AI & Tech Weekly (8.11–8.15): Grok4 Expert opens, AI stocks rally, Cohere funding, Tencent earnings, Apple AI hardware, Meta DINOv3
Over the past week, generative AI and hardware ecosystems accelerated on two fronts: platforms balanced “lowering entry barriers to acquire users” with “upgrading experience to drive conversion,” while capital markets sent an even clearer signal around AI as a core theme. Meanwhile, large models and self-supervised vision kept iterating, widening real-world use cases. Here are the highlights and takeaways.
1) Grok4 opens Expert mode to free users
Grok announced that Expert mode (based on Grok4) is now available to free users, including image-drawing features. Usage is limited; after hitting the cap, users must upgrade to SuperGrok.
Commentary:
With ChatGPT-5 topping app stores, the “Grok4 vs. ChatGPT-5” conversation is heating up. Grok’s move is both feature iteration and growth strategy: reduce the trial threshold to capture more potential paid users. Amid increasingly homogeneous competition, whoever builds stickier habits—retention and repurchase—earlier will have a commercial edge.
2) AI stocks continue to attract capital
Bridgewater Associates’ Q2 holdings show Alphabet (GOOGL) as the fifth-largest position, up 84.08% QoQ by shares; NVIDIA (NVDA) holdings rose 154.37% to become the third-largest position; Microsoft (MSFT), Meta (META), and Salesforce (CRM) also ranked in the top ten. Value investor Duan Yongping likewise added NVDA and Google in Q2.
Commentary:
Concentrated allocations by top funds signal a shift from “theme trade” to “core asset.” With technical dividends and compute moats stacking, leaders’ earnings power and valuation frameworks still have room to expand.
3) Cohere raises $500M at a $6.8B valuation
Canada’s Cohere closed a $500M round at a $6.8B valuation, led by Radical Ventures and Inovia Capital, with AMD Ventures, NVIDIA, PSP Investments, and Salesforce Ventures participating. Cohere focuses on enterprise AI models and named Joelle Pineau as Chief AI Officer and Francois Chadwick as CFO.
Commentary:
Unlike general-purpose foundation-model paths, Cohere emphasizes enterprise customization—building a moat in focused scenarios. Participation by supply-chain giants strengthens its commercialization outlook, though ultimate leadership still hinges on deployment speed and ecosystem integration.
4) Tencent earnings: AI as a growth engine
Tencent’s Q2 2025 results: operating capex RMB 17.9B (+149% YoY), mainly for GPUs and servers; free cash flow RMB 43B (+7% YoY). AI applications continued to land—Yuanbao now supports Official Accounts content, Video Accounts comments, and Mini Shop customer service. In gaming, AI boosts content creation, engagement, and monetization. Ad revenue reached RMB 36B (+20% YoY); Video Accounts marketing services grew ~50% YoY; cloud saw gains alongside GPU and API token demand.
Commentary:
With social traffic and the content ecosystem as a steady base, AI acts as a dual amplifier for efficiency and monetization: higher CTR/conversion in ads, scaled compute/model services in cloud. The growth curve looks more resilient—with more upside optionality.
5) Apple plans a new AI hardware lineup
Apple is advancing an AI hardware roadmap: a desktop robot (with a human-like Siri, targeted for 2027), a smart speaker with display (targeted for next year), and a home security camera (as the core of Apple’s security system).
Commentary:
Apple remains committed to an integrated “software + hardware + services” playbook. The challenge is to define “true user needs” and high-frequency use cases. If Apple delivers a step-change experience, AI hardware could see an iPhone-style category leap.
6) Meta unveils 7B-parameter DINOv3, pushing self-supervised vision
Meta introduced DINOv3, a self-supervised vision model that produces high-resolution image features and leads on dense prediction tasks, while being directly deployable across diverse vision tasks with minimal or no fine-tuning. Reports suggest Meta may further reorganize its AI org over the next six months.
Commentary:
Self-supervised learning (SSL) reduces labeling costs and has huge potential, but vision progress has lagged. DINOv3 offers a more efficient SSL path—lower compute cost, broader transferability—providing a general backbone for many vision scenarios.
That’s our AI & tech weekly wrap from the iaiseek team. For more cutting-edge updates, business insights, and tech trends, visit:
https://iaiseek.com
Read last week’s report:
AI & Tech Weekly (8.4–8.8): GPT-5 sparks global buzz, Apple bets $100B on U.S. manufacturing
See you next week—we’ll keep tracking model upgrades, AI hardware & compute investment, and how capital flows reshape the industry.