Every region moved this week — chips, capital, labour, and the rulebook all at once. Here's what actually happened, verified, region by region, plus our read on who's winning the race and the one gap almost nobody has closed yet: governance.
Key Takeaways
- UAE: Washington eased export rules, letting the UAE government and approved companies buy advanced Nvidia chips and AI servers without individual licences. Abu Dhabi is consolidating AI policy under one new federal AI and Data Authority.
- Saudi Arabia: HUMAIN has committed 50MW of compute to Cohere for Arabic-language models and is now also building a data centre with xAI.
- US: OpenAI has pitched Washington a 5% equity stake (~$42.6B at its $852B valuation). No deal is signed; congressional approval would likely be required.
- Europe: The Cloud and AI Development Act publishes in the Official Journal on 15 July. Separately, the EU formally agreed in May to push AI Act high-risk deadlines back to December 2027 (stand-alone systems) and August 2028 (embedded systems).
- China: Xi Jinping delivers the keynote at Shanghai's World AI Conference (17–20 July) — his first appearance at the event since 2018 — alongside a High-Level Meeting on Global AI Governance.
- Revenue: Anthropic's annualised revenue run rate is reportedly ~$69B, up from ~$47B in May. TSMC posted its largest monthly revenue in company history on AI chip demand.
- Governance gap: Only ~8% of organisations globally have a comprehensive AI governance framework, even though 88% are already using AI across business functions.
UAE & GCC — access and architecture
US export rules now let the UAE government and vetted companies import advanced Nvidia AI chips and servers without seeking individual licences each time. In parallel, the UAE is merging the Office of Artificial Intelligence, the Digital Economy office, and the UAE Data Office into one federal AI and Data Authority, giving the country a single front door for AI governance as it scales.
Saudi Arabia isn't standing still either. HUMAIN, the PIF-owned AI champion, has moved from its Cohere compute-sharing deal (50MW earmarked for Arabic-language models) straight into a data-centre partnership with xAI — locking in multiple frontier-lab relationships rather than betting on one.
United States — who owns the upside, who owns the risk
OpenAI's proposed 5% government equity stake was reportedly discussed with Commerce Secretary Lutnick and Treasury Secretary Bessent, modelled on Alaska's Permanent Fund. It remains unconfirmed and would likely need congressional approval — treat it as a signal of where political pressure is heading, not a done deal.
TSMC's June revenue was its largest monthly total in company history, up 67.9% year-on-year, with advanced chip lines reportedly sold out through year-end — a reminder that the AI boom still runs through physical manufacturing capacity. Meanwhile, in the Bronx, Montefiore Medical Center laid off 12 utilization-review nurses and shifted the work to AI software from Datavant; the nurses' union, NYSNA, alleges this breaches AI-protection language won after this year's strike, and filed a grievance on June 1.
"If AI is touching roles covered by a union contract, get legal and HR aligned on the 'diminishment' clause before you announce anything. Montefiore is the cautionary tale every company deploying AI near organised labour should be reading closely."
— Lisa Warren, Founder & CEO, Neural Horizons AI
Europe — the rulebook moves in two directions at once
The Cloud and AI Development Act publishes in the Official Journal on 15 July, introducing new obligations for cloud providers and AI developers, with the first cloud-sovereignty tier applying from February 2028 and the strictest tier from August 2029. At the same time, the EU has formally agreed to delay the separate AI Act's high-risk obligations — stand-alone Annex III systems (recruitment, credit scoring, education, border control) now need to comply by December 2027, and AI embedded in regulated products (medical devices, machinery, vehicles) by August 2028, rather than this August as originally planned.
One caveat worth flagging: the delay only takes legal effect once formally published in the Official Journal. If that slips, the original August 2026 deadline could technically still apply — worth tracking if you operate in the EU.
