☕️ Midjourney's V7 Debut Puts it Back in The Limelight After Nearly a Year

Image generation continues to rule the headlines as Midjourney returns to the scene with the alpha release of V7, its first major model release after nearly a year. Other key highlights include:

• Intel and TSMC reach a preliminary agreement to launch a joint venture to run Intel's US chip factories

• Gemini's executive lead, Sissie Hsaio, steps down as Google shuffles around its AI leadership ranks

• Nvidia finds itself facing $16 billion worth of orders from China for its H20 server chips for AI

Join us at AI Tangle as we untangle this week's happenings in AI!

Midjourney releases its next image generation AI model after nearly a year since its last

After nearly a year of absence from the scene, one of the leaders in the GenAI image generation race, Midjourney, has resurfaced to release its first new AI image model in ages. Sticking to its usual naming scheme, Midjourney V7 began rolling out as part of an early alpha-test phase starting on Thursday, a week after OpenAI's image generator in ChatGPT went viral and three weeks after Google began experimenting with its own.

​

Where does Midjourney V7 currently stand, and what has changed?

Built on a "totally different architecture," V7 is designed to better adhere to text prompts, provide even greater image quality with better textures, and more coherent rendering of bodies, hands, and objects. Midjourney V7's alpha release also comes with its "personalization" feature turned on by default, along with offering two modes, Turbo and Relax, as well as a new Draft Mode that speeds up rendering and cuts costs with the downside of lower image quality.

However, it does have some minor downsides, as to even begin using it, users first have to rate around 200 images to build that aforementioned "personalization" profile, if they haven't already, to tune the model each user's individual preferences. Additionally, many of Midjourney's otherwise standard features, such as image upscaling and retexturing, are currently missing from V7, though company CEO David Holz said they are in the works and could arrive within two months.

Intel and TSMC shake hands to launch a joint venture to operate Intel's chip factories in the US

Two giants in the chip world, Intel and TSMC, have reportedly reached a preliminary agreement to launch a joint venture aimed at running Intel's American fabs. According to sources cited by Reuters and The Information, TSMC will hold a 20% stake in the new venture, and though the fate of the remaining 80% is unclear, potential investments from other major US chip designers are being discussed. The arrangement was allegedly influenced by the US government's efforts to stabilize Intel under its IDM 2.0 strategy as it continues to struggle quarter after quarter.

​

Google shuffles around its AI division leadership as Gemini's executive lead, Sissie Hsaio, steps down

Google is reshuffling the top order of its AI division as Sissie Hsaio, the longtime executive lead behind Bard, better known as Gemini these days, steps down from her position. Google Labs vice president Josh Woodward is reported to take over her post, with Google DeepMind CEO Demis Hassabis stating the change will "sharpen our focus on the next evolution of the Gemini app." A separate staff memo details that, as for Hsaio herself, she will be taking a "short break" before returning in a new role.

​

Nvidia faces overwhelming demand in China as tech giants place orders for $16 billion worth of AI chips

Demand for Nvidia's server chips may be at an all-time high as Chinese tech giants have reportedly placed orders for a staggering $16 billion worth of server chips in the first quarter of 2025, says The Information. From Tencent to Alibaba to ByteDance, analysts have seen a surge in orders for Nvidia's H20, the most advanced powerful AI processor the company is allowed to legally sell to the mainland under US export controls, which can at least be partially attributed to DeepSeek's blockbuster debut with its low-cost AI models.

​

Anthropic launches its own take on a subscription plan for education called Claude for Education

As an answer to OpenAI's Edu plan for ChatGPT, Anthropic announced on Wednesday that it is releasing its own take on it, launching a new Claude tier called Claude for Education. The tier comes with a few extra things, such as a "Learning Mode" to enhance critical thinking by asking probing questions and providing research templates. Additionally, it also comes with enterprise-grade security and privacy and is already being adopted by universities like Northeastern University, the London School of Economics and Political Science, and Champlain College.

​

Qualcomm acquires Vingroup's AI research division, VinAI, to bring on board its top talent

Led by Vietnam's wealthiest man, Pham Nhat Vuong, Vingroup has recently sold VinAI, its generative AI division based in Hanoi, to US chip giant Qualcomm for an undisclosed amount, as announced by Qualcomm in a blog post. The deal is set to "bring in high-caliber talent from VinAI" into Qualcomm's fold, including the AI division's CEO and founder Hung Bui, who previously worked at Google DeepMind as a research scientist. Bui later added that he looks forward to contributing to a number of Qualcomm's products, ranging from its software and chips to smartphones and computers.

​

Runway rakes in $308 million in Series D funding off the back of its Gen-4 video generation AI model

Just days after releasing its standard-setting Gen-4 media model, Runway announced that it had raised a sum of $308 million in Series D funding, led by growth equity firm General Atlantic with participation from Nvidia, SoftBank, and others. Runway said in a press release that the extra cash flow will be put toward further developing its AI research and hiring, along with expanding its ts Runway Studios film and animation production arm. According to Crunchbase, a business information provider, Runway has raised just above half a billion in its lifetime, $538 million to be exact.

Lindy - Automate your workflows and help your business grow with Lindy's all-in-one AI agent suite - build agents in minutes to automate lead generation, customer support, outreach, and more.

​Lectura - From detailed video summaries to personalized note-taking, Lectura provides the AI-powered tools needed to stay organized and focused, making it easier to achieve your educational goals.

​Markable - Markable is an AI-enhanced search engine for your bookmarks that quickly and accurately indexes your bookmarks and categorizes them to ensure fast retrieval and a smooth user experience.

DeepMind Predicts AGI by 2030 (4-min read)

​A new 145-page paper by Google DeepMind urges a renewed focus on AI safety planning for the long term as it outlines how it believes the world could begin to prepare for the arrival of AGI as early as 2030.

The Consequences of AI Crawlers on The Web (6-min read)

The need for AI training data has seen many companies send waves of AI bots to crawl the web aggressively, and Wikipedia is feeling the brunt of it as relentless bot scraping has surged its bandwidth use by 50% and caused its operating costs to skyrocket - "Our content is free, our infrastructure is not."