Mint Explainer: Will the $500 billion Stargate plan help the US trump China?

What is the Stargate Project?

Announced on Tuesday, it is a new company that intends to invest $500 billion over the next four years to build new artificial intelligence (AI)-focused data centres across the US.

The joint venture among OpenAI, Oracle Corp., SoftBank Group Corp., and MGX, which will begin deploying $100 billion immediately, aims to strengthen the country’s AI infrastructure, create 100,000 jobs, and enhance its competitiveness against China in AI development. 

Masayoshi Son will be the chairman.

Why is the US backing it?

The US has a good reason to be antsy in the AI space. 

It currently has the world’s most robust AI ecosystem, leading significantly across key metrics, according to Stanford University’s Global AI Vibrancy Tool, which ranks countries on AI research output, economic activity, and infrastructure. 

As of November 2024, the US led in producing high-quality AI research, building notable machine learning (ML) models, private investment, AI-related mergers/acquisitions, job postings, and newly funded AI startups.

That said, though China-based Baidu Inc.’s Ernie bot may not hold up to OpenAI’s ChatGPT, China ranks second in AI and is providing credible competition, attracting $67.2 billion in AI-related private investment in 2023 compared to a mere $7.8 billion for the US, according to the Stanford report. 

China also produced more notable ML models (61 vs 15) and led in AI-related patents.

Moreover, China continues to innovate in the space, which is worrying the US.

Chinese AI lab DeepSeek, for instance, is hitting the headlines with its new DeepSeek-R1, an open-source reasoning model that rivals OpenAI’s o1. This model has been trained using reinforcement learning (RL) without supervised fine-tuning (SFT).

The innovation is important since it avoids dependence on pre-labelled datasets, allowing the model to explore solutions independently and adapt to new challenges more flexibly. The company claims that using this method, DeepSeek-R1 has achieved performance comparable to OpenAI-o1 across math, code, and reasoning tasks while costing just 5-10% of o1’s API (application programming interface) price ($0.14 vs $7.50 per million input tokens).

Further, while the US hosts Nvidia Corp., the world’s largest AI company, most of the latter’s chips are produced by Taiwan Semiconductor Manufacturing Co. Ltd (TSMC). 

Since China considers Taiwan a breakaway province, navigating these complexities will shape the US semiconductor and AI landscape in the years to come.

What are its implications for the global AI race?

Governments across the world are sharpening their focus on AI data centres, which are equipped to host advanced AI workloads since the latter guzzle computing power. Sample this: while earlier a 30-megawatt (MW) data centre was considered large, now even a 200-MW  is considered “normal”, according to McKinsey & Co.

The consultancy firm estimated that even if all currently known plans are delivered on time, there could still be a data centre supply deficit of more than 15 gigawatts (GW) in the US alone by 2030.

McKinsey analysis suggested that demand for AI-ready data centre capacity is expected to grow at an average annual rate of 33% from 2023 to 2030. By that year, approximately 70% of the total data centre capacity demand will be for facilities equipped to handle advanced AI workloads. 

Generative AI (GenAI), the fastest-growing advanced AI use case, is projected to account for about 40% of this demand. Cloud service providers (CSPs) such as Amazon Web Services (AWS) Inc., Google Cloud Platform, Microsoft Azure, and Baidu are the companies fuelling most of today’s incremental demand for AI-ready data centres.

What are other countries doing? 

The UK ranks third in the Global AI Vibrancy Ranking 2023. India ranks fourth, followed by the United Arab Emirates (UAE), which is heavily investing in AI research through institutions like the Technology Innovation Institute. 

The UK government has announced a substantial expansion of data centres to attract AI firms. This initiative aims to remove planning obstacles to facilitate private-sector investments in AI infrastructure. Additionally, the European Union (EU) has been actively investing in AI infrastructure.

What does India stand? 

India, too, has initiated several significant plans to bolster its AI infrastructure. RackBank Datacenters Pvt. Ltd has announced its 80MW AI-focused data centre in Madya Pradesh with a 60,000 GPU (graphics processing unit) capacity. 

The country currently has about 150 data centres, including those from big companies—also known as hyperscalers due to their size—such as AWS, Microsoft Azure, Google Cloud, NTT DATA Group Corp., Sify Technologies Ltd, and CtrlS Datacenters Ltd, and those from newer entrants like Yotta Infrastructure, Digital Connexion, and Lumina CloudInfra Pvt. Ltd.

All of them are rapidly expanding their capacities to account for increasing digitalization, demand for AI and GenAI projects, nationwide rollout of 5G, laws mandating certain data be stored locally, and the need for edge computing that allows data processing on devices themselves.

Other than that, Microsoft Corp. recently announced a $3 billion investment to expand its cloud and AI services in India over the next two years, aiming to accelerate AI innovation and equip 10 million people with AI skills by 2030. 

Additionally, the Indian government has launched the IndiaAI Mission, a comprehensive national-level initiative with a budget outlay of ₹10,371.92 crore (approximately $1.25 billion), which focuses on enhancing computing infrastructure, supporting deep-tech startups, and developing data platforms to foster AI development across various sectors.

To be sure, India boasts a strong and cost-effective talent pool in software engineering, integrated circuit (IC) and manufacturing equipment design, with over 2,000 semiconductor chip design engineers. Leading semiconductor firms, including Intel Corp., Texas Instruments Inc., Nvidia Corp., Advanced Micro Devices Inc., and Qualcomm Technologies Inc., have established design, research and development centres in the country.

India is also advancing its semiconductor manufacturing capabilities. It has approved four major projects worth over ₹1.5 trillion, including ventures by Micron Technology Inc. and a partnership between Tata Electronics and Taiwan’s Powerchip Semiconductor Manufacturing Corp (PSMC). These initiatives aim to produce 180,000 wafers monthly upon completion. 

Besides, the Adani Group, in collaboration with Israel’s Tower Semiconductor Ltd, is planning a $10 billion fab in Panvel, Maharashtra, establishing a foundation for local chip-making and a full semiconductor value chain, including design, fabrication, and ATMP (assembly, testing, marking, and packaging). India has also set aside a $10 billion fund to attract global chipmakers.

India has also forged semiconductor agreements with the US, Singapore, and the EU, signalling its intent to solidify its global position in a sector challenged by raw material shortages and geopolitical tensions. Its focus on producing 28 nm and higher chips is strategic, balancing cost efficiency with longer shelf life. However, swift execution of these projects is essential for success.

What challenges does India need to address?

The National Mission for AI further underscores India’s ambitions, with plans to build infrastructure using 10,000 GPUs through public-private partnerships, supported by a ₹10,000 crore investment. However, India must address restrictions on US AI chip exports, tightened under the Joe Biden administration, to ensure access to critical technology.

Stargate shows the US’ “serious intent” to take control of AI and own all large language models (LLMs), said Ajai Chowdhry, founder of HCL (Hindustan Computers Ltd) and chairman of EPIC Foundation and the National Quantum Mission of India.

Dubbing Stargate as the “weaponization of tech”, he said India must create its “own AI doctrine” by controlling its data and domestic hardware for data centres.

Geopolitics, as we see, again remains a significant factor.