WHAT IS GOING ON

Zuckerberg shrugs off DeepSeek. While markets panicked over DeepSeek’s potential to reduce GPU demand, Meta CEO Mark Zuckerberg reaffirmed the company’s commitment to AI, pledging to invest “hundreds of billions of dollars” over the long term. Despite Nvidia’s stock plunge, he dismissed concerns that AI infrastructure spending would slow down, emphasizing Meta’s need for large-scale computing to support billions of users. With Llama 4 on the horizon, Zuckerberg aims to make Meta’s AI models the most competitive, challenging both open and closed alternatives.

Microsoft’s AI revenue soars. Microsoft reported $69.6 billion in Q2 2025 revenue, with AI-driven business surpassing a $10 billion annual run rate. CEO Satya Nadella emphasized AI as a key growth driver, contrasting with the company’s gaming division, which lags behind despite upcoming titles like Ninja Gaiden 4 and Doom: The Dark Ages. While Microsoft’s AI investments continue to pay off, the gaming sector remains a challenge in an otherwise strong earnings report. More here.

Apple Glowtime event. Apple's annual event was a tech buffet, showcasing the iPhone 16, AI-centric iOS 18 updates, AirPods 4, and Apple Watch Series 10. The iPhone 16, designed “ground-up for Apple Intelligence”, boasts a side camera control and a new A18 chip, making it 30% faster than its predecessor. Apple Intelligence, first revealed at WWDC, is set to enhance user experience across apps and features. The AirPods 4 promises personalized Spatial Audio and machine learning-enabled Siri interactions. And let's not forget the Apple Watch Series 10, with its AI-curated photos and machine learning-boosted translation tools. More here.

Digital temples.  The Sri Mandir app is helping Hindus connect with their faith, offering virtual visits to over 50 temples in India. The app, launched by AppsForBharat, allows users to participate in prayers, offer donations, and access devotional content straight from their phones. It's a hit with the Trivedi family in the U.S., who use it for last-minute prayers and donations. However, the app isn't cheap, with the average monthly spend outside India being $100.  A small price to pay for divine connection.  More here.

Try Guys try subscription.  The Try Guys, YouTube's popular creator group, have ditched Google's unpredictable algorithms and ad revenue, opting for their own streaming service, 2nd Try. With YouTube's ad revenue growth slowing and potential TikTok ban threatening creators' income, the Try Guys have found success in a subscription model. Since launching 2nd Try, they've seen steady income from exclusive content behind a $5 monthly paywall. They're not alone, other creators are also exploring subscription models for more reliable income. Despite the shift, YouTube still remains a part of the Try Guys' business model, but the focus is on growing 2nd Try.  More here.

AI CORNER

DeepSeek data leak raises security alarms. Cloud security firm Wiz discovered that DeepSeek left a critical database exposed online, leaking over 1 million records, including user prompts, API keys, and system logs. The vulnerability, found with minimal effort, raises concerns about DeepSeek’s security maturity, especially as it rapidly gains global adoption. This comes amid growing scrutiny over the company’s ties to China, with regulators and military officials warning of potential security risks.

Microsoft integrates DeepSeek’s R1 model into Azure AI and GitHub. In an uncharacteristically rapid move, Microsoft has added DeepSeek’s cost-effective R1 AI model to its Azure AI Foundry and GitHub, allowing developers to experiment with and integrate it into their applications. The company is also preparing a smaller version of R1 for local use on Copilot Plus PCs. On a related note, OpenAI and Microsoft are reportedly probing whether DeepSeek trained its models using OpenAI’s API, following security concerns about data usage patterns. More here.

DeepSeek's rise sparks AI export control debate. Simon Willison breaks down Anthropic CEO Dario Amodei’s perspective on DeepSeek’s AI advancements, arguing they follow expected scaling trends rather than signaling a fundamental shift in AI economics. Amodei stresses the need for strict export controls on advanced chips to China, citing concerns over potential military applications. Meanwhile, Willison highlights how DeepSeek’s open-weight approach sets it apart from closed AI labs like OpenAI and Anthropic. More here.

SoftBank eyes massive OpenAI investment. SoftBank is in talks to invest up to $25 billion in OpenAI, potentially making it the startup’s largest backer. The deal, still in negotiation, follows SoftBank’s prior $500 million investment and comes as OpenAI shifts toward a for-profit structure to compete in the AI arms race. This move aligns with the Stargate project, OpenAI’s joint venture with SoftBank and Oracle to expand AI infrastructure in the U.S.

DeepSeek in India. India’s IT minister announced plans to host DeepSeek’s AI models on domestic servers, marking a rare approval for Chinese technology in the country. Despite previous bans on Chinese apps, India sees DeepSeek’s cost-effective AI as a strategic opportunity, contingent on strict data localization requirements. The models will be hosted on India’s new AI Compute Facility, which offers discounted AI computing services to local firms. More here.

