Tech • AI • Robotics • Game
US AI Is Exploding From Within, China Has the Winner's Smile
7/10TL;DR
Salesforce has launched a reasoning model built with Nvidia that targets sales, marketing and customer support, as geopolitical tensions over AI governance harden between the United States and China.
KEY POINTS
Salesforce enters frontier AI
Salesforce, a major CRM software group with $38 billion in annual revenue, $4 billion in profit and about 75,000 employees, unveiled its first in-house reasoning model, Koa, inside Agentforce. The system is designed for long, multi-step business tasks across prospecting, campaign planning, lead enrichment, email outreach and customer support. Unlike general chatbots, the goal is not benchmark glory but direct execution of office work.
A vertical model aimed at white-collar roles
The model is specialized for sales, marketing and support functions, the core workflows already embedded in Salesforce products across large companies worldwide. That makes it a notable shift from AI as an assistant used by employees to AI embedded in enterprise software that can replace parts of those roles outright. The immediate question for companies is no longer which worker uses AI best, but how many tasks the software can handle without a worker.
Nvidia moves beyond chips
Nvidia appears as more than a hardware supplier in the launch. Koa was built on Nemotron 3 Super, one of Nvidia’s open-weight models, which places the chipmaker in more direct competition with the same AI developers that depend on its infrastructure. Salesforce used that base model and further trained it on synthetic business scenarios drawn from decades of expertise in customer management rather than client data.
Specialized AI may beat general models in business use
The larger strategic implication is that highly verticalized models can outperform broader models on specific job families. In practical terms, a business-trained model can be more useful for replacing a sales or support workflow than a more powerful general-purpose system. That strengthens the position of software incumbents that own domain data, customer relationships and workflow distribution.
China rejects calls to slow advanced AI
Beijing responded sharply to renewed calls from parts of the US AI sector to pace frontier-model releases. Chinese officials framed those demands as selective and self-serving, arguing that Washington opposed stronger international intervention only days earlier and now backs controls that could preserve US advantage. China’s answer is a push for open-weight AI deployment rather than a slowdown.
An organized Chinese-led AI bloc is taking shape
China is promoting cooperation through the World Artificial Intelligence Cooperation Organization and broader BRICS alignment. The pitch combines shared training infrastructure, lower-cost model deployment and sovereign adaptation of Chinese models by partner countries. Beijing’s message is that AI safety and strategic scale can coexist under a state-backed, coordinated system.
Real-time voice AI is accelerating
The race between OpenAI and Google also intensified with new real-time voice systems. OpenAI released GPT Live 1 as an API, while Google answered with Gemini 3.8 Live. These are duplex models that can speak and listen simultaneously, making conversations sound far more natural than older turn-by-turn assistants and opening the door to automated call centers, reception desks and support lines.
The economics favor rapid deployment
The new voice models are becoming cheap enough for mass integration into software products and phone systems. That matters because cost has been a major brake on replacing human-handled calls at scale. Combined with agentic features that can search records, pull customer files and trigger actions, the technology directly reinforces the automation trend already highlighted by Salesforce’s launch.
OpenAI is also investing in machine vision
OpenAI reportedly acquired Glass Imaging for about $300 million. The company specializes in AI that improves image quality at the sensor-processing stage rather than by simply retouching finished pictures. The move suggests interest in future AI-native hardware that can continuously see and interpret the surrounding world, extending beyond phones and PCs.
The US signals a harder military edge in space
In a separate escalation, US Air Force Secretary Troy Meink publicly stated that the United States has “space control” weapons in orbit. The declaration implies capabilities to disrupt, degrade or possibly destroy adversary systems in space, including through electronic warfare or more direct means. The announcement adds a military layer to a broader contest already spanning semiconductors, AI models and global digital infrastructure.
CONCLUSION
The latest moves show AI competition shifting from standalone models to integrated systems that automate real business functions and reshape strategic power. The result is a tighter link between enterprise software, state rivalry and the growing militarization of advanced technology.
Explain this