Enterprises are no longer asking which AI model wins. They are cutting software budgets to fund AI, consolidating legacy vendors, and renegotiating contracts to free up room. The belief is that AI will lead to smaller, more specialized technology stacks.
We recently hosted a dinner in Washington, D.C. with CIOs and technology leaders from the World Bank, DOJ/Interpol, BAE Systems, Athena Health, Virginia Tech, and Graham Holdings.
These five observations stood out:
1. Enterprises are consolidating legacy vendors and renegotiating contracts to free budget for AI. Fewer tools, more automation.
2. Foundation models have reached escape velocity. Most organizations have not. They still struggle to access, prepare, and govern core structured data at scale, and they increasingly treat the models as replaceable – they don’t want to get locked into one provider. What they guard is everything around the model: data, workflows, governance, security, integrations, and business outcomes.
3. Leaders dislike today’s AI pricing because costs are hard to predict and not tied to outcomes. That is driving interest in open-source models like Llama, Mistral, DeepSeek, and Qwen, as leaders think about cost control, pricing predictability, and vendor independence.
4. Claude remains the preferred model for many, though confidence in Anthropic has been shaken by Mythos and the broader implications of increasingly autonomous AI systems.
5. The CIO is expanding rapidly from a technology to a business transformation leader, being accountable for shaping outcomes.
One theme rose above the rest. AI is spreading among business users faster than enterprises anticipated, and change management is now the bigger challenge. Analysts, PMs, marketers, and operators are adopting AI personally and bringing it to work on their own.
Below, we go deeper on the moves driving the reset: Oracle cut 21,000 jobs and tied the reductions directly to AI-driven productivity; Samsung extended ChatGPT Enterprise and Codex to its entire workforce, putting AI in the hands of far more than just engineers; Anthropic shipped Claude Tag, an always-on AI teammate embedded in Slack with persistent company context; and Mistral’s open OCR 4 underscored the pull toward open, swappable models for core enterprise workflows, fewer tools, more automation, and AI landing on every desk.
Full Weekend Edition below. 👇
Signals Shaping the Future of AI:
Infrastructure
OpenAI unveils Jalapeño, its first custom AI inference chip, co-designed with Broadcom and manufactured by Celestica. The chip is designed to power inference for ChatGPT, Codex, and the API, with deployment expected by the end of 2026. Click here
Reflection AI signs a compute agreement with SpaceX worth up to $6.3 billion through 2029. The deal provides access to NVIDIA GB300 GPUs at the Colossus 2 data center to support large-scale AI model training and inference. Click here
NVIDIA launches Halos, a robotics operating system derived from its autonomous driving software, alongside a humanoid robot safety lab. The platform is designed to support safe deployment of physical AI systems running on NVIDIA IGX Thor hardware. Click here
Microsoft and Chevron sign a 20-year natural gas supply agreement to power a proposed AI data center in West Texas. The facility is expected to become one of the largest data centers in the United States as demand for AI infrastructure continues to grow. Click here
Enterprise
Oracle has reduced its workforce by 21,000 employees over the past year, citing AI-driven productivity gains. The company disclosed the reductions in a regulatory filing as it continues expanding AI across its operations. Click here
Meta reportedly plans to have AI handle 90% of content and advertising review by the end of 2026. The shift would make AI the primary decision-maker for moderation across Facebook and Instagram, significantly reducing reliance on human reviewers. Click here
Samsung deploys ChatGPT Enterprise and Codex across its South Korean workforce and global Device eXperience division. The rollout extends AI coding and workflow automation tools beyond engineers to employees across the organization. Click here
Anthropic launches Claude Tag, an always-on AI teammate for Slack with persistent memory and organizational context. The service allows teams to assign tasks, collaborate through a shared Claude identity, and enable AI agents to work across enterprise workflows with administrator-controlled access. Click here
Air Space Intelligence wins an $875 million, 12-year FAA contract to deploy AI across U.S. air traffic operations. The system will map flight trajectories and identify airspace congestion to improve efficiency and reduce delays. Click here
Eli Lilly plans to invest part of its $7.3 billion cash reserve in an AI marketplace for biotech researchers. The platform will allow scientists to deploy AI tools against Lilly’s proprietary biological datasets for drug discovery. Click here
Capital Flows
Abu Dhabi’s MGX raises a ~$50 billion fund to accelerate AI investments. The sovereign-backed vehicle is one of the largest dedicated AI investment funds announced outside the United States. Click here
Qualcomm confirms its acquisition of Modular, valued at approximately $3.9 billion. The deal adds Modular’s AI software platform, hardware abstraction layer, and Mojo programming language to Qualcomm’s AI portfolio. Click here
Upscale AI raises a $190 million funding extension at a $2 billion valuation to expand AI networking infrastructure. The company builds hardware and software that connect AI chips, memory, and storage, with the new funding supporting the deployment of its AI-native networking platform. Click here
Groq raises $650 million to expand custom AI inference infrastructure. The company plans to build 200 megawatts of inference capacity by the end of 2027 following its partnership with NVIDIA. Click here
Agility Robotics agrees to go public through a $2.5 billion SPAC merger. The transaction includes a PIPE financing of more than $200 million led by Foxconn. Click here
General Intuition raises $320 million at a $2.3 billion valuation to develop AI agents that transfer skills from video games to real-world robotics. The company is using gameplay data to train generalized world models for robotics, automation, and embodied AI, with much of the funding dedicated to expanding compute infrastructure. Click here
Hang Ten Systems launches with a $32 million seed round to build AI-native enterprise software services. Founded by former Infosys CEO Vishal Sikka, the company uses agentic AI to automate software development, modernization, and enterprise IT delivery. Click here
