Why $130bn of AI Projects Are Stuck — and It's Not the Chips

The story dominating this edition is one that belongs squarely in the built environment: over $130bn of AI data centre projects have been blocked or delayed in 2026, and the resistance is coming from local communities, councils, and planning authorities, not from a shortage of capital or chips. Alongside that, we look at Anthropic's new Claude Tag agent for Slack, OpenAI's first custom inference chip, the talent war reshaping frontier AI labs, and the geopolitical flashpoint of Anthropic's accusation that Alibaba ran a 28.8 million-exchange distillation attack on Claude. We also examine what the US government's decision to gate OpenAI's latest model releases means for the pace of AI innovation globally.

A Harvard study published this week puts hard numbers on what many in the built environment have been watching develop for months. Active opposition groups fighting data-centre projects have more than doubled, from 396 to 833, now spanning 49 US states. More than 75 projects worth over $130bn have been stalled or cancelled in the first half of 2026 alone. The flashpoints are familiar: noise, water consumption, grid strain, and the visual impact on communities that were never consulted. Seattle paused five projects in a single week. Local bans, water and sewer moratoria, and council removals are spreading across jurisdictions that once competed to attract hyperscale investment.

For project delivery leaders, this is a planning, power, and water story as much as a technology story. The AI infrastructure build-out is now running into the same community opposition, permitting complexity, and infrastructure constraints that have long defined large-scale construction. The firms that will deliver these projects are the ones with deep experience in stakeholder engagement, environmental impact assessment, and phased delivery strategies. The data centre pipeline is enormous; the path through the planning system is not.

Read the full story

Anthropic has replaced its Slack integration with Claude Tag, a meaningfully different product. Tag @Claude in any channel and the model breaks the work into stages, runs connected tools, learns the channel's context over time, and operates asynchronously, with full action logs and admin token controls. Anthropic reports that 65% of its own product team's code now comes from its internal version. For AEC and project delivery teams already running Slack as their coordination layer, this is a practical deployment option worth evaluating now rather than later.

Read more →

OpenAI has unveiled Jalapeño, a reticle-sized inference ASIC co-designed with Broadcom and taken from schematic to fab-ready in roughly nine months, with OpenAI's own models assisting in the design process. Early results show significantly better performance-per-watt than current alternatives, with first deployment targeted by end-2026. The move is a direct effort to reduce dependence on Nvidia and to control the full stack from model to silicon. For organisations planning AI infrastructure procurement, this signals that the hardware economics of inference are about to shift.

Read more →

John Jumper, who shared the 2024 Nobel Prize in Chemistry for his work on AlphaFold2, is leaving DeepMind after nine years to join Anthropic's AI-for-science programme. His departure is part of a broader talent reshuffle this week: Noam Shazeer moved from Google to OpenAI, and Dean Ball also joined OpenAI. The pattern is consistent. Frontier labs are competing aggressively for the researchers who have already demonstrated they can produce work of Nobel-level significance, and Anthropic is winning some of those bids.

Read more →

In a letter to the US Senate Banking Committee, Anthropic alleges that operators linked to Alibaba's Qwen lab ran 28.8 million exchanges through approximately 25,000 fake accounts between 22 April and 5 June, with the intent of extracting Claude's strongest capabilities. Senators are moving to attach sanctions to defence legislation in response. The case raises concrete questions about model security, IP protection, and the geopolitics of frontier AI, and it is likely to shape how enterprise AI contracts are written and audited going forward.

Read more →

OpenAI announced three new models, GPT-5.6 Sol, Terra, and Luna, but complied with a US government request to limit the initial rollout to a small group of trusted partners. The company said it does not believe this kind of access process should become the long-term default, and it is working with the Trump administration to establish a repeatable framework for future releases. The precedent matters: if government pre-clearance becomes standard for frontier model launches, the pace of deployment for enterprise and project delivery teams will be shaped as much by geopolitics as by technical readiness.

