Introduction
In late July 2025 the artificial intelligence landscape witnessed several landmark announcements. Privacy advocates cheered the launch of Proton’s end‑to‑end encrypted assistant Lumo, cloud infrastructure watchers tracked OpenAI and Oracle’s massive 4.5 GW data‑center expansion, policy analysts pored over the Trump administration’s AI action plan, while hardware enthusiasts debated Huawei’s new CloudMatrix 384 system. This article brings together the key details and context behind each story and places them in the wider AI trajectory.
Proton unveils privacy‑first AI assistant Lumo
Proton, the Swiss company known for its secure email and VPN services, launched Lumo on July 23. According to MacRumors, Proton built Lumo as an alternative to mainstream chatbots and large language models that harvest user data【937228695729200†L146-L160】. Lumo can perform typical assistant tasks like document analysis, email rewriting and code generation【937228695729200†L146-L154】. What sets it apart is that Proton promises not to log or use user queries to train its models; conversations are encrypted end‑to‑end using the same technology underpinning Proton’s other services【937228695729200†L159-L173】.
Lumo runs on open‑source models such as Mistral’s Nemo and Nvidia’s OpenHands 32B and operates from Proton‑controlled European data centres【937228695729200†L159-L173】. A “Ghost mode” deletes chats when the window is closed, and optional web search uses privacy‑friendly engines【937228695729200†L165-L168】. Lumo is free via the web and has iOS and Android apps; paying $12.99/month for Lumo Plus removes limits on chats and file uploads【937228695729200†L170-L179】. Proton CEO Andy Yen said the company wanted to ensure AI doesn’t become a surveillance tool and emphasised building an assistant that protects users【937228695729200†L175-L180】.
OpenAI and Oracle add 4.5 GW of capacity to Stargate
On July 22, OpenAI and Oracle announced that their Stargate partnership will develop an additional 4.5 gigawatts of data‑center capacity【617924818770170†L163-L169】. Reuters reported that the expansion is part of the up to $500 billion Stargate project, which includes SoftBank and aims to keep the U.S. ahead in AI computing【617924818770170†L163-L177】. The new capacity will be spread across facilities whose locations and funding details have not been disclosed【617924818770170†L168-L170】.
With more than 5 GW now under development, the Stargate network will rely on over 2 million chips to power generative AI services like ChatGPT and Copilot【617924818770170†L190-L193】. The announcement underscores how data‑center infrastructure has become a strategic asset in the global AI race, though analysts have questioned whether the consortium can secure the hundreds of billions of dollars required【617924818770170†L196-L207】.
Trump administration’s AI action plan
During the “Winning the AI Race” summit in Washington on July 23, the Trump administration released a sweeping AI blueprint intended to expand AI exports to allies and loosen environmental regulations【265178656263048†L155-L175】. The plan contains roughly 90 recommendations and calls for a single federal standard to replace varied state regulations【265178656263048†L172-L178】. It seeks to export U.S. AI hardware and software to partner nations and crack down on state laws perceived as overly restrictive【265178656263048†L172-L175】.
Michael Kratsios, head of the Office of Science and Technology Policy, said Commerce and State Departments would work with industry to deliver “secure full‑stack AI export packages”【265178656263048†L180-L184】. The blueprint includes three executive orders: loosening environmental rules to fast‑track data‑center construction, establishing rules for chip exports and limiting perceived political bias in AI【265178656263048†L194-L199】. The strategy marks a departure from former President Biden’s cautious “high fence” export control approach and reflects the administration’s belief that winning the AI race requires aggressive expansion【265178656263048†L195-L207】.
Huawei’s CloudMatrix 384 system
At the World Artificial Intelligence Conference in Shanghai on July 26, Huawei showcased its CloudMatrix 384 AI computing system, which some analysts view as a direct competitor to Nvidia’s top‑of‑the‑line GB200 NVL72 system【40207786345666†L172-L176】. Reuters notes that CloudMatrix 384 uses 384 of Huawei’s 910C chips and is reported to outperform Nvidia’s system on some benchmarks【40207786345666†L172-L194】. Huawei developed a supernode architecture to connect the chips at high speed, compensating for weaker individual chip performance through system‑level innovations【40207786345666†L196-L201】.
The system drew significant attention from attendees and highlights China’s push to develop domestic AI hardware despite U.S. export restrictions【40207786345666†L187-L193】. Nvidia CEO Jensen Huang remarked in May that Huawei had been “moving quite fast”【40207786345666†L187-L190】. Analysts see CloudMatrix 384 as evidence that Chinese tech firms are narrowing the gap with American chipmakers and diversifying the global supply of AI computing platforms.
Broader AI trends and context
These announcements arrive amid broader trends documented in Stanford’s 2025 AI Index. The report shows that AI performance on benchmarks improved dramatically between 2023 and 2024; scores on the MMLU, GPQA and SWE‑bench benchmarks jumped by up to 67 percentage points【556274532921870†L73-L80】. Generative AI attracted $33.9 billion in private investment in 2024 and 78% of surveyed organizations reported using AI【556274532921870†L95-L103】. The U.S. produced 40 notable AI models in 2024 versus China’s 15【556274532921870†L108-L116】, but Chinese models are closing the quality gap【556274532921870†L108-L116】. Optimism about AI is growing globally, though regional divides remain; 83% of Chinese respondents say AI is more beneficial than harmful compared with only 39% in the United States【556274532921870†L136-L141】.
At the same time, the responsible AI ecosystem is developing unevenly: new benchmarks like HELM Safety and FACTS are emerging but major companies seldom conduct standardized responsible AI evaluations【556274532921870†L122-L133】. Governments are increasing regulation and investment; U.S. federal agencies issued 59 AI‑related regulations in 2024, more than double the previous year, while countries like Canada, France, and Saudi Arabia announced multibillion‑dollar AI initiatives【556274532921870†L160-L168】.
Conclusion
July 2025 illustrates the multifaceted evolution of artificial intelligence. Consumer‑facing tools like Proton’s Lumo demonstrate growing demand for privacy‑respecting AI assistants. Infrastructure expansions such as OpenAI and Oracle’s Stargate highlight the enormous resources required to support generative AI. The Trump administration’s action plan shows how AI is now a central element of U.S. policy and geopolitical strategy, and China’s unveiling of CloudMatrix 384 underscores the intensifying hardware race. As AI performance climbs and investment soars, thoughtful governance and responsible deployment will be crucial to ensuring these advances benefit society.