HPE and AMD are pushing enterprises to adopt AI and high-performance computing through on-premises infrastructure. Their combined hardware, including AMD EPYC processors and HPE Cray systems, supports demanding workloads like tsunami forecasting and drug discovery. Many organizations now prefer keeping AI work in-house rather than relying on public cloud, even if it means accepting slower storage. This approach helps meet privacy, security, and regulatory requirements while managing energy costs.
NVIDIA is making moves in both scientific research and robotics. Its cuPhoton tool helps scientists process large multidimensional datasets using GPUs, with Princeton and Harvard using it for astronomy and dark energy research. The company also released Isaac Lab 3.0, an open framework that trains robots for precise assembly tasks. In tests, one policy learned 20 assembly tasks with over 98% success while using significantly less GPU time.
Google DeepMind used its AlphaProof Code framework to formally verify that Signal's encryption protocol meets its security goals. The four-week project confirmed the messaging app's post-quantum PQXDH handshake and Triple Ratchet design. The AI also caught two reintroduced bugs, flagging them as insecure. Meanwhile, DeepMind co-founder Demis Hassabis warned that recursive self-improvement in AI carries serious risks, comparing unchecked development to a potential AI oil spill.
OpenAI acquired AI travel platform Odessia, founded earlier this year, which raised a $6 million seed round led by Sequoia Capital. Founder Francis Davidson will join OpenAI to focus on the future of travel planning. In healthcare, experts note that measuring return on investment for AI remains difficult due to inconsistent metrics and varying project goals across organizations.
Berkeley graduates Jai Bhatia and Aryan Vij turned down corporate job offers to launch Turnstone AI, joining a growing wave of young adults starting businesses in 2025. In Puerto Rico, business leaders at The Enterprise of the Future forum emphasized that corporate governance and human oversight are essential when deploying AI. Mastercard's Pablo Cuarón stressed that client privacy and data protection are non-negotiable in any AI strategy.
Key Takeaways
- HPE and AMD promote on-prem AI infrastructure using AMD EPYC processors and HPE Cray systems for workloads like drug discovery and tsunami forecasting
- NVIDIA's cuPhoton enables up to 14,900x faster image subtraction for scientific research at Princeton and Harvard
- NVIDIA's Isaac Lab 3.0 trains robots for assembly tasks with over 98% success while reducing GPU usage
- Google DeepMind's AlphaProof Code formally verified Signal's encryption, including its post-quantum PQXDH handshake
- DeepMind co-founder Demis Hassabis warns that recursive self-improvement in AI poses serious risks comparable to an oil spill
- OpenAI acquired AI travel platform Odessia, which raised a $6 million seed round from Sequoia Capital
- Healthcare organizations struggle to measure AI return on investment due to inconsistent metrics and differing project goals
- Berkeley graduates Jai Bhatia and Aryan Vij launched Turnstone AI after declining corporate job offers
- Puerto Rico business leaders say AI deployment requires strong corporate governance and human decision-making oversight
- Most organizations now prefer on-site AI infrastructure over public cloud to maintain data sovereignty and regulatory compliance
HPE and AMD help scale AI without straining power grids
HPE and AMD report that corporate leaders must now deploy AI and high-performance computing to drive real results. Their combined hardware, including AMD EPYC processors and HPE Cray systems, supports work like tsunami forecasting and drug discovery. The focus is on helping businesses use AI efficiently while managing energy costs and security.
Data sovereignty reshapes how organizations build AI systems
A study found that most organizations keep high-performance AI work on-site instead of using public cloud. Many buyers accept slower storage performance just to keep data under their own control. HPE Cray systems with AMD EPYC CPUs and AMD Instinct GPUs aim to deliver fast AI-ready performance on premises. This approach helps meet privacy, security, and regulatory rules.