NEWS YOU CAN USE

Move over ChatGPT. Chinese AI startup DeepSeek has overtaken OpenAI’s ChatGPT as the #1 downloaded free app on the U.S. App Store, triggering global tech stock sell-offs. Founded in 2023, DeepSeek has impressed with its open-source R1 model, offering top-tier AI performance at a fraction of the cost of competitors like OpenAI and Meta. This milestone raises questions about the future of proprietary AI models versus open-source ones. U.S. tech leaders, including Meta’s Yann LeCun and Microsoft’s Satya Nadella, acknowledge the significance of DeepSeek’s achievement while predicting AI’s increasing accessibility will only expand its use and value.

AI-Powered VC dashboard.  TRAC, a four-year-old VC firm, is using artificial intelligence to pick its investments. Unlike its competitors, TRAC shares a piece of its AI algorithm with founders, creating a custom "intelligence dashboard" for each startup it invests in. The dashboard includes tools that help startups accurately value themselves and find new investors. TRAC's AI-powered approach seems to be working – one startup said TRAC's estimated valuation was within $3 million of what they actually raised.  More here.

Epic Systems mythic headquarters.  Tucked away in Verona, Wisconsin, Epic Systems, a leader in electronic medical records software, boasts a 1,670-acre campus that looks like it's straight out of a sci-fi novel. With offices inspired by "The Wizard of Oz" and "Harry Potter", the company's whimsical approach extends to its business model too. It's never accepted venture capitalist investment, made zero acquisitions, and insists on staying private. Epic's software is used across hospitals and clinics, storing medical records of over 280 million people in the U.S. Despite its fantastical surroundings, Epic has serious responsibilities and critics, with calls to streamline patient data exchange.

Bridging AI and corporate data.  Ragie, a startup offering a Retrieval Augmented Generation (RAG) as a service platform, is officially launching today. The platform connects enterprise data with generative AI large language models to provide updated and relevant information. Ragie is also announcing a $5.5 million seed round of investment led by Craft Ventures, Saga VC, Chapter One, and Valor. Ragie's technology is already in use as a core element of the Glue AI chat platform. The platform offers a simple, turnkey approach to enabling a data pipeline for RAG applications, simplifying the process for developers and organizations.

Apple’s ChatGPT prompts.  Apple's latest developer betas come with some intriguing AI features, and guess what, the prompts for these are stored right on your Mac!

On macOS, instructions programmed into the Apple Intelligence model can be read, showing up as default prompts that precede user inputs to a chatbot. A member of the macOS 15.1 beta subreddit discovered the files containing these backend prompts. Although these files can't be altered, they offer insights into the AI's functionality and operations.  Snippet above.

COOL NEWS

AI-powered robots and deep-sea discovery. Researchers in California are using AI-enabled robots like MiniROV to autonomously track and study marine life, addressing crucial gaps in ocean research. To train AI faster, scientists launched FathomVerse, a mobile game where players classify deep-sea creatures, improving machine learning models. With nearly 50,000 images identified, the collaboration is revolutionizing marine biology and climate research. More here.

Founder Mode" frenzy.  Memes are flooding the internet about "founder mode," a concept coined by Y Combinator founder, Paul Graham. The debate was sparked by an essay called "Founder Mode" by industry veteran Paul Graham. His argument? There are two ways to run a growing company: Founder Mode (hands-on) and Manager Mode (hands-off). While some tech leaders are all for rolling up their sleeves, others argue that managers play a crucial role. The tech world seems to be buzzing about this concept, with countless memes and hot takes making the rounds. More here and here.

Fish bot makes waves. Researchers at ETH Zurich have created Eve, a robotic fish designed to collect environmental DNA (eDNA) from aquatic ecosystems without causing a disturbance. Eve's design, complete with a silicone tail and noiseless internal pumps, allows it to blend in with other sea creatures. The bot navigates with sonar and an algorithm, and uses a camera to record underwater footage. Its main job is to collect DNA samples shed by marine life into the water, providing valuable data on species present without direct observation. This tech could revolutionize how we learn about our largely unexplored oceans.

DNA decoded with AI.  Researchers at Dresden University of Technology have developed GROVER, an AI model that reads human DNA like a language. By treating DNA as text, GROVER can decode the complex genomic information hidden within our genome. This tech tool is set to revolutionize genomics and fast-track personalized medicine. GROVER can also predict DNA sequences and extract biological meaning from them. So, in essence, GROVER has learned to 'speak' DNA. This could be a game-changer in unlocking the layers of genetic code that dictate our disease predispositions and responses to treatments – AI in genomics, folks.

Jackets swapped, tomatoes sliced.  At the SIGGRAPH 2024 conference, Nvidia CEO Jensen Huang and Meta CEO Mark Zuckerberg had an unusual fireside chat. The discussion began with tech talk, but quickly devolved into a jacket swap, inspired by a viral meme. The CEOs also shared anecdotes about Zuckerberg's precise tomato slicing habits and mutual compliments on their physiques. Huang's playful roasts of Zuckerberg were also a highlight, possibly reflecting Meta's reliance on Nvidia for model-training hardware. The chat was filled with unexpected turns, from F-bomb drops to Zuckerberg's style makeover.  Check it out here.