By
Ayush Banerjee
Africa-Press – Eritrea. The rise of large language models has led to a new wave of global inequality, where data, compute, and capital are concentrated in the hands of a few dominant powers.
According to Stanford’s 2025 AI Index, the U.S. alone accounted for 40 out of 149 significant models in 2024—dwarfing Europe’s three. This disproportionate control risks mirroring historic power asymmetries. As AI permeates governance, media, education, and national security, unchecked concentration may shape global thought and policy in ways that echo post-war Western dominance, only now encoded in algorithms.
This dominance is largely due to the substantial investments and resources. U.S. private AI investment reached $109.1 billion in 2024, which is nearly 12 times China’s $9.3 billion and 24 times the UK’s $4.5 billion. Such capital enables leading Western labs such as OpenAI, Google, and Meta to operate on scales inaccessible to most global researchers.
Modern AI pivots a lot on semiconductor hardware. Over 90% of these chips are produced in just five countries: the U.S., Taiwan, China, South Korea, and Japan. Taiwan’s TSMC produces 55% of advanced-node chips, Samsung 18%, Intel 13%, and SMIC about 6%. U.S. export controls on NVIDIA GPUs cut China’s access by an estimated 40% in 2023.
From 2013 to 2024, U.S. private AI funding totaled ~$470.9 billion. China followed at $119.3 billion, with the UK ($28.2B), Canada ($15.3B), Israel ($15B), Germany ($11.3B), and France ($9B) trailing behind. In 2023 alone, U.S. AI startups raised $75 billion versus China’s $18 billion, and in Q1 2024, the U.S. led again with $20 billion while China drew $5 billion.
China now accounts for nearly 70% of global AI patent filings, up from 40% in 2010. The U.S. share has declined to 14% as of 2023 (WIPO IP Statistics Data Center). In 2023, Chinese entities filed 45,000 AI-related patents compared to 9,000 (U.S.), 5,500 (Japan), and 4,200 (EPO). Major patent filers included Huawei (8,000), Baidu (6,500), IBM (4,200), Microsoft (3,800), and Tencent (3,500).
LLMs are trained primarily on Western-centric data—mostly in English, German, Spanish, and Chinese. Over 3,000 languages are poorly represented. This causes output biases that privilege U.S./European norms and marginalize local or indigenous knowledge systems.
Meanwhile, hyperscalers earn roughly $30 billion annually from non-home markets via API and cloud subscriptions while extracting local data for training.
In response, some regions are asserting technological sovereignty. The EU AI Act (2024) applies to all AI systems within the EU and imposes penalties up to 6% of global turnover for violations. The European Chips Act similarly boosts domestic semiconductor manufacturing. In the Global South, projects like Masakhane, Argos Translate, and AI4Bharat (which received Rs. 200 crore, or $25 million, in Indian public funding) are building open-source models for African and Indian languages.
Post-WWII Bretton Woods institutions such as the IMF and World Bank cemented U.S. financial control via conditional aid. Today’s AI governance forums, such as the G7 AI Safety Summit 2023, OECD AI Principles 2019, and the upcoming U.N. Global Digital Compact, risk doing the same unless structural changes allow Global South vetoes or mandatory IP cross-licensing.
Without action, LLM hegemony could homogenize global discourse. Western-centric moderation policies and disinformation filters may suppress dissent or non-mainstream views. Embedding diverse voices through frameworks like UNESCO’s Recommendation on AI Ethics or India’s DPDP Act is vital.
Pathways Toward an Inclusive AI Ecosystem
Creating an inclusive AI ecosystem calls for systemic reforms that address both access and equity. A fundamental starting point is investing in compute infrastructure and digital literacy in underserved regions—particularly the Global South. Over 2.7 billion people globally still lack internet access, according to the International Telecommunication Union (2023), leaving large swathes of the world excluded from the digital revolution.
Open-source AI models such as BLOOM, Mistral in 2023, and Meta’s LLaMA are pivotal in democratizing access. They come with permissive licenses, enabling researchers and smaller actors to experiment and build without the high costs or legal constraints of proprietary models.
However, equitable AI governance requires enforceable international frameworks. Proposals like the 1% hyperscaler revenue levy, suggested by researchers such as Meredith Whittaker and the AI Now Institute, aim to redistribute compute resources and fund global AI infrastructure.
Transparency through public audits, open training datasets, and algorithmic explainability is essential for ethical AI. Additionally, only 3% of global AI research funding reaches non-OECD countries, according to the Wellcome Trust’s 2023 global funding report.
Currently, AI’s core assets, LLMs, compute power, and capital, remain highly concentrated. But through open models, inclusive datasets, and equitable funding, AI can evolve into a shared global public good. The future of intelligence must be collective.
Building an inclusive AI ecosystem requires structural reforms. First, investing in compute infrastructure and AI literacy across underserved regions is critical; over 2.7 billion people still lack internet access.
Open-source models like BLOOM and Mistral lower barriers through permissive licensing. Equitable governance demands enforceable global treaties, including proposals like a 1% hyperscaler revenue levy to fund shared AI resources. Ethical AI requires public audits and transparency in datasets and training methods. Finally, funding must flow to the Global South—less than 3% of AI research grants currently reach non-OECD nations (Wellcome Trust). These steps can close the AI divide and promote global participation.
The data is clear: AI’s core assets, LLMs, chips, and capital, are clustered in the hands of a few powers. This mirrors historical hegemony, reinforcing inequalities in knowledge and power. Yet the tide can shift. Through equitable governance, inclusive datasets, and transparent technologies, we can reshape AI as a global public good.
The future of intelligence should belong to all.
moderndiplomacy
For More News And Analysis About Eritrea Follow Africa-Press
