The Sequence Radar #783: Softbank, DeepSeek, MiniMax and The Sequence 2026

2025-07-08 · Source: TheSequence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation, Cloud Computing & IT Infrastructure · Depth: Advanced, medium

Summary

The first week of 2026 established a clear bifurcation in the AI ecosystem, marked by SoftBank's aggressive infrastructure consolidation and DeepSeek's advancements in algorithmic efficiency. SoftBank Group committed $41 billion to OpenAI and acquired DigitalBridge for $4 billion, aiming to vertically integrate AI models with their underlying data center and edge infrastructure. Concurrently, DeepSeek released DeepSeek-V3.2, an improved Mixture-of-Experts (MoE) architecture featuring Group Relative Policy Optimization (GRPO). This innovation allows DeepSeek to match the reasoning capabilities of models like GPT-5 and Gemini 3.0 Pro with significantly reduced GPU hours. Additionally, Chinese AI unicorn MiniMax filed for a $539 million IPO on the Hong Kong Stock Exchange, signaling a major test for the applied AI market. These events highlight a dual trend of capital-driven dominance and innovation-led efficiency in the evolving AI landscape.

Key takeaway

For AI Architects and Machine Learning Engineers evaluating model deployment strategies, recognize that the industry is splitting between large-scale infrastructure plays and algorithmic efficiency. You should investigate advanced MoE architectures like DeepSeek-V3.2 to achieve competitive reasoning capabilities with fewer compute resources, potentially reducing operational costs and hardware dependencies, especially when facing capital constraints or export controls.

Key insights

AI's future is bifurcated between capital-intensive infrastructure consolidation and efficiency-driven algorithmic innovation.

Principles

• Algorithmic innovation can rival capital dominance.
• Vertical integration blurs model and infrastructure providers.

Method

DeepSeek-V3.2 utilizes Group Relative Policy Optimization (GRPO) within a Mixture-of-Experts architecture, enabling self-evaluation of group outputs without a resource-heavy critic model during reinforcement learning.

In practice

• Explore MoE architectures for training efficiency.
• Consider vertical integration for AI deployments.

Topics

• AI Investment Strategy
• Algorithmic Efficiency
• AI Infrastructure
• Large Language Models
• World Models

Best for: AI Architect, Machine Learning Engineer, NLP Engineer, AI Engineer, AI Researcher, CTO

Related on AIssential

• The Sequence Radar #816: Last Week in AI: $110B Bets, Nano Banana 2, and the New Economic Reality — AI is rapidly reshaping economic structures and geopolitical dynamics through massive investment and advanced model capabilities.
• MLWhiz Weekly AI/ML Newsletter # 3 — The AI industry is rapidly fragmenting across infrastructure, models, and applications, moving beyond single-vendor dominance.
• Google’s $185B Infra Bet & DeepMind’s Memory Fix, India’s GPU Surge, and the Mobile RAM Crisis — AI's rapid growth is driving massive infrastructure investments and creating critical supply chain pressures across the tech industry.
• OpenAI Kills Sora | $15M/Day Burn, Musk’s $75B IPO & India’s AI Rise — AI product viability increasingly depends on clear business outcomes and cost-justified scalability over mere capability showcases.
• The Sequence Radar #837: Last Week in AI: From Model Releases to Market Structure — AI is evolving into an infrastructure play, prioritizing scale, integration, and open models over raw product demos.
• 2025 Interconnects year in review — DeepSeek's open-weight, efficient MoE and MLA architectures challenge AI's cost curve, intensifying the global AI race and geopolitical tech competition.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by TheSequence.