Jensen Huang told Donald Trump that “we’re not going to let” an AI slowdown happen and argued that companies building AI products—not new regulators—should engineer their safety. The Nvidia CEO described AI as hardware and software, not an “alien mind.”
That puts Huang at odds with Dario Amodei, whose calls to slow AI development have support from Elon Musk and Sam Altman. Huang made the argument during Salesforce’s Dreamforce, where Salesforce announced Koa, its first CRM reasoning model, built on NVIDIA Nemotron 3 Super.
Huang also took a live call from Trump during the event, putting the Nvidia CEO’s position in a direct conversation with the president.
Why it matters: Huang’s proposal would leave AI product makers responsible for engineering safety rather than relying on new regulation, and he made that case to Trump as Salesforce introduced Koa.
U.S. data centers could consume more natural gas than Germany and Japan combined by 2035, according to TechCrunch AI. The report says the AI frenzy could make data centers one of the world's largest natural-gas consumers.
NIST awarded more than $30 million to Manufacturing Extension Partnership Centers in 11 states and Puerto Rico. The awards will help small and medium-sized manufacturers increase their adoption of advanced manufacturing technology.
Why it matters: The 2035 projection gives U.S. data-center operators a concrete natural-gas benchmark, while NIST's awards give small and medium-sized manufacturers support for adopting advanced manufacturing technology.
Amazon's Bedrock materials cover three operational pieces for AI agents: prompt caching that can cut repeated input-token costs by up to 90%, a managed OAuth consent portal for services such as GitHub and Slack, and ephemeral compute for agent workflows.
AWS documents implementations rather than just feature descriptions. Abnormal AI uses Bedrock AgentCore Code Interpreter as an ephemeral scratch pad for real-time email threat detection at billion-message scale. Ninth Wave uses Compass, a multi-agent assistant, to validate bank APIs against Financial Data Exchange standards and compress open-finance onboarding from weeks to minutes.
The performance figures come from AWS-published implementation examples: the caching result is an "up to" claim, and the onboarding result describes Ninth Wave's deployment.
Why it matters: Teams building Bedrock agents can reuse context at up to 90% lower input-token cost, collect GitHub and Slack authorization through a managed consent flow, and use ephemeral Code Interpreter compute for production email-security workloads.
The US and China agree that advanced AI poses serious risks. Beijing, however, is deeply skeptical of a deal that prioritizes keeping US companies ahead.
The disagreement is over the proposed deal’s competitive priority—not whether advanced AI carries risks.
Why it matters: For US companies, a deal that prioritizes keeping them ahead faces Beijing’s skepticism despite shared US-China concern about advanced AI risks.
TypeSafe has left stealth with Jev, a specialised System One Model designed to automate programmatic decisions inside production software. The model uses a parallel-sampling architecture to execute structured probabilistic decisions directly.
The release targets software systems that require automated deterministic logic. TypeSafe says those systems can bypass conversational language models and run structured decisions directly inside production.
Why it matters: Software teams building production systems that require automated deterministic logic can use Jev to execute structured probabilistic decisions directly instead of relying on conversational language models.