Other International Models & The Global AI Landscape

Learning Objectives

• Identify key models from Cohere and other international providers and understand their strategic positioning
• Explain the geopolitical dimensions of the global AI race — US-China competition, EU positioning, and export controls
• Articulate why AI development is now treated as a matter of national security and economic strategy

Canada: Cohere

Cohere is a Canadian AI company founded in 2019 by Aidan Gomez, Nick Frosst, and Ivan Zhang — all of whom worked at Google Brain and were co-authors on the original Transformer paper (the 2017 "Attention Is All You Need" paper that defines modern AI).

Cohere occupies a distinctive market position: while OpenAI, Anthropic, and Google compete in the consumer and general-purpose API market, Cohere has focused intensely on enterprise requirements — data privacy, deployment flexibility, and RAG performance.

In September 2025, Cohere reached a $7 billion valuation after raising $600 million across two rounds led by Radical Ventures and Inovia Capital, with participation from AMD Ventures, NVIDIA, and Salesforce Ventures. Revenue crossed $240 million ARR by end of 2025 — with 50%+ quarter-over-quarter growth. An IPO is widely expected in 2026.

Key leadership additions: Joelle Pineau (former head of Meta's FAIR research lab) joined as Chief AI Officer, and Francois Chadwick (who took Uber public) joined as CFO — both signaling IPO readiness.

Command A — The New Enterprise Flagship

Command A (March 2025) replaced Command R+ as Cohere's flagship model:

• 111 billion parameters with a 256K context window
• Runs on just 2 GPUs — a 150% throughput improvement over its predecessor
• Matches or exceeds GPT-4o and DeepSeek V3 on enterprise agentic tasks

The Command A family now includes:

• Command A Translate: First dedicated machine translation model, achieving state-of-the-art results across 23 languages
• Command A Reasoning: Hybrid reasoning model for complex agentic tasks

Any deployment environment: Unlike OpenAI and Anthropic, which require API access to their hosted servers, Cohere deploys on any cloud (AWS, Azure, Google Cloud), on-premise, or in air-gapped environments. For financial institutions, healthcare organizations, and government agencies that cannot transmit sensitive data to external APIs, this is a critical capability.

North — Agentic AI Platform

North (August 2025) is Cohere's agentic AI workspace, deployed entirely behind customer firewalls. Key customers include RBC (Royal Bank of Canada), Dell, and LG CNS. North for Telecom was co-built with Saudi Telecom (STC).

This behind-the-firewall approach is Cohere's key differentiator against Microsoft Copilot agents, Google Vertex AI agents, and Salesforce Agentforce — all of which involve sending data to cloud providers.

Command A

Cohere

Closed

Strengths

111 billion params; 256K context; runs on 2 GPUs; any cloud/on-premise/air-gapped deployment; enterprise RAG leader

Context Window

256K tokens

Pricing

Enterprise pricing; API available

Cohere →

Aya Expanse and Tiny Aya — Multilingual Leadership

Aya Expanse addresses one of the most significant gaps in the global AI landscape: the treatment of non-English languages.

The major US foundation models were trained predominantly on English text and perform significantly better in English than in other languages. Aya Expanse supports 23 languages with particular attention to low-resource languages that other models underserve. It was developed with contributions from researchers globally, including from regions where those languages are spoken.

Aya Vision (March 2025) extended this to multimodal — accepting images and text across 23 languages.

Tiny Aya (February 2026) is a family of open-weight 3.35 billion parameter models designed for 70+ languages on laptops and edge devices. Regional variants — Global, Earth (Africa), Fire (South Asia), Water (Asia-Pacific + Europe) — are tailored to specific language clusters. This is the broadest language coverage of any edge-deployable model.

Aya Expanse

Cohere

Open Source

Strengths

23 languages with strong low-resource language support; open source; multilingual optimization across global regions

Context Window

32K tokens

Pricing

Free (open source)

Cohere →

UAE and the Middle East

The United Arab Emirates has emerged as a surprising force in AI, leveraging sovereign wealth to build both models and infrastructure.

Falcon 3 — Open-Source from Abu Dhabi

The Technology Innovation Institute (TII) in Abu Dhabi released Falcon 3 in December 2024 — a family of open-source models at 1 billion, 3 billion, 7 billion, and 10 billion parameters, trained on 14 trillion tokens.

