Welcome to “This Week in AI” for July 20–26, 2025, a week where global policy ambitions clashed and converged, infrastructure scaled up, and corporate AI leadership made bold moves. From Washington’s newly unveiled “America’s AI Action Plan” to China’s counterproposal for a global AI governance body, geopolitics loomed large over technological progress. Meanwhile, AWS debuted powerful new tools for building agentic AI, Meta poached a key OpenAI scientist, and Google touted massive user growth in Gemini and AI search. As AI becomes the battleground of innovation and influence, this week proved one thing: the race for intelligent leadership is accelerating on every front.
Global AI Diplomacy & Policy
U.S. unveils “Winning the Race: America’s AI Action Plan” (July 23, 2025)
The Trump Administration released a 28‑page federal roadmap laying out 90+ action items to reinforce U.S. AI leadership. The three pillars:
- Accelerate innovation through deregulation and fast-tracked R&D cooperation.
- Build infrastructure via expedited permits for data centers and chip fabs, and support workforce training.
- Lead global AI diplomacy and security, including AI export packages in partnership with allies.
The plan underscores deregulation ambition while preserving export controls for advanced AI compute hardware.
China proposes global AI governance body (July 26, 2025)
At the World Artificial Intelligence Conference in Shanghai, Premier Li Qiang proposed launching an international AI cooperation organization, headquartered in Shanghai. The aim: create a globally inclusive, UN‑aligned AI governance framework, not dominated by Western interests, and foster broader access especially for the Global South.
BRICS declaration reinforces inclusive AI oversight (July 6–7, 2025)
Though slightly earlier, the Rio summit’s BRICS nations formally called on the UN to lead global AI governance, emphasizing inclusive, equitable development and cautioning that unchecked AI could worsen the digital divide. This resonates with China’s Wednesday proposal and reflects a growing geopolitical alignment around AI norms.
🏢 Corporate & Industry Moves
Alphabet (Google): AI momentum fuels strong earnings
In its Q2 2025 earnings, Alphabet posted $96.4 billion revenue (+14% YoY), with $54.1 billion from search. CEO Sundar Pichai highlighted AI’s impact:
- AI Overviews now serve over 2 billion monthly users.
- AI Mode has surpassed 100 million users in the U.S. and India.
- Gemini app reached 450 million monthly active users.
Google plans to hike capex to $85 billion to scale its AI‑smart cloud and data center infrastructure.
Meta hires ChatGPT co‑creator as Chief Scientist
On July 26, Meta tapped Shengjia Zhao, a GPT‑4 co‑creator and former OpenAI lead, to lead its Superintelligence Lab. Mark Zuckerberg hailed him as a pioneer, underlining Meta’s escalating talent competition with OpenAI.
AMD embraces AI‑accelerated GPU design
At the “Winning the AI Race” summit, AMD CEO Lisa Su discussed using AI tools to speed and enhance the design process of next‑gen GPUs. While not replacing engineers, AI is positioned as a force multiplier that could drive more efficient and cost‑effective chip development PC Gamer.
☁️ Cloud & Infrastructure Innovations
AWS Summit NY reveals major AI platform enhancements
Amazon Web Services introduced three significant new offerings aimed at developer productivity and agentic AI:
- Bedrock AgentCore – a seven-part toolkit for secure AI agents with governance features like API Gateway, memory, and browsing.
- AI Marketplace in AWS – enables discovery and purchase of third‑party AI agents from providers like Anthropic, IBM, Brave.
- S3 Vectors – native vector‑storage in S3 offering up to 90% cost savings for vector workloads, integrated with Bedrock and OpenSearch.
OpenAI + SoftBank team up on localized AI infrastructure
Announced July 21, a compact data center pilot for “Stargate”, funded by SoftBank and OpenAI, aims to explore decentralized, energy‑efficient compute hubs by end of 2025. It’s part of SoftBank’s broader “Crystal Land” AI infrastructure vision Crescendo.