Read more →

NVIDIA launches Halos for Robotics, the first full-stack safety system for physical AI. Unveiled at Automate 2026, Halos extends NVIDIA's autonomous-vehicle safety stack to robots operating around people, giving developers a common safety architecture built on IGX Thor and Holoscan Sensor Bridge. Agility, Boston Dynamics, and 40+ partners are on board. Read more →

Micron selects Bechtel for its $100bn New York megafab; Bechtel and NVIDIA modularise the gigawatt data-centre design. Bechtel has been appointed EPC contractor for the first phase of Micron's $100bn Clay, NY memory complex, while a parallel workstream with NVIDIA is turning the gigawatt-scale AI factory reference design into a repeatable, modular build template. Read more →

BMW deploys AEON humanoids in production at Plant Leipzig. BMW is running a pilot with two Hexagon AEON humanoids supporting high-voltage battery assembly at its Leipzig facility, its first humanoid deployment in Europe. The pilot runs to end 2026 and follows an earlier deployment at Spartanburg, providing a concrete data point on physical AI in manufacturing environments. Read more →

How AI and automation are changing financial planning in construction. A practical look at how AI is reshaping cost forecasting, financial planning, and commercial controls in construction, directly relevant to QS and PM workflows. Read more →

New APM research warns that AI adoption in project management must not come at the cost of practitioner well-being. The report, AI's Dual Impact on Project Managers' Well-Being, authored by Dr Xiangming Tao, examines the psychological effects of AI on project managers and calls for governance frameworks that account for human factors alongside productivity gains. Read more →

How AI is being applied to construction tendering. Volve sets out how AI is changing bid analysis, document review, and submission preparation in construction procurement, a concrete use case for commercial teams evaluating AI adoption. Read more →

Five stories worth reading before you get back to delivery:

SpaceX signs up to $6.3bn compute deal with Reflection AI at Colossus: Reflection AI will pay SpaceX roughly $150m per month from July for NVIDIA GB300 capacity at Colossus 2, taking SpaceX's committed external compute revenue above $80bn and confirming Colossus as one of the world's largest third-party AI infrastructure platforms.

OpenAI ships GPT-5.5-Cyber and launches 'Patch the Planet' to fix open-source vulnerabilities: OpenAI and Trail of Bits are pairing AI-assisted vulnerability discovery with expert human review across 30+ open-source projects including cURL, Go, and Python, marking a shift from finding bugs to verified, human-reviewed remediation at scale.

Santander plans £430m of AI-driven cost cuts: Banco Santander is rolling out AI to all 185,000 staff and targeting 40% of annual calls resolved by self-service, providing one of the clearest enterprise-scale data points yet on AI's measurable cost impact.

OpenAI and Anthropic face new AI reality as users shift from 'tokenmaxxing' to efficiency: Companies are pulling back on uncapped AI spend as CFOs demand ROI, with open-source alternatives gaining ground and both frontier labs now racing to IPO before the growth curve flattens.

How agents are transforming work: OpenAI's economic research paper on Codex adoption shows that 80% of individual users are now delegating tasks estimated to take more than 30 minutes of human work and that non-developer adoption has grown 137-fold since August 2025, with legal, finance, and recruiting teams now using Codex as their primary AI tool.

Mistral has released OCR 4, a self-hostable document intelligence model that extracts structured content with paragraph-level bounding boxes, block classification (titles, tables, equations, signatures), and per-word confidence scores across 170 languages. It runs in a single container, keeping document data within your own infrastructure, and is priced at $4 per 1,000 pages via API ($2 with batch discount). For AEC teams handling contracts, drawing registers, RFIs, and O&M manuals where data sovereignty matters, this is a practical drop-in for document-heavy workflows.

Learn more →

Rob Butler shares his journey from project management to data analytics and AI [Podcast]

Build a custom Claude 'voice' skill that writes like you

Stop fighting generic AI output. In Claude's Settings, go to Capabilities, then Skills, and use the skill creator to upload three to five samples of your best writing. Claude will learn your tone and apply it automatically to emails, posts, and reports. The setup takes around 15 minutes. Exclude any AI-assisted samples so the model learns your patterns, not its own.

Source →

Now in its 17th year, the AGC Technology Conference is the construction industry's flagship event for IT and technology leaders. This year's programme in Minneapolis covers AI readiness, cybersecurity, data-driven construction, and tech lessons learned on large projects, with sessions designed for practitioners implementing technology in the field rather than evaluating it from the sidelines. If your organisation is working through AI adoption in a construction context, this is the room to be in.

4–6 August 2026 | Minneapolis Marriott City Center, Minneapolis, MN

Details →

Till next time,

_{All content reflects our personal views and is not intended as professional advice or to represent any organisation.}