• Falcon3-10 billion leads the Hugging Face leaderboard for models under 13 billion parameters
• Multimodal variants (January 2025) process images, video, and audio
• Falcon-Edge uses Microsoft's BitNet 1.58-bit architecture for ultra-low-power deployment
• Falcon Arabic (May 2025) is a dedicated Arabic-language model outperforming other regional alternatives

G42 and Stargate UAE

G42, an Abu Dhabi-based AI company, has assembled a remarkable coalition:

• Stargate UAE (May 2025): A joint venture with OpenAI, Oracle, NVIDIA, SoftBank, and Cisco to build a 5 gigawatt AI campus across 10 square miles — the first 200MW cluster going live in 2026
• Microsoft is investing $15.2 billion in the UAE including a 200MW datacenter expansion through G42's Khazna subsidiary
• G42 co-founded MGX, a $25 billion investment fund targeting AI infrastructure and semiconductors
• G42-Mistral AI partnership for sovereign AI in the Middle East and Global South

The UAE's strategy is distinctive: rather than building its own frontier models, it is becoming the AI infrastructure hub for the Middle East, Africa, and South Asia — hosting and financing compute for regions that lack domestic AI capability.

United Kingdom

The UK remains Europe's leading AI market, raising over $6 billion in AI venture capital in 2025.

Key players:

• Google DeepMind (London) — arguably the world's leading AI research lab, despite being a US company
• ARM — chip architecture company critical to edge AI and mobile computing
• Wayve — autonomous driving AI startup with significant funding

Stability AI, the company behind Stable Diffusion, stabilized under new CEO Prem Akkaraju after founder Emad Mostaque's departure in March 2024. By end of 2024, the company had eliminated all debt and reported triple-digit revenue growth. However, an ongoing Getty Images copyright lawsuit in UK High Court (since June 2025) highlights unresolved legal questions around AI training data. Black Forest Labs' Flux has gained significant ground as an alternative open-source image model.

UK AI regulation takes a different approach from the EU: no single binding AI law, but a principles-based framework from the 2023 AI White Paper, relying on existing regulators (Ofcom, FCA, CMA) to apply AI-specific guidance within their domains. The government published an AI Opportunities Action Plan progress report in early 2026.

The AI Geopolitical Landscape

Understanding international AI models requires stepping back to understand why national origin matters in the AI era.

AI as National Strategic Asset

Governments around the world have concluded that AI is a strategic asset — not just economically important, but relevant to national security, military capability, economic competitiveness, and diplomatic influence. This is why:

• The US restricts export of advanced AI chips to China (and is tightening these restrictions)
• China invests billions in domestic AI semiconductor development (Huawei Ascend, Kunlunxin, Cambricon)
• The EU designates Mistral AI as Europe's AI champion and funds AI Factories across member states
• The UAE builds 5GW AI campuses and $25 billion infrastructure funds
• Japan commits $135 billion to sovereign AI infrastructure
• South Korea allocates $735 billion for sovereign AI development
• India backs Sarvam AI's sovereign model for 22 official languages

The era of AI as purely a commercial technology, developed by private companies for global markets, is over. AI is now also a geopolitical arena.

The US-China Competition

The United States and China are in the most consequential AI competition in history.

US advantages:

• Current frontier model leadership (OpenAI, Anthropic, Google DeepMind)
• NVIDIA chip dominance — H100/H200/Blackwell/GB200 are the training chips of choice
• Deep talent concentration in Silicon Valley, Seattle, and Boston
• Established venture capital and commercial ecosystem
• Alliance network (Five Eyes, EU, Japan, South Korea) with coordinated chip export controls

China's advantages:

• World's largest AI user base — ByteDance Doubao alone has 100 million+ daily active users
• Significant government investment: hundreds of billions committed to AI development
• Domestic market large enough to develop and scale AI applications independently
• Engineering talent at scale — China graduates more computer science engineers annually than any other country
• DeepSeek demonstrated that algorithmic efficiency can partially compensate for hardware constraints
• Huawei Ascend chips approaching NVIDIA parity, with a roadmap to 4 zettaflops by 2028
• Two AI companies IPO'd in Hong Kong in January 2026 alone (Zhipu AI, MiniMax)

The export control dynamic: US restrictions on exporting frontier chips to China are intended to slow China's AI development. The partial reversal on H200 exports (December 2025, with a 25% surcharge) reflected a pragmatic calculation: better to sell older chips than accelerate China's domestic chip development. Blackwell-class chips remain fully restricted.

China's counter-strategy: invest in domestic chip alternatives (Huawei Ascend 950PR with in-house HBM), develop more algorithmically efficient training methods (DeepSeek's $5.9 million training claim), and acquire non-restricted chips through third countries.

