Hello Project AI enthusiasts,
This week's developments signal a maturation in how AI is deployed and constrained. The global data centre build-out faces its first statutory pause in New York, setting a template for how jurisdictions might manage grid and community pressures. Meanwhile, the frontier model race tightens as Moonshot's Kimi K3 closes the gap with US labs, and Anthropic moves toward monetisation with a services joint venture and metered billing. We also look at the growing debate over data sovereignty, sparked by Satya Nadella's warnings and the Grok Build leak, and OpenAI's push into physical hardware.
📌 Featured This Week
The global data centre boom just hit its first hard stop
The world's data-centre build-out met its first statutory pause. New York's Governor signed Executive Order 62 on 14 July, introducing a 12-month moratorium on environmental permits for new hyperscale data centres while the state develops impact rules, a Grid Acceleration Fund and community investment requirements. Over 12 proposed projects exceeding 50MW are affected, drawing opposition from construction unions.
This is a critical global signal. Grid, water and community pressures are building in the UK, Ireland and Singapore, but New York is the first jurisdiction to turn this backlash into policy. It serves as a live case study and a likely template that other regions will copy as they balance AI infrastructure demands against local resources.
🔗 Editor’s Picks
This week's must-read stories on AI, project delivery, and infrastructure:
Moonshot's Kimi K3 launches, closing the frontier gap with US labs
Moonshot released Kimi K3 on 16 July: a 2.8-trillion-parameter open model (weights due 27 July) with 1M-token context that developers preferred over Fable 5 and GPT-5.6 Sol for coding in blind Arena tests, despite scoring 57 on the Artificial Analysis index behind only Fable 5 (60) and Sol (59) - at $3/$15 per million tokens. For AEC firms it's 2026's DeepSeek moment: near-frontier capability at commodity pricing, days before Beijing's World AI Conference.
Anthropic monetises: IPO bankers, a $1.5bn services JV, and Fable 5 goes on the meter
Three connected moves in one week: Goldman, Morgan Stanley and JPMorgan are lining up investor meetings for a potential Anthropic IPO as soon as October (after the $965bn-valuation round); Anthropic launched 'Ode with Anthropic', a $1.5bn AI-deployment services firm with Blackstone and Hellman & Friedman; and Fable 5 moves to metered billing from 19 July at roughly 2x Opus pricing. The signal for buyers: implementation services and usage-based pricing are where frontier AI is being monetised.
Who owns your AI exhaust? Nadella's warning and the Grok Build leak
Satya Nadella's 'Reverse Information Paradox' post argues firms pay for AI twice - in money and in proprietary knowledge flowing one way into model providers - echoing Palantir's Karp and Benioff's refusal to let labs train on Salesforce customer data. The warning got concrete the same week: wire-level analysis showed xAI's Grok Build CLI uploading entire repositories (including unread files and .env secrets) to a Google Cloud bucket regardless of the privacy toggle; Musk promised deletion. For AEC firms feeding project data into models, data governance just became a board issue.
OpenAI's first hardware device is reportedly a screenless speaker that can move
OpenAI's hardware week: Bloomberg reports its first device (with Jony Ive) is a screenless, movable smart speaker with cameras, sensors and GPT-Live voice, expected ~2027 - and it separately shipped Codex Micro, a $230 keypad with Work Louder for controlling Codex agents (agent-status keys, reasoning dial, joystick), which sold out. OpenAI goes physical while fighting Apple's trade-secrets suit.
🔨 AI in AEC and Projects
Here are some intriguing reads, specifically for AEC and project delivery professionals:
Nvidia and Japan launch a 140MW national 'AI factory': Japan's government and industrial leaders launched a Vera Rubin-based national AI infrastructure programme with Nvidia - 27,500 Rubin GPUs aimed at physical AI, robotics and open multimodal models.
TSMC adds $100bn to US investment for Arizona fabs and packaging: TSMC committed an additional $100bn at its Q2 earnings (profit +77%) for four-plus 2nm-and-below Arizona fabs and advanced packaging capacity, taking its total US commitment to $265bn.
Uber's forward-deployed AI pods closed its ROI gap in 60 days: After May headlines said Uber had burned its 2026 AI budget, its CTO paired top AI engineers with domain experts in 16 two-week 'pods' across finance, legal, ops and support.
Meta's Hyperion data centre grows to $50bn and 5GW in Louisiana: Meta expanded its Hyperion campus in Richland Parish, Louisiana from 2GW/$27bn to 5GW and $50bn+, with full build-out running to about 2032.
Nemetschek takes stake in data exchange platform Dawex: Nemetschek has taken a stake in data-exchange platform Dawex, positioning governed data exchange as the foundation for agentic AI across the building lifecycle.
Nemetschek Arabia launches Bluebeam Max across the Middle East: Nemetschek Arabia announced the regional availability of Bluebeam Max across the Middle East, bringing advanced automation, intelligent markups and AI-powered reviews to AEC professionals.