UK–OpenAI strategic partnership signed
The UK government signed a memorandum (July 22) with OpenAI, committing to explore hosting data centers, integrate AI tools across public services (justice, defence, education), and collaborate via the UK’s AI Security Institute. Critics warned over privacy risks and growing dependence on a single AI vendor Financial Times.
🇨🇳 China’s AI ecosystem response
Chinese firms form alliances amid U.S. export curbs
At the Shanghai conference, two industry groups launched:
- Model‑Chip Ecosystem Alliance linking LLM developers with domestic chipmakers like Huawei, Biren, Moore Threads.
- Shanghai Chamber AI Committee focusing on industrial AI integration, with firms like SenseTime, Tencent, Baidu.
Highlights included Huawei’s CloudMatrix 384 cluster (384 of its 910C chips), reportedly rivalling top Nvidia hardware. Alibaba demoed Quark AI Glasses for navigation and QR scanning, and Baidu presented “digital human” livestream tech. Tencent unveiled Hunyuan3D World Model for interactive 3D environments.
🧬 Sector Use Cases & Startups
Healthcare & diagnostics AI
- On July 21, researchers unveiled an AI platform that estimates tumor stemness to gauge cancer aggressiveness and recurrence risk, supporting personalized oncology planning Crescendo.
- Portuguese startup MedTiles (July 20) launched an AI platform for faster analysis of dermatology, radiology, and pathology scans, targeting early diagnostics across European clinics € Crescendo.
- Everlab secured a $10 million seed round (July 21) to scale its preventive health platform, offering personalized diagnostics and lifestyle guidance from biomarker data, backed by Pear VC and Quiet Capital.
🧪 Research & Safety Signals
Medical AI ethics study exposes fragility
A July 24 study demonstrated that models like ChatGPT falter on slight tweaks to standard medical‑ethics scenarios, a reminder that generative AI can make surprisingly flawed decisions in sensitive contexts ScienceDaily.
DeepMind releases AlphaGenome (late June context)
While technically from late June, the AlphaGenome model, capable of interpreting non‑coding regions of DNA (the “dark matter”) and predicting expression effects, continues to influence genomic AI discussions into mid‑July.
📈 Market & Investor Climate
Wall Street demands bolder AI strategy from tech giants
With Q2 earnings underway, analysts closely watched Microsoft, Meta, Apple, Amazon, and Alphabet. Concerns center on monetization of AI, capex plans, and hardware readiness. Microsoft’s Azure AI and Meta’s ad‑AI boost are promising, while Apple is urged to catch up. Collectively, these firms may invest $300B+ in AI this year.
Industry trend: companies naming AI a top priority
As of mid‑2025, ~83% of businesses see AI as central to strategy; the AI market is worth ~$391 billion and expected to quintuple over five years, employing ~97 million people globally explodingtopics.com.
📋 Summary & Outlook
This week illustrated how AI is shifting simultaneously on multiple fronts:
- Governance & policy: U.S. and China racing to shape AI norms; international institutions (UN/BRICS) taking stakes.
- Corporate investment: From Google and Meta’s expansion of user‑facing AI services, to Amazon’s secure agent tools, to geopolitical infrastructure plays via SoftBank/OpenAI.
- Domestic resilience: China doubling down on a homegrown AI ecosystem via alliances and hardware-performance showcases.
- Vertical innovation: Healthcare startups carving niches in diagnostics and biomarker‑driven preventive care.
- Ethics & research caution: Studies showing generative AI’s vulnerability in critical decisions, reminding us that oversight is essential.
Together, these developments underscore a pivotal moment: AI is both embedding deeply into business and public infrastructure, and becoming a geopolitical instrument. The companies and nations that align innovation with robust governance are positioning themselves to shape AI’s future, while gaps in safety, access, or infrastructure could create risks or widen the divide.
No responses yet