edge hardware and software, often without clear short-term benefits. However, the potential for long-term gains and maintaining a competitive edge keeps the momentum going.
Tech giants form an industry group to help develop next-gen AI chip components
TechCrunch (May 30, 2024)
AI accelerators are specialized devices designed to speed up the training and fine-tuning of AI models. Big tech companies heavily invest in these accelerators, with Gartner estimating $21 billion spent on AI accelerators in servers this year. If linked together, these accelerators could create a value-adding ecosystem, fostering innovation and competition against Nvidia's NVLink.
Ultra Accelerator Link (UALink) aims to achieve this by forming an industry group, including Intel, Google, Microsoft, and Meta, to develop tools that connect AI accelerators via a single Ultra Accelerator Switch. This will enhance speed and reduce data transfer latency by enabling direct memory interactions between AI accelerators. UALink 1.0, the first version, supports over 1,024 AI accelerators based on open standards. A consortium will be established in Q3 to oversee UALink’s progress.
Nvidia's dominance in the AI accelerators field, holding 80-90% of the market share, and a surge of 400% in sales in a year puts the company in a position of not having to worry about being part of UALink. AWS, another absentee, is evaluating its own hardware accelerators performance first and may avoid opposing Nvidia. However, Nvidia's GPUs’ key users—Microsoft, Meta, and Google—are also important UALink beneficiaries, seeking to challenge Nvidia's market power.
Saudi fund invests in China effort to create rival to OpenAI
Financial Times (May 31, 2024)
Zhipu, the Chinese startup aiming to rival OpenAI, has secured a $400 million investment from Prosperity7, the venture capital arm of Saudi Arabia-based oil company Aramco, valuing the company at $3 billion. This marks a significant international investment in a Chinese AI firm amidst U.S. bans on certain investments and tightened export controls. The move underscores Saudi Arabia’s intent to challenge U.S. dominance in the AI industry.
Chinese AI startups have primarily relied on local funding and cloud providers, but international investments can open new, more lucrative markets. This investment signals a potential shift, enabling Chinese firms to expand beyond domestic borders.
Additionally, Lenovo announced the issuance of $2 billion in bonds to Alat, a subsidiary of Saudi Arabia’s Public Investment Fund, and plans to expand its headquarters to the country. However, Saudi Arabia's alignment with U.S. interests remains flexible. For example, investment firm G42 divested from Chinese tech companies like ByteDance following a $1.5 billion investment from Microsoft. Alat's CEO, Amit Midha, indicated willingness to withdraw from Chinese investments if it conflicts with Saudi Arabia’s plans to establish a semiconductor industry.
VCs are selling shares of hot AI companies like Anthropic and xAI to small investors in a wild SPV market
TechCrunch (June 1, 2024)
Despite major venture capitalists struggling to secure stakes in promising AI startups, smaller, lesser-known investors are finding their way into companies like Anthropic, Groq and OpenAI. Special Purpose Vehicles (SPVs) play a crucial role in this process. SPVs, which pool money from multiple parties to invest in a single company, are used by investors with access to specific shares. They then sell these shares to smaller investors, charging high fees and keeping a profit share, known as carry. Part of xAI’s $6 billion capital raise was through SPVs, which included up to 5% of the total investment upfront fee, management fees, and carried interest.
Often, early investors in popular AI companies cannot afford to exercise their pro-rata rights to buy more shares in subsequent rounds. Instead of giving up on their shares, they create SPVs to raise the necessary funds. Sometimes, even secondary SPVs are created, as seen when Menlo Ventures raised a $750 million SPV to invest in Anthropic , and some funds resold parts of their SPV allocation to other investors. However, a downside of SPVs is that backers do not receive direct information from the company.
Commenting on the resurgence of SPVs, Jack Selby, managing director at Thiel Capital and founder of AZ-VC Fund, said, “It boggles my mind that just a few years after the excesses of the 2020 and 2021 period, when people were essentially investing blindly into SPVs, with fees on fees on fees, into vehicles that were totally opaque, people are doing that all over again with everything that is a shiny toy: AI.”
AI is promoted from back-office duties to investment decisions
Financial Times (June 2, 2024)
JPMorgan's plans to use the AI tool "Moneyball," which leverages about 40 years of data to analyze and question analysts’ decisions, and Voya Investment Management’s virtual assistant designed to analyze stocks and alert about potential risks, indicate a shift in AI competition among asset management companies. This shift moves away from paperwork-intensive tasks towards enabling professionals to make smarter decisions.
Despite the success of tools like Voya’s AI analyst, which has shown a good ratio of right to wrong decisions according to Voya’s co-head of machine intelligence Gareth Shepherd, skepticism remains. Veteran portfolio manager David Giroux, for example, doubts AI's ability to significantly reduce inefficiencies in estimating long-term earnings potential, stating, “I think there is very, very little that AI is going to do to make that inefficiency go away.”
Nevertheless, there are believers in AI's potential. South Korean conglomerate LG's AI-powered exchange-traded fund and SoftBank-backed Qraft Technologies are not only implementing AI but also granting it more decision power. Their AI systems are tasked with stock picking and generating monthly holdings reports, demonstrating a more hands-off approach and greater trust in AI's capabilities in asset management.
