Gemini and ChatGPT hit 1 billion users, NVIDIA launches trillion-parameter Nemotron, and regulators force Meta to unwind Manus acquisition.
Google CEO Sundar Pichai announced that Gemini has surpassed 1 billion monthly active users, making it the fastest-growing product in Google's history. OpenAI's ChatGPT also crossed the 1 billion mark almost simultaneously. With two major AI assistants hitting this ceiling on the same day, the tipping point where AI shifts from a geek toy to a universal tool arrived earlier than many expected. For the application layer, the boundary of the user base has been redrawn; whoever can retain billions of users holds the key to the next phase of monetization.
The Information, citing project participants, reported that NVIDIA is developing a new generation of open-source model series, Nemotron 4, aiming to challenge the world's most advanced open-source models. On launch day, its open-source version Nemotron 3.5 Lightning was also released, with Jensen Huang personally endorsing open source. NVIDIA is one of the few large tech companies in the US actively pushing open-source models; by raising the stakes to trillion parameters, they are clearly using open-source models to defend their ecosystem position.
OpenAI announced it has begun testing ads within ChatGPT while launching the Daybreak model on AWS. Pushing forward on both commercial monetization and cloud channels, the path for charging AI applications is becoming clearer. For users accustomed to free lunches, ads entering conversations are inevitable; the key is how to do it without being intrusive.
Manus stated on Tuesday that it will soon resume operations as an independent company. In April this year, Chinese regulators required Meta to unwind its $2 billion acquisition of the firm, forcing the deal to be dismantled. This marks the first time Chinese regulators have enforced the unwinding of cross-border AI asset acquisitions, clearly signaling the sovereign nature of AI application-layer assets and interrupting Meta's global path of supplementing agent capabilities through acquisitions.
Hotel orders redirected from Doubao to Douyin Laike now enforce an independent fee rate, totaling around 12%, compared to the 8% cut taken by Douyin's information flow entry point. The 4 percentage point difference reflects the anchor set by ByteDance for large model commercialization. AI dialogue entry points are shifting from traffic funnels to toll gates, with every transaction on this chain being repriced.
Unitree disclosed the online lottery results for its STAR Market IPO, with 19,414 winning numbers, an issue price of 150.80 yuan/share, requiring payment of 75,400 yuan per lot, and a final online subscription win rate of only 0.018%, about 3.8% of CXMT's. The popularity of the "first humanoid robot stock" is evident, with primary market enthusiasm amplified by retail investor sentiment. For the secondary market, the first-day performance after listing at a high valuation will test the absorption capacity of the humanoid robot sector.
Uber sold its entire holding in autonomous delivery robot company Serve Robotics, exiting completely after five years. On the same day, primary market enthusiasm for the humanoid robot sector remained undiminished. The former parent company stepping away indicates that commercial realization in the unmanned delivery track is slower than imagined, and capital is beginning to requeue.
The XPeng G9L Golden Five-Seater Flagship began pre-sales with a price starting at 259,800 yuan, equipped with 480 LED beads and digital projection headlights, tested for 3 years and compliant with over 3,000 regulations across 64 countries. From today, a 1,000 yuan deposit offsets 3,000 yuan off the purchase price, with Fjord Gray paint and Sahara Orange interior available for free for a limited time. He Xiaopeng said, "It's not just about having strengths, but more importantly, having no weaknesses," clearly targeting full-scenario smart driving on highways and in cities.
Envision Technology announced in Ulanqab the landing of "Envision Ulanqab," the world's largest AI compute super monolith, revealing an integrated "New Energy + Compute" model. The key variable in the compute race is shifting from chips to electricity. Whoever holds sufficient green power holds the ticket for the next round of expansion. For northwest regions rich in green power but lacking industry, compute landing has become a new industrial variable.
Latest research shows AI can extend tornado warning times to 30 hours, implementing physical AI in disaster prevention scenarios. These extra hours gained could mean the difference between life and death for evacuation and transfer.
NVIDIA announced partnerships with Apollo, Blackstone, BlackRock, Brookfield, Goldman Sachs, KKR, etc., to establish an AI computing infrastructure financing platform, planning to mobilize over $500 billion in third-party capital. Wall Street endorses Jensen Huang's compute financialization plan with "big concepts," as AI infrastructure shifts from capital expenditure to a new asset class competing for layout. Compute providers can use off-balance-sheet funds to expand production, alleviating balance sheet pressure on giants, while demand-side confidence is amplified by financial leverage. Directly benefiting NVIDIA and the hardware chain, financing channels for compute leasing firms like CoreWeave are broadened, and the financeability of capex conversely supports cloud vendor valuations.
CoreWeave released its Q2 earnings report, with revenue doubling year-over-year and beating Wall Street expectations due to accelerating AI infrastructure demand, rising up to 10% after hours. As the purest AI compute leasing target, its earnings report is viewed as a litmus test for "the authenticity of compute demand," directly supporting the logic behind NVIDIA's $500 billion financing platform.
📰 EastVoice — today's AI headlines, decoded
Five minutes a day, the global AI shift →