China — claiming the room
Xi Jinping will deliver the keynote at Shanghai's World AI Conference (17–20 July) — the first time China's top leader has personally opened the event since it began in 2018, a clear signal of how central AI has become to national strategy amid intensifying US-China tech rivalry. The conference runs alongside a High-Level Meeting on Global AI Governance, with organisers expecting 1,400+ guests, 12 government ministries, and 300+ product debuts.
Who's winning the AI race
On raw model capability, the US still leads, and the chip gap is the reason why — American and allied manufacturing capacity for advanced AI processors runs an estimated 35–38x China's, quality-adjusted. That's a structural advantage that doesn't disappear quickly.
But China is closing the performance gap faster than almost anyone predicted — from a ~30-point benchmark gap in 2023 to roughly 2.7% by March 2026. DeepSeek-V4 reportedly matches frontier US benchmark scores at a fraction of the API cost, and it's the clearest signal of a "good enough, cheaper, wider distribution" strategy working.
Our read: the US is winning the sprint, China is building for the marathon — and the Gulf isn't racing either of them. It's positioning itself as the infrastructure and capital layer both sides need. The UAE's eased chip access and Saudi's compute partnerships with Cohere and xAI are a bet that whoever ends up ahead will still need somewhere to build.
Companies leading on upskilling & retraining
The AI race isn't only being run in labs — it's being run in HR departments:
- Microsoft — $4B Elevate initiative, targeting AI credentials for 20 million people within two years, with LinkedIn Learning, GitHub, and Code.org.
- Accenture — $1.2B+ invested in AI training; AI proficiency now a formal promotion requirement across 700,000+ employees; ~40% of the workforce uses AI tools daily on client work.
- Amazon — $700M committed to upskill 100,000 employees in AI and machine learning.
- Meta — $115M for training tradespeople for data-centre construction, the largest private job-guarantee training program in US history.
- JPMorgan Chase — mandatory AI training for all new hires in asset management.
- RAISE US — a coalition backed by OpenAI, Anthropic, Microsoft, and Amazon has secured over $500M toward a $1B fund to retrain US workers displaced by AI.
Governance: the part nobody can afford to skip
Three frameworks are converging as the global baseline: the NIST AI Risk Management Framework (Govern, Map, Measure, Manage) dominates US enterprise practice; ISO/IEC 42001 offers a certifiable management standard already adopted by Microsoft and Google; and the EU AI Act, even with its deadlines pushed to December 2027 and August 2028, still sets the direction of travel for risk-based regulation globally. Layer on the UAE's new federal AI and Data Authority, Singapore's world-first governance framework for agentic AI, and whatever China puts forward at this week's WAIC, and the picture is clear: governance is becoming the actual terrain of competition between countries.
What to do, concretely
- Stand up an AI governance owner or committee now, even if it's small.
- Maintain a live inventory of every AI system in use, including vendor-procured ones.
- Map each system against every jurisdiction you operate in — UAE, US, EU, and China are now four distinct rulebooks.
- Build an incident and grievance process before you need one — see Montefiore, above.
Lisa Warren: what's really shaping AI right now
"The gap that matters most this week isn't between the best model and the second-best model. It's the gap between the 88% of organisations already using AI and the 8% who've actually governed it. That's where the next AI failure comes from — not a weaker model, an ungoverned one."
— Lisa Warren, Founder & CEO, Neural Horizons AIWhat leaders should take away this week
The first takeaway is that infrastructure access is shifting fast — the UAE's eased chip rules and Saudi's multi-lab compute deals mean the Gulf is a genuinely live build zone right now, not a future one.
The second is that labour and AI deployment are now a compliance issue, not just an HR one — Montefiore shows what happens when contract language and AI rollout collide without a process in place.
The third is that Europe's regulatory relief is real but temporary — the extra runway on AI Act deadlines should be used to get compliant, not to deprioritise it.
And finally, governance is the actual competitive edge — not model access, not compute. The organisations and countries that operationalise governance fastest will set the template everyone else copies.