Welcome back. Microsoft just laid off 9,000 employees despite posting an 18% jump in profits to $25.8 billion last quarter, bringing total job cuts this year to 15,000. CEO Satya Nadella recently revealed that 30% of Microsoft's code is now written by AI tools, and while the company won't directly say AI is replacing workers, the timing speaks volumes. Apparently, record profits and AI productivity gains mean fewer humans are needed to achieve them.

Have a great 4th of July! We’ll see you on Monday.

In today’s newsletter:

• 🖌️ AI for Good: AI finds paint formula that keeps buildings cool

• 💻 Microsoft scales back AI chip ambitions to overcome delays

• 📹 AI VTubers are now raking in millions on YouTube

🖌️ AI for Good: AI finds paint formula that keeps buildings cool

On sweltering summer afternoons in cities like Rio or Bangkok, the sun bakes rooftops and buildings, raising urban temperatures by several degrees. But a new paint developed using AI may help turn down the heat — and the energy bills.

What happened: Researchers from the University of Texas at Austin, the University of Shanghai Jiao Tong, the National University of Singapore and Umea University in Sweden have designed a new machine learning-based approach for creating complex, three-dimensional thermal meta-emitters that can cool buildings by 5 to 20 degrees Celsius compared to normal paint.

• The team developed more than 1,500 different materials capable of emitting heat at various levels using machine learning algorithms to predict optimal chemical structures and material compositions.

• When tested on model houses, surfaces coated with the AI-designed paint remained 5 to 20 degrees Celsius cooler than those with regular white and grey paints after four hours of direct midday sunlight.

• According to the researchers, this level of cooling can save approximately 15,800 kilowatt-hours per year in an apartment building in a hot climate. A typical air conditioner uses approximately 1,500 kilowatt-hours annually.

Why it matters: The breakthrough addresses a major bottleneck in materials science where traditional trial-and-error approaches have been "slow and labor-intensive," according to Yuebing Zheng, a co-leader on the study published in Nature. Kan Yao, a co-author and research fellow in Zheng's group, noted that "the unique spectral requirements of thermal management make it particularly suitable for designing high-performance thermal emitters" using machine learning. With 17% of all residential electricity use in the U.S. going toward air conditioning, AI-designed cooling materials could deliver substantial energy savings while helping cities adapt to rising temperatures.

🗣️ AI voices with personality

Rime offers the most realistic and expressive AI voices on the market today, creating agentic experiences that sound like everyday people, not robots or voice actors — while driving double digit increases in conversion for brands like Dominos.

Use Rime’s platform to:

• Deliver natural voice AI agents that laugh, breathe, sigh, and sound human

• Easily integrate multilingual text-to-speech (TTS) via API or on-prem

• Start building today with a generous free tier

Try Rime’s live chat (or +1-662-727-8948) and start creating agentic experiences with personality today.

💻 Microsoft scales back AI chip ambitions to overcome delays

Microsoft is revising its roadmap for internally developed AI server chips and will focus on releasing less ambitious designs through 2028, aiming to overcome the issues that have caused delays in development.

Here's what's changing: Microsoft executives told engineers in its silicon team about the new plans in a meeting last week, according to The Information. The decision comes after Microsoft had to push back the release of its latest-generation AI chip, Maia 200, from 2025 to 2026.

• The company launched its first AI chip, Maia 100, in late 2023 and immediately began working on three successors — codenamed Braga, Braga-R and Clea — due for release in 2025, 2026 and 2027, respectively.

• Braga's design was only completed in June, missing a year-end deadline by around six months.

• Microsoft is now considering developing an intermediary chip for release in 2027 that will sit between Braga and Braga-R in terms of performance, likely called Maia 280.

• The release of Microsoft's third-generation AI chip, Clea, has been pushed beyond 2028.

Why it matters: Like Google and Amazon, Microsoft designs its own chips to power AI services, such as OpenAI's ChatGPT, in hopes of creating an alternative to Nvidia's chips, which currently dominate the market. Microsoft was Nvidia's largest customer by revenue last year and spends billions of dollars annually buying Nvidia AI chips for its Azure cloud service.

Microsoft executives believe the Maia 280 approach will still deliver between 20% and 30% better performance per watt compared to the chips that Nvidia will release in 2027.

Missed Plaid Effects? Your competitors didn’t.

Plaid just launched a wave of new product updates—and the smartest teams are already moving.

From their most powerful fraud model yet to real-time cash flow insights and AI-powered tools, this year’s updates are built to give you an edge. 

Catch up now to see what’s possible. 

• Stop fraud at every step of the user journey

• Lend smarter with cash flow insights

• Speed up your integration with new AI tools

It’s all on demand: Zach Perret’s keynote, product deep dives with Plaid experts, and a fireside chat with Rocket’s CEO on the future of mortgage lending. 

You’ve got the link—now use it. 