Runlayer raises a $30 million Series A to expand its infrastructure platform for enterprise AI agents. The company provides tools to monitor, govern, and orchestrate AI agents in production environments. Click here
Research
OpenAI reports GPT-5.5-Cyber autonomously discovered dozens of security vulnerabilities across open-source software. The model identified kernel exploits, privilege escalation techniques, and vulnerabilities that were later patched by software maintainers. Click here
Unitree and Hugging Face release HIW-500, the largest open dataset for humanoid robot training to date. The dataset includes more than 500 hours and 23,000 episodes of teleoperated household robotics data. Click here
Two key Google Gemini researchers are reportedly leaving to join Anthropic. The departures of Jonas Adler and Alexander Pritzel add to a growing wave of high-profile AI talent moving from Google to frontier AI startups amid intensifying competition for model researchers. Click here
Policy
OpenAI is reportedly delaying the broader release of GPT-5.6 at the request of the U.S. government. The model will first launch through a limited preview with selected partners while federal agencies review access before a wider rollout. Click here
The Trump administration is urging Meta to participate in its voluntary AI model review program ahead of major releases. OpenAI, Anthropic, Google, xAI, and Microsoft have already agreed to the government’s pre-release evaluation process. Click here
CISA, the NSA, and Five Eyes partners warn that frontier AI models will soon accelerate cyberattacks. The joint advisory urges organizations to strengthen cybersecurity defenses, patch systems faster, and improve identity protections. Click here
OpenAI, Anthropic, Amazon, Microsoft, and other major employers back a new $500 million initiative to help workers adapt to AI-driven job disruption. Led by former U.S. Commerce Secretary Gina Raimondo, the nonprofit will partner with states and employers to fund workforce training, reskilling, and AI transition programs. Click here
Global AI Strategy
China’s ARM-based LineShine supercomputer debuts at the top of the June 2026 TOP500 rankings. Built entirely with domestically developed processors, interconnects, and operating systems, it surpasses the U.S. El Capitan system to become the world’s fastest supercomputer. Click here
The European Union and Germany join the U.S.- led Pax Silica initiative to strengthen AI and semiconductor supply chain coordination. The coalition now includes 24 participants working together on critical AI infrastructure and technology supply chains. Click here
Anthropic accuses Alibaba of using Claude for large-scale “adversarial distillation” in a letter to U.S. officials. Anthropic alleges Alibaba accessed Claude nearly 29 million times through about 25,000 accounts between April and June, as concerns over AI model replication and intellectual property intensify. Click here
JD.com founder Richard Liu says robots will eventually replace the company’s 700,000 delivery workers. He said employees displaced by automation would be retrained for roles such as robot maintenance. Click here
Talent Signals
Each week, we spotlight key roles tied to the themes shaping this week’s AI headlines, connecting talent to the companies driving the news.
UpscaleAIis building the networking layer for the AI era, helping data centers scale AI workloads across high-performance Ethernet infrastructure from scale-up to scale-out. Following its Mayfield-led investment and NVIDIA partnership around open AI networking, Upscale AI is hiring across engineering, systems, networking, and go-to-market roles. Open roles are listed on its careers page. Click here
Replicate is building infrastructure that makes it easier for developers to run, fine-tune, and deploy open-source AI models through simple APIs. As more teams bring multimodal and generative AI models into production, Replicate is hiring across engineering, product, infrastructure, design, and go-to-market roles. Open roles are listed on its careers page. Click here
Anyscale is building the AI compute platform behind Ray, helping teams scale, deploy, and manage AI and Python workloads across training, tuning, inference, and data processing. As enterprises look to run AI workloads more efficiently across distributed infrastructure, Anyscale is hiring across engineering, product, customer engineering, sales, and go-to-market roles. Open roles are listed on its careers page. Click here
You can see all the opportunities at Mayfield-backed AI companies here.
Social Signals
The most important conversations in AI are unfolding across social media, where top voices are shaping the next wave of signals and strategy. Here are some of the top social signals and their takes from the past week.
Nikesh Arora (Click here) — “The AI business model trap: LLMs want cash flow to fund the race to AGI. Free consumer AI feeds post-training needs, but the monetization challenge is being pushed to enterprises. If you want to win enterprise, you should be forward pricing tokens. The cheaper the tokens for enterprises, the more experimentation and workflow reinvention you unlock.” In a post viewed 2.8M+ times, Arora argues that the next competitive battle in AI will center on enterprise economics, not model quality. He suggests lower token costs, better enterprise context, and stronger workflow integration will determine which AI platforms become embedded across large organizations.
Satya Nadella (Click here) — “Every company is going to have to build what I think of as human capital and token capital. Human capital does not become less valuable as token capital grows. It only becomes more valuable. The real opportunity is not in picking the best model but in building a learning loop where human capital and token capital compound.” In a post viewed more than 66 million times, Nadella argues that the future advantage for enterprises will come from owning their learning loop, turning institutional knowledge, workflows, and judgment into AI systems that improve over time. He suggests that durable value will accrue to companies that compound their own expertise rather than simply adopting the latest frontier model.
Paul Graham (Click here) — “College students use AI to do most of their writing. An increasing number of professors secretly use it for grading. In the limit case, AIs do all the work, and all the humans do is transmit what they create. A good compiler would recognize this as dead code and remove it.” Graham uses a programming analogy to question what happens when AI intermediates both the creation and evaluation of knowledge. The post reflects a growing debate over how AI will reshape education, knowledge work, and the role humans play as these systems become more deeply embedded in everyday workflows.