NVIDIA cuPhoton speeds up scientific data processing
NVIDIA cuPhoton is a tool that helps scientists process large multidimensional datasets using GPUs instead of CPUs. It is being used by Princeton University and Harvard University for astronomy and dark energy research. The system claims up to 14,900x faster image subtraction and much faster data loading and analysis. It supports formats like FITS and HDF5 and works with tools including PyTorch.
Isaac Lab 3.0 trains robots for delicate assembly tasks
Isaac Lab 3.0 is an open robot learning framework from NVIDIA that helps train robots for precise assembly work. It supports physics engines like PhysX and Newton and improves how robots learn through practice. In tests, one policy learned 20 assembly tasks with over 98% success while using far less GPU time. New evaluation tools like DexBench and VABAR help measure robot performance across many settings.
Measuring return on investment for AI in healthcare is difficult
Healthcare organizations are using more AI tools, but calculating return on investment remains challenging. Different projects use different goals, such as better patient outcomes or lower costs, making comparisons hard. There is also no standard set of metrics, which leads to inconsistent results. Experts say better evaluation methods are needed to guide spending decisions.
Google DeepMind AI proves Signal encryption is secure
Google DeepMind used a multi-agent AI framework called AlphaProof Code to formally prove that libsignal, the code behind Signal's messaging protocol, meets its security goals. The four-week project covered Signal's post-quantum PQXDH handshake and Triple Ratchet. The team reintroduced two known bugs and the AI correctly flagged them as insecure. Signal CTO Ehren Kret said the work expands earlier verification efforts. DeepMind notes the proof assumes perfect cryptography and does not cover the full mobile app.
DeepMind's Hassabis warns of AI oil spill risk
Demis Hassabis, DeepMind co-founder, warned that the race to build AI that improves itself is outpacing safeguards. He described recursive self-improvement as useful but carrying the most risk. Hassabis said AI currently makes engineers three to five times more productive, but removing human reviewers from the loop is a very dangerous next step. He compared the potential outcome to an AI oil spill unless the industry changes course.
Berkeley grads launch AI startup Turnstone
Berkeley graduates Jai Bhatia and Aryan Vij turned down lucrative corporate job offers to launch their AI startup called Turnstone AI. The two founders appeared on Fox and Friends to explain their decision. They also discussed a study showing a surge of young adults starting businesses in 2025. The founders said they are betting on themselves with their new artificial intelligence company.
OpenAI acquires AI travel platform Odessia
OpenAI acquired the AI travel planning platform Odessia, announced by founder Francis Davidson on Tuesday. Davidson said he will join OpenAI to focus on the future of travel. Odessia uses AI to help users plan and book trips. The platform was founded earlier this year and raised a $6 million seed round from investors including Sequoia Capital.
Puerto Rico leaders discuss AI competitiveness rules
Business leaders in Puerto Rico gathered at the Sheraton Puerto Rico Convention District Hotel for The Enterprise of the Future executive forum. The event was hosted by the Official Spanish Chamber of Commerce in Puerto Rico in partnership with Citi. Pablo Cuarón of Mastercard said client privacy and data protection are non-negotiable when deploying AI. Adriana Ramirez of Abarca Health and Graham Castillo of Estudios Tecnicos said AI effectiveness depends on corporate governance and keeping humans in the final decision.
Sources
- The CIO dilemma: Scaling production AI without breaking the power grid
- Data sovereignty is reshaping AI infrastructure
- NVIDIA cuPhoton: GPU-Accelerated Scientific Data Processing
- Isaac Lab 3.0 trains robots for delicate assembly work
- Why evaluating ROI from AI isn’t always straightforward
- Google DeepMind Uses AI to Formally Prove Signal's Encryption Is Secure
- DeepMind's Demis Hassabis Warns an AI ‘Oil Spill’ Is Coming Unless the Industry Changes Course
- Berkeley grads turn down corporate jobs to launch AI startup
- OpenAI Acquires AI Travel Agent Odessia
- Puerto Rico business leaders analyze new rules of competitiveness amidst advancing artificial intelligence
Comments
Please log in to post a comment.