The AI race is not a two-horse contest. For the past three years, most coverage has framed it as America vs China — OpenAI vs DeepSeek, Silicon Valley vs Beijing. That framing is increasingly incomplete. India has quietly built the world’s third-largest AI startup ecosystem, attracted $50 billion in global AI investment, hosted the first global AI summit in the Global South, and had Sam Altman, Sundar Pichai, and Dario Amodei all fly in to attend. That’s not a country watching from the sidelines.

But “winning” means different things in each country. The US is building the frontier — the most powerful models, the most capital, the most infrastructure. China is building efficiency — doing more with less, bypassing chip bans through innovation. India is building for a billion people — multilingual AI, frugal compute, sovereign models for a market that doesn’t look or sound like California. These are three genuinely different bets on what AI is actually for.

The Three Contestants — At a Glance

The United States: Unlimited Capital, Unlimited Ambition

The US AI story in 2026 is fundamentally a story about capital concentration at a scale that has no historical precedent. OpenAI raised $122 billion in March 2026 alone. Anthropic — now the most valuable AI company in Silicon Valley at a $965 billion valuation — raised $65 billion in May. Google committed $190 billion in capital expenditure this year. Microsoft, Meta, and Amazon are deploying comparable sums. Combined, the US tech sector is investing somewhere north of $600 billion in AI infrastructure in 2026.

That capital buys two things nobody else can easily replicate: compute and talent. US labs have unconstrained access to Nvidia’s most advanced chips — the H100, H200, and now the Blackwell B200. OpenAI’s Stargate project envisions 33 gigawatts of AI compute. Anthropic has secured 1 million Google TPUs and 1 gigawatt of compute. These aren’t incremental upgrades — they’re civilisation-scale infrastructure investments.

The frontier models that result — GPT-4o, Claude Opus 4.8, Gemini 2.5 — remain the global benchmark for reasoning, coding, multimodal understanding, and agent behaviour. Every other AI lab in the world measures itself against these. That’s a genuine competitive moat.

China: The Chip Ban That Became a Superpower

The conventional narrative about Chinese AI is that US export controls on advanced chips will gradually choke its development. That narrative is being revised in real time. DeepSeek’s January 2025 announcement — matching OpenAI o1 performance at a fraction of the cost, using older chips — was the clearest proof yet that constraints can drive innovation rather than simply limit it.

DeepSeek’s R1 model, and now its successor models, demonstrated that the efficiency gap between constrained and unconstrained AI development is much smaller than Silicon Valley assumed. The company is now raising $7.4 billion — one of China’s largest ever startup rounds — from Tencent, CATL, and the National AI Industry Investment Fund. Its open-weight models have reached approximately 30% of global AI token usage at their peak.

Meanwhile, ByteDance’s Doubao has over 200 million monthly users. Alibaba’s Qwen models are among the most capable open-weight models available globally. Moonshot’s Kimi is expanding internationally. The Chinese AI ecosystem isn’t waiting for permission from Nvidia.

India: The Billion-User Opportunity Nobody Is Writing About

India’s AI story is the least covered of the three — and arguably the most interesting for the next decade. Here’s the starting point most Western coverage misses: India has 22 official languages, 1.4 billion people, and the world’s largest English-speaking engineering workforce. That combination creates an AI market that looks nothing like California — and that US and Chinese models are genuinely bad at serving.

Sarvam AI’s Vision OCR model scored 84.3% on olmOCR-Bench for Indian language tasks — outperforming Gemini 3 Pro (80.2%) and ChatGPT (69.8%). That’s not a minor benchmark difference. It means an Indian bank deploying AI for Hindi-speaking customers gets meaningfully better results from Sarvam than from OpenAI. That’s a real market advantage that compounds as India’s digital economy scales.

Krutrim — India’s first AI unicorn, backed by Ola founder Bhavish Aggarwal — built Krutrim-2, a 12-billion-parameter multilingual model trained on 22 Indian languages. In early 2026 it looked like a direct challenger to OpenAI for Indian language AI. But in May 2026, TechCrunch reported that Krutrim has pivoted away from frontier model development toward cloud infrastructure and compute services — pulling its Kruti AI assistant app from stores, scrapping its chip design programme, and refocusing on sovereign GPU cloud services for Indian enterprises. The company posted ₹3 billion (~$31.5M) in revenue for FY2026 — a threefold increase — and its first annual profit. The pivot is actually revealing: building sovereign AI models is harder and more expensive than building the cloud infrastructure that runs them. Krutrim chose the commercially viable path. That’s not failure — it’s the same path AWS took before Amazon became the dominant cloud provider.

The Global Investment Flowing Into India

The India AI Impact Summit in February 2026 — the first global AI summit ever held in the Global South — generated over $200 billion in investment commitments. Google committed $15 billion over five years to build an AI hub in Visakhapatnam. Anthropic opened a Bengaluru office — its second in Asia — after CEO Dario Amodei met Prime Minister Modi to discuss AI applications in education, healthcare, and agriculture. Microsoft, Nvidia, and Qualcomm all have major India AI programmes. The IndiaAI Mission has allocated ₹10,000 crore ($1.25 billion) in government funding, with the potential to double.

Head-to-Head: Where Each Country Actually Leads

What This Means for You — Wherever You Are

If you’re a developer or startup founder, the practical implication is that the best AI tool for your use case depends heavily on what language you’re working in and who your users are. Building in English for a Western audience? US models are still the best. Building for Hindi, Tamil, or Telugu speakers? Sarvam outperforms OpenAI on your actual use case. Building something that needs to work cheaply at scale? DeepSeek’s open-weight models are worth serious consideration.

If you’re watching this as a technology observer, the most important thing to understand is that the race is not converging — it’s diverging. Each country is increasingly building AI for its own market, its own language, its own regulatory environment. The idea of a single global AI winner — one model that everyone everywhere uses — is looking less likely by the month. What’s more likely is a fragmented world of regional AI ecosystems, each optimised for its own context.

If you’re an investor or business leader, the India story is the most underpriced of the three. The US story is fully priced — Anthropic at $965 billion, OpenAI at $852 billion. China is geopolitically constrained in Western markets. India has $50 billion in investment but is still in early innings relative to its population scale. The India AI Impact Summit’s $200 billion in commitment is a signal, not an endpoint.

My Take — Who’s Actually Winning?

Nobody. And that’s the honest answer.

The US is winning the frontier. If you define “winning” as building the most capable models and attracting the most capital, the US is so far ahead it’s not a race. OpenAI and Anthropic’s combined valuation exceeds the GDP of most countries. The compute advantage is real and compounding.

China is winning efficiency and domestic scale. DeepSeek’s January 2025 moment was genuine. The ability to match frontier performance at a fraction of the cost, on restricted chips, is an extraordinary engineering achievement. For the 1.4 billion people in China’s domestic market, Chinese AI is not inferior — it’s optimised for them in ways US models aren’t.

India is winning the long game. The combination of a 1.4 billion person market, the world’s largest engineering talent pool, multilingual-first AI models, and $200 billion in incoming investment is a foundation that could produce the most impactful AI applications of the 2030s — healthcare, agriculture, financial inclusion, education at a scale that Silicon Valley isn’t building for. India’s AI isn’t about prestige benchmarks. It’s about solving real problems for a billion people who have been largely invisible to the first wave of AI development.

The race has three different finishing lines. And all three contestants are still running.

Mr Wangdoo

Founder and Editor-in-Chief of Wangdoo.com. Independent tech journalist covering AI, EVs, gadgets, and emerging tech.