AMD Unveils Latest AI Chips, Accelerated Chip Update Timeline
Wall Street Journal (June 3, 2024)
At a trade show in Taipei on Monday, Advanced Micro Devices (AMD) showcased its newest accelerator chip, the MI325X, as part of its strategy to challenge Nvidia’s dominance in the AI market. Taking advantage of Nvidia's chip shortages and the increasing demand for alternatives, AMD aims to capture a portion of Nvidia's 80% market share, which has been crucial for developing AI tools like ChatGPT.
AMD plans to release an upgraded version, the MI350 series, in 2025, which is expected to improve data processing through a trained model due to its new architecture. Following this, the MI400 series is slated for release in 2026. AMD's Chief Executive Lisa Su has projected that the company's AI chip sales for data centers will surpass $4 billion this year, as stated during the company's first-quarter results in April.
Extra News Highlights
AI models have favorite numbers, because they think they’re people
TechCrunch (May 28, 2024)
Even though it might seem like AI models think they are people, they do not actually think. However, they are programmed to act as if they do, providing solutions or answers based on human-made content available on the internet, which they are trained on. In this context, an interesting phenomenon has been observed in AI models' behavior regarding the randomization of numbers.
Due to the inherently non-random nature of randomness (for example, tossing a coin 100 times might result in sequences such as seven consecutive tails), humans tend to misunderstand and overthink it. Humans have standard biases towards certain numbers, such as rarely choosing multiples of 5, repeating digits like 77, and, when asked to pick a number between 1 and 100, often avoiding the extremes. This is due to certain qualities of these numbers: they are small, large, or distinctive.
It has been proven that AI exhibits the same biases in its answers. Engineers at Gramener conducted an experiment asking LLM chatbots to pick random numbers from 1 to 100, and the results showed that their choices were not truly random. ChatGPT-3.5 Turbo frequently chose 47, and previously favored 42, a number made famous by Douglas Adams in “The Hitchhiker’s Guide to the Galaxy” as the answer to life, the universe, and everything. Anthropic also frequently chose 42, while Gemini often chose 72. The biases mentioned earlier were evident in the models' answers, even at higher "temperatures," a setting that increases the model’s variability. Claude never chose numbers above or below 27.
More interestingly, all three models demonstrated human-like biases in the other numbers they selected, even at high temperature settings. Double digits were scrupulously avoided: no 33s, 55s, or 66s, but 77 appeared. Round numbers were rarely chosen, with the exception of Gemini choosing 0 once at the higher temperature.
Credits:
- Rafaella Andrade
- GANGA SINGH
- GDSC Events Team
References
- Jin, B., Bobrowsky, M., & Kao, K. (2024, May 27). Elon Musk's xAI will raise $6 billion in latest fundraising round. The Wall Street Journal. https://www.wsj.com/tech/ai/elon-musks-xai-will-raise-6-billion-in-latest-fundraising-round-fcdd722d
- Coldewey, D. (2024, May 28). AI models have favorite numbers, because they think they’re people. TechCrunch.https://techcrunch.com/2024/05/28/ai-models-have-favorite-numbers-because-they-think-theyre-people/#:~:text=AImodels are always surprising,which is to say%2C badly.
- Heim, A. (2024, May 28). China’s $47B semiconductor fund puts chip sovereignty front and center. TechCrunch. https://techcrunch.com/2024/05/28/chinas-47b-semiconductor-fund-puts-chip-sovereignty-front-and-center/
- Elder, B. (2024, May 29). The next wave of AI hype will be geopolitical. You’re paying. Financial Times. https://www.ft.com/content/a60c3c7b-1c48-485d-adb7-5bc2b7b1b650
- Wiggers, K. (2024, May 30). Tech giants form an industry group to help develop next-gen AI chip components. https://techcrunch.com/2024/05/30/tech-giants-form-new-group-in-effort-to-wean-off-of-nvidia-hardware/#:~:text=Intel%2C Google%2C Microsoft%2C Meta,accelerator chips in data centers.
- Olcott, E. (2024, May 31). Saudi fund invests in China effort to create rival to OpenAI. https://www.ft.com/content/87a40ffe-c791-4c90-8123-3f75aa0ed26b
- Temkin, M. (2024, June 1). VCs are selling shares of hot AI companies like Anthropic and xAI to small investors in a wild SPV market. https://techcrunch.com/2024/06/01/vcs-are-selling-shares-of-hot-ai-companies-like-anthropic-and-xai-to-small-investors-in-a-wild-spv-market/#:~:text=TechCrunch-,VCs are selling shares of hot AI companies like Anthropic,in a wild SPV market&text=VCs are clamoring to invest,into such deals at all.
- Schmitt, W. & Masters B. (2024, June 2). AI is promoted from back-office duties to investment decisions. https://www.ft.com/content/3d82ea9f-f040-47aa-9b9d-0be9decdbb14
- Qin, S. (2024, June 3). AMD Unveils Latest AI Chips, Accelerated Chip Update Timeline. https://www.wsj.com/tech/amd-unveils-latest-ai-chips-accelerated-chip-update-timeline-d2130ee9