Rocketlane secures Atlassian Ventures investment: Atlassian Ventures has invested in Rocketlane, the professional-services delivery platform, as AI reshapes how client work is scoped, staffed and delivered.
🧭 Curated Links
Five stories worth reading before you get back to delivery:
SK Hynix's record $26.5bn US IPO - and a memory shortage forecast to 2030
SK Hynix raised $26.5bn in the biggest foreign IPO in US history, whilst its CEO warned the HBM memory crunch will peak in 2027 and last into 2030. Memory, not GPUs, may be the binding constraint on the data-centre pipeline, with knock-on price rises already hitting hardware buyers.
IBM's worst day on record: -25% as clients shift spend to AI hardware
IBM lost a quarter of its market value in a day after warning that customers are reprioritising budgets toward AI compute, dragging Oracle, Salesforce and ServiceNow with it. The capex reprioritisation squeezing enterprise software is the same force inflating data-centre construction demand.
29 countries sign agreement to establish a global AI cooperation body
Twenty-nine countries signed an agreement establishing a new international AI governance body, announced ahead of Shanghai's World AI Conference. Early-stage, but it's the first multilateral structure with signatures rather than communiqués – one to track for future compliance obligations.
Suno hack reveals scraped training data from YouTube, Deezer and Genius
Leaked source code given to 404 Media indicates AI music generator Suno scraped YouTube Music, Deezer, Genius, podcasts and stock-music libraries for training data. A cautionary datapoint for anyone building on generative vendors: provenance risk is real and discoverable.
We Must Act Now: 200+ economists and 16 Nobel laureates put AI's job shock on the clock
A Stanford-organised statement signed by 200+ researchers, economists and lab leaders (including Jeff Dean and Jack Clark) warns societies may have only a few years to adapt labour policy before AI reshapes work faster than the Industrial Revolution. For an industry already short of skilled trades, the call for retraining incentives and safety nets lands directly on AEC workforce planning.
🔧 Tool of the Week
Gemini Notebook (formerly NotebookLM)
NotebookLM has been renamed Gemini Notebook (16 July): existing notebooks and features carry over, and it gains native code execution for in-notebook data analysis, with Gemini 3.5 and an Antigravity upgrade coming to AI Pro. With 30M+ users and 600k organisations, it remains the easiest drop-in for synthesising project documents, specs and reports – now with analysis built in.
🏛️ Governance
Illinois enacts first US law mandating annual AI safety audits
Governor Pritzker signed the AI Safety Measures Act (SB 315) on 6 July - the first US state law requiring annual independent third-party safety audits for frontier AI developers with over $500m revenue, plus 72-hour/24-hour incident reporting and fines up to $3m. It takes effect 1 January 2028, giving any firm building on frontier models a concrete compliance date to plan against.
DeepMind CEO calls for an independent standards body to regulate frontier AI
Demis Hassabis proposed a FINRA-style industry-funded AI standards body: a voluntary 30-day pre-release model review for cyber/bio dangers, moving to mandatory operational by year-end – publicly backed by Altman and Nadella.
🧑🏫 Training
UST to train 20,000 engineers, architects and consultants on Claude
Anthropic's newest Global Premier Partner UST is training 20,000 engineers, architects and consultants on Claude and embedding it in its engineering platforms across semiconductors, telecom and manufacturing (iDEC chip validation, IntelliOps, FinX). One of the largest single AI-upskilling commitments yet in engineering services - a template for what workforce AI enablement looks like at delivery-firm scale.
👀 Also This Week
Alex Luketa from Xerini shares insights on AI, data management, and the future of technology [Podcast]
💡 Quick Win
Point ChatGPT Work at a messy project folder, get a status dashboard
Futurepedia's walkthrough shows the 6-step workflow: point ChatGPT Work at a folder of invoices, notes and PDFs, assign an outcome in plain English, review the first pass, turn the resulting spreadsheet into a hosted Site dashboard, and save the whole thing as a scheduled, reusable Skill. Maps one-to-one onto project-folder chaos: drawings register, cost reports, meeting notes in, live project status out.
📰 Event of The Week
AGC Construction Demo Day – Pre-Construction Technology (22 July, virtual)
AGC of America runs a free virtual Construction Demo Day on pre-construction technology on Wednesday, 22 July (10:00-16:00 CT), showcasing estimating, takeoff and preconstruction AI tools. Book-ahead option: AGC Technology Conference 2026, 4-6 August, Minneapolis – the main US construction-tech event of the summer.
📊 Hiring: Power BI & Project Data Reporting Specialists
Pace Global is recruiting data reporting talent for major project delivery programmes across London, Manchester and Birmingham.
If you know your way around Power BI and you've worked on the data behind real projects — schedules, cost, performance — they want to talk to you.
£45k–£70k · permanent · project controls
See the roles → https://www.paceglobal.org/hot-jobs
Till next time,
Project Flux
All content reflects our personal views and is not intended as professional advice or to represent any organisation.