The EU: Regulatory Leadership as Strategy

The European Union's approach to AI geopolitics is distinctive: regulation as a form of soft power.

The EU AI Act has entered phased implementation:

• February 2, 2025: Prohibited AI practices banned (social scoring, real-time biometric ID in public spaces with limited exceptions, manipulation of vulnerable groups). AI literacy obligations took effect.
• August 2, 2025: General-Purpose AI (GPAI) model obligations became applicable. The AI Office became operational. Penalty regime took effect — fines up to 7% of global turnover for prohibited practices, 3% for other violations.
• August 2, 2026: Full applicability for most operators. Transparency duties. Each Member State must establish at least one AI regulatory sandbox.
• August 2, 2027: Extended transition for high-risk AI in regulated products (medical devices, vehicles).

When a company wants to operate in the EU market, it must comply with these requirements. This "Brussels effect" means European standards, in practice, become global standards for companies wanting access to the EU's 450 million consumers.

Mistral AI is central to this strategy: having a capable European model provider allows the EU to argue for regulatory standards that a European company can meet.

Middle Powers: Japan, South Korea, India, Canada

Several countries have made significant moves since early 2025:

Japan has committed $135 billion to sovereign AI infrastructure (including quantum and GPU buildout). Sakana AI — founded by Llion Jones, a co-author of the original Transformer paper — has become Japan's most valuable unicorn at $2.65 billion (November 2025). NTT released tsuzumi 2 (30 billion parameters, runs on a single H100 GPU) and committed $59 billion over 5 years to AI. The government's "Gennai" platform selected 7 AI vendors to serve 180,000 government staff, including NTT, Preferred Networks, and KDDI. Japan maintains a permissive regulatory approach to AI training on copyrighted content.

South Korea announced a $735 billion sovereign AI initiative and enacted the AI Basic Act, establishing the country's AI regulatory framework. Samsung launched the Mach-1 AI inference accelerator for edge computing and is building an AI factory with 50,000+ GPUs plus a $14.3 billion R&D center focused on AI semiconductors. Upstage released Solar Pro 2 (31 billion parameters) which beats GPT-4.1 on benchmarks — targeting Korea's first generative AI IPO in H2 2026. Naver unveiled Agent N as its AI strategy, while Kakao is developing on-device model Kanana for KakaoTalk integration.

India has the world's largest pool of English-speaking technology engineers. Sarvam AI (backed by Lightspeed and Khosla Ventures) launched Sarvam-105 billion (February 2026) — a 105 billion parameter model trained entirely from scratch on government-subsidized GPUs, competitive with global benchmarks, and supporting all 22 official Indian languages. Sarvam signed an MoU with Tamil Nadu for India's first Sovereign AI Park (~$1.2 billion). However, India's first AI unicorn Krutrim (founded by the Ola CEO) has faced senior engineer departures and layoffs by December 2025 — highlighting the difficulty of building AI companies outside established hubs.

Canada has deep AI research roots — Geoffrey Hinton's group at the University of Toronto was foundational to modern deep learning, and Yoshua Bengio's Mila institute in Montréal remains a world-leading AI research center. The government committed C$925 million for sovereign AI compute and C$1.7 billion for research talent. Cohere remains Canada's leading commercial AI company.

A Framework for Evaluating International Models

When evaluating an international foundation model for use in your organization, consider:

Key Takeaways

• Cohere (Canada) has reached $7 billion valuation and $240 million ARR with Command A (111 billion, 256K context, runs on 2 GPUs) and the North agentic platform deployed behind customer firewalls — IPO expected in 2026
• The UAE has emerged as an AI infrastructure hub through TII's Falcon 3 open-source models and G42's Stargate UAE partnership (5GW campus with OpenAI, Oracle, NVIDIA)
• The UK leads European AI VC ($6 billion+ in 2025) and hosts Google DeepMind, while taking a principles-based regulatory approach distinct from the EU AI Act
• The EU AI Act is now in active enforcement: prohibited practices banned (Feb 2025), GPAI rules live (Aug 2025), full applicability coming August 2026 with fines up to 7% of global turnover
• Japan ($135 billion sovereign AI), South Korea ($735 billion initiative + AI Basic Act), and India (Sarvam-105 billion sovereign model) are making major national AI investments
• AI has become a matter of national strategic interest — the US-China chip competition, sovereign AI initiatives, and regulatory frameworks reflect a permanent geopolitical shift