By the TensorMax editorial team
· Drawing from sources across the AI industry
Today's top story
regulatory action
$100,000
The Trump administration has proposed a new fee for H-1B visas, while considering a $100,000 fee on Optional Practical Training, which allows international graduates to work in the US for one to three years after finishing their degrees.
Why it matters. The Trump administration's proposal for a new fee for H-1B visas and a $100,000 fee for Optional Practical Training could drive away future AI leaders, with roughly 419,000 people currently working on OPT in 2024. This move may exacerbate the existing issue of the US losing talented international students, such as Yang Zhilin, who chose to return to China and set up Moonshot AI, a company whose open model has drawn comparisons to those from OpenAI and Anthropic. The US risks losing its competitive edge in the AI market, with 40% of Nobel Prizes won by Americans in chemistry, medicine, and physics since 2000 having gone to immigrants.
The Trump administration has proposed a new fee for H-1B visas, while considering a $100,000 fee on Optional Practical Training, which allows international graduates to work in the US for one to three years after finishing their degrees. This move is seen as counterintuitive, given that the US has historically been a hub for international talent, with many foreign students choosing to study and work in the country. However, with the current visa system, many are being driven away, including Yang Zhilin, who despite getting his Ph.D. at Carnegie Mellon University and interning at Google and Meta, chose to return to China and set up his own company, Moonshot AI. The US is also losing out on talent due to its slow and restrictive immigration system, with the window for students to remain in the country after finishing a degree being cut from 60 days to 30. Other nations, such as Canada, Australia, and France, are taking advantage of the US's restrictive policies, actively recruiting researchers and international students. China, in particular, has been successful in pulling back talent, with its Thousand Talents Plan launched in 2008, and recently increased funding for research, making it an attractive destination for many. The US needs to rethink its immigration policies to remain competitive in the AI market, with some experts suggesting the creation of an Office of Strategic Human Capital to identify and support brilliant international AI students.
More from today
model release
Why it matters. The release of Jalapeño, OpenAI's first custom inference chip, marks a significant milestone in the company's compute strategy, with the chip delivering industry-leading speed and efficiency in AI inference, including more peak throughput per kilowatt and lower token latency than commercial systems on the InferenceX benchmark, and providing OpenAI with greater control over its models and serving economics, which is crucial as the company aims to continually seek the strongest mix of capability, speed, reliability, efficiency, and cost for each workload, with the goal of staying on the Pareto frontier, and as evidenced by the 54% fewer output tokens used by GPT-5.6 Sol with max reasoning on the Artificial Analysis Coding Agent Index.
model release
$899
Why it matters. The release of Apple's M6 and M5 Ultra chips marks a significant improvement in performance and AI capabilities, with the M6 chip offering up to 4x faster AI performance and the M5 Ultra chip providing up to 4.3x faster AI performance. This enhancement is crucial for Apple's position in the AI market, particularly with the M6 chip starting at $899 and the M5 Ultra chip starting at $5,499. The improved performance and AI capabilities will enable users to run more complex models and workflows, making Apple's devices more attractive to professionals and developers in the AI field.
model release
Why it matters. OpenAI's development of a custom inference chip, Jalapeño, in just nine months, has significant implications for the AI industry. With Jalapeño, OpenAI has achieved substantially better performance per watt than existing accelerators, handling 1.5 to 1.9 times more work per watt while cutting end-to-end latency by 1.7 to 3.6 times. This breakthrough has the potential to revolutionize the way AI models are deployed and used, particularly in interactive workloads where response time is critical. By building its own chip, OpenAI gains more control over the hardware evolution alongside its models, allowing for more efficient and responsive AI systems.
product launch
Why it matters. The strategic stake of the Cisco and NVIDIA partnership lies in its ability to compress the path between ordering infrastructure and producing the first token, with the full rack-scale Secure AI Factory solution set to be orderable through Cisco in September, marking a significant shift in the AI infrastructure market where time to first token becomes the new infrastructure metric, and winners will be determined by how quickly they turn capital expenditures into productive capacity, with the cost of consuming models through external APIs growing and sovereign AI programs requiring balanced performance, data control, and national infrastructure requirements.
product launch
Why it matters. The launch of Gemini Enterprise for Financial Services by Google Cloud is a strategic move to provide financial institutions with an AI-powered platform that meets their specific needs for speed, precision, and security. With 66 global systems integrators and fintech providers already on board, this platform has the potential to transform high-value workflows in capital markets and corporate banking, such as credit risk assessment and portfolio monitoring. For instance, the platform can reduce complex bond portfolio risk exposure analysis to under 5 minutes, complete with automated duration-hedging strategy suggestions. This can enable trading desks to deal with sudden macroeconomic shocks more effectively.
benchmark result
Why it matters. The strategic stake of this signal lies in its demonstration of scalable video captioning, with Anyscale and CoreWeave processing 600 TB of data in 95 minutes using 1,600 GPUs. This achievement has significant implications for the AI industry, particularly in applications such as physical AI, drug discovery, and creative generation, which require large-scale multimodal data processing. The use of GPUs for data processing has become essential, and the ability to scale infrastructure to meet this demand is crucial. With the ability to stand up a production-ready pipeline in under 24 hours, this solution addresses a significant pain point for teams working with large datasets.
Catch up quick
-
Researchers from multiple institutions publish a paper on indirect prompt-injection exposure in hidden states of agentic LLMs, including models like GLM-5.2 and Kimi-K3, achieving 0.90+ AUROC on unseen attacks
-
Researchers found that ordinary WiFi routers can identify people with nearly 100% accuracy
-
Stanford University economists find AI is causing significant entry-level job losses for younger workers in some fields, with employment levels 19% below those in less AI-exposed occupations
-
Google Cloud partners with leading law firms, including Cleary Gottlieb, Freshfields, Weil, and Williams & Connolly, to develop and launch Gemini Enterprise for Legal
-
Mexico provides $130 billion bailout to state-owned oil giant Petroleos Mexicanos
-
Stripe agrees to acquire OpenRouter for $8 billion
-
Tim Cook to step down as Apple CEO on September 1, with John Ternus taking over
-
Omar Yaghi, a Nobel laureate, leaves his faculty post at the University of California, Berkeley, to lead an institute in Beijing
-
A fully autonomous Russian drone killed three Ukrainians
-
The Trump administration's proposed rule to restrict mail-in voting through the U.S. Postal Service could exert federal control over state-run elections
-
The US State Department is set to revoke up to 200,000 B1 and B2 visas from foreigners who have applied for or are seeking asylum status
-
Europe's electric vehicle market share hits 25% in July with sales up 51% year-over-year
-
Secondary market for private company shares becomes increasingly active, with AI companies commanding premiums
-
China holds a significant share of capacity for robotics parts, including 90% of permanent-magnet processing, posing a challenge to US robotics companies.
-
AMD's share of x86 client CPU shipments tops 30% for the first time
Also on the desk
Who else should be reading this?Hit reply with the name and email of one person you think belongs on this list. We'll reach out personally, with your intro if you want.
|
|