👉 Watch now

• Perplexity introduces its most powerful subscription tier yet

• ChatGPT is sending more traffic to news sites

• OpenAI signs $30B data centre deal with Oracle

• AI could help cut global carbon emissions, study suggests

• Meta says Zuckerberg didn’t offer $300M to lure top AI talent

• Clothoff is a dubious “nudify” app that has millions of visitors

• Meta seeks $29B from private credit giants to fund AI data centres

• ASML, SAP, Mistral Ask EU to delay start of AI Act rules

• The hidden labor that makes AI work

• OpenAI says Robinhood’s tokens aren’t equity in the company

AI Engineers Ready to Join Your Team

Brought to you by Athyna.

• Pedro, São Paulo: Senior ML Ops engineer with 7+ years’ experience in AWS and Python. – $45/h

• Felipe, Buenos Aires: Senior AI specialist with 6+ years in NLP, and 3+ years in LLMs and Generative AI. – $50/h

• Maria, Rio de Janeiro: Senior computer vision engineer with 8+ years using TensorFlow and image recognition tools. – $43/h

Need a different role? Let’s talk!

• PlayHT: Generate realistic AI voice overs from text using cloned or stock voices

• Durable: Build an AI-generated website in 30 seconds

• Supernormal: Automatically generate meeting notes with AI and sync them to your tools

📹 AI VTubers are now raking in millions on YouTube

AI VTubers are now raking in millions on YouTube

Bloo has blue hair, animated eyes, and a fan base of more than 2.5 million subscribers. He plays Grand Theft Auto, Roblox and Minecraft. His videos have garnered over 700 million views. But Bloo is not a person. He is a fully AI-powered virtual YouTuber.

Bloo was created by Jordi van den Bussche, a long-time YouTuber known as Kwebbelkop. After years of struggling to meet content demands, van den Bussche built Bloo to take over. The character now anchors an entire channel that combines human voice control with AI-driven scripts, visuals and automation.

Bloo uses AI tools like ChatGPT, Gemini and ElevenLabs to generate voiceovers, create thumbnails, and translate content for his global audience. His creator has experimented with fully AI-generated episodes, but says they’re not yet as strong as ones guided by humans. Key word: yet.

The details: The VTuber boom is part of a broader trend where AI is used to scale digital personalities and eliminate production bottlenecks.

• Bloo has generated seven figures in revenue without a human on camera

• Hedra’s Character-3 model animates fully AI-powered characters in real time

• Comedian Jon Lajoie’s Talking Baby Podcast uses Character-3 for a hyper-realistic virtual infant host

• Virtual singer Milla Sofia builds music videos with AI choreography and vocals

• Startup TubeChef offers tools for creating faceless AI videos for as little as $18 per month

Faceless channels are growing fast. Some creators run networks with dozens of automated channels. One creator based in Spain said he publishes up to 80 videos per day using AI for everything except the idea. His content ranges from audiobooks to storytelling clips targeted at older audiences. His goal is to scale to 50 channels.

Van den Bussche's approach reflects a broader shift in the economics of content creation. "Turns out, the flaw in this equation is the human," he said in an interview. "We need to somehow remove the human." The 29-year-old Amsterdam-based creator invested millions of euros into developing Bloo after experiencing burnout from daily uploads over nearly a decade.

Why it matters: This wave of AI-generated video lowers the cost of content and accelerates production. It opens the door to creators who prefer not to be on camera and provides professionals with new ways to scale their digital media.

The rise of what users call “AI slop” is already drawing complaints. Some viewers claim their feeds are being inundated with low-effort AI-generated content that lacks originality or intent. Others worry about misinformation, especially when synthetic content becomes indistinguishable from real footage.

YouTube's monetization policies treat faceless content creators the same as traditional creators, meaning channels that meet subscriber and view thresholds can still earn ad revenue. This creates a powerful incentive for creators to automate production at scale.

The technology is improving fast, the audience is growing, and the incentive is clear. Faceless AI creators are quickly becoming a category.

Not every video will be good. Some will be engineered to hit the lowest common denominator. However, the infrastructure now exists for creators to build entire content operations without the need for studios, editors, or actors. What comes next is a race for differentiation.

🤔 Your thought process:

Selected Image 1 (Left):

• “The head of the beekeeper in the [other] image seemed out of proportion and just weird. Also, the frame that he is holding seemed too small for the size of the hive.”

• “The sunlight through the honeycomb frame is at the wrong angle for the fake photo.”

Selected Image 2 (Right):

• “Dang, I second-guessed myself because I looked at the hands and thought that’s not how I would hold a tray of bees. What do I know about holding a tray of bees?? Less than AI, apparently.”

• “Darn. I liked the quantity of honey on the frame but the keeper’s weird face should’ve been the tell.”

💭 A poll before you go

The Deep View is written by Faris Kojok, Chris Bibey and The Deep View crew. Please reply with any feedback.

Thanks for reading today’s edition of The Deep View! We’ll see you in the next one.

P.S. Enjoyed reading? Take The Deep View with you on the go! We’ve got exclusive, in-depth interviews for you on The Deep View: Conversations podcast every Tuesday morning. Subscribe here!

P.P.S. If you want to get in front of an audience of 450,000+ developers, business leaders and tech enthusiasts, get in touch with us here.