Michael T Nguyen

Echoing NVIDIA's CEO Jensen Huang keynote address at #CES last week Larry Dignan of Constellation Research, Inc. wrote a great article around "Why enterprise AI leaders need to bank on open-source LLMs" I think he is spot on and believe enterprises will treat LLMs like cloud compute — commodity where possible, differentiated where necessary. That means: ✅ multi-model strategies   ✅ portability (avoid lock-in)   ✅ hard questions on provenance, security, and SLAs https://lnkd.in/dyPwmd_C #EnterpriseAI #OpenSource #LLM #NVIDIA #GenAI #CIO #CTO

3

These interviews provide the underlying cognitive and hardware science that confirms the macroeconomic fractures in the volume-centric model by explicitly validating the mechanical failure of brute-force scaling and the impending financial trap for massive compute landlords. YouTube: https://lnkd.in/erpxthzi ​1. The Trillion Pound Baby: The failure of linear scaling is no longer a theoretical debate; it is a measurable reality. Marcus describes the industry's reliance on brute-force GPU scaling as the trillion pound baby fallacy - the naive assumption that because a model scales rapidly in its infancy, pouring infinite data and compute into it will endlessly yield exponential intelligence. The evident diminishing returns between iterations like GPT-4 and GPT-5 physically prove that volume-centric inference is collapsing under its own weight. Just as projects like Stargate hit an architectural wall, the scientific community is openly acknowledging that indiscriminately throwing hardware at the problem yields exponential energy and capital costs for rapidly shrinking marginal gains. ​2. Hardware Depreciation Trap and Stranded Assets: The structural exposure inside the balance sheets of major compute landlords is directly tied to the physical lifespan of the hardware they are heavily financing. Marcus acutely points out that the current capital expenditure cycle is entirely ignoring hardware depreciation. Plowing hundreds of billions of dollars into chips that rapidly lose their technical relevance creates a massive, ticking vulnerability. When hyperscalers spend $500 billion on GPUs based purely on the speculation of infinite, linear demand, they are manufacturing their own stranded assets. As the ROI on brute-force infrastructure fails to materialize at the application layer, the data centers financed by insurmountable non-current debt are on a direct path to becoming digital empty malls. ​3. The Inevitable Pivot to World Models and Efficiency: The most critical alignment is the absolute necessity of moving away from brute-force statistical prediction toward systems that route information precisely. Marcus explicitly argues that functional AI requires "world models"- architectures that understand causality, physical boundaries, and spatial relationships rather than just diluting noise across massive GPU clusters. Relying purely on volume-centric scaling is wildly inefficient and economically unviable. The industry is facing a hard financial mandate to pivot toward methods that are inherently more efficient and economical. This maps perfectly onto the necessary transition toward approaches that bypass the massive compute tax by understanding physical reality from the start rather than trying to memorize it via brute force. ​ These interviews confirm that the market correction won't just be financial; it will be driven by a hard architectural shift away from the very infrastructure that current valuations are built upon.

3

1 Comment

The race toward autonomous platforms is accelerating rapidly. The industry broadly recognizes the transformative growth potential of this market and is actively seeking accelerated paths to establish leadership in the domain.

5

Great connecting with Penguin Solutions at #NVIDIAGTC! Excited to advance sovereign AI through strong partnerships, practical infrastructure, and real-world performance. The next wave of AI will be built together, not in isolation.

17

1 Comment

Watching Jensen Huang’s keynote at GTC Paris this morning, one thing is clear: AI infrastructure is now economic infrastructure. And agentic AI is its next great workload. At KX, we’re proud to be building for this new era. Today we launched the AI Banker Agent—our first production-grade Agentic AI Blueprint, built in collaboration with NVIDIA. It brings together: ⚡ NVIDIA NeMo, Nemotron & NIM microservices 📊 KX’s high-performance time series & vector database 🧠 Domain-specific AI workflows for global markets trading This is real-time, applied AI—engineered for the speed, complexity, and compliance rigor of capital markets. 📣 Read the full press release: https://lnkd.in/e_JSnsxJ 🎥 Watch Jensen’s keynote from GTC Paris: https://lnkd.in/ekH_N4CV This is how we make the AI Factory real. #AgenticAI #AIInfrastructure #KX #NVIDIA #GTCParis #AIinFinance #KDBAI #RealTimeAI #FinancialServices #TradingAI #AppliedAI #AITransformation

38

1 Comment

Nice demo from Embedded World of how DepthIQ™ can be accelerated with an FPGA, in this case by Lattice Semiconductor.

10

1 Comment

Unlocking GPU Power for Data Processing with Dell AI Data Platform, Dell ObjectScale Storage and Apache Arrow Processing large datasets at scale is a challenge faced by every large enterprise. We announced Data AI Data Platform accelerated with NVIDIA GPUs at NVIDIA GTC Conference in March 2026. It accelerates data processing in Apache Spark by 3.4x by leveraging NVIDIA RAPIDS and cuDF (Cuda for Data Frames). Now we are further accelerating that performance to 5.7x by leveraging Apache Arrow Columnar format and integrations with Dell ObjectScale Storage Platform. Read this article written by me and my colleagues Chetan Somaiah to understand details and visit us at Dell Technologies World, May 18-21 to see a technical preview of this and other AI Data Platform capabilities. https://lnkd.in/g-CdYBib

12

2 Comments

Broadcom just gave markets a useful reminder: AI demand can be massive while margins get pressured. As AI shifts from “chips only” to full systems + custom silicon, the value pool moves — and investors are starting to price that reality in. The takeaway for CEOs/boards isn’t “AI is slowing.” It’s this: the winning AI strategy is the one that turns infrastructure spend into measurable productivity + defensible unit economics. Curious: where do you think the next margin moats will sit — chips, systems integration, or software/workflows? #AI #Semiconductors #Strategy #Board #CEOs #Tech #Markets #Investment #DigitalTransformation Broadcom Nasdaq S&P Global https://lnkd.in/d2WayVhh

3

True physical models, based on methods such as PDEs are the most useful way to predict behaviors for completely new shapes or operating regimes, as opposed to trained regression methods. So this is good news! We are big fans of #Fujitsu architectures such as the FX1000 and the #Fugaku for that reason. I had the opportunity at #SC25 to hear the #NVIDIA booth presentation about the collaboration between Fujitsu and NV and I asked both Yoshida-san of Fujitsu and Dan Ernst of NV about this very topic and their replies were encouraging. #Kombyne

3

A great read from Jean-Philippe Goyet on how Agentic AI in OSS must be grounded in operational realities not just innovation headlines. #Telecom #OSS #AgenticAI #AutonomousNetworks

41

1 Comment

👀 From grid-free Monte Carlo simulation to animatable avatars, #NVIDIAResearch papers at #SIGGRAPH2025 are redefining AI and graphics. See what’s new ➡️ https://bit.ly/44YdOkZ

2

NVIDIA GTC, CEO Jensen Huang’s Keynote on how physical AI is powering the next generation of intelligent systems with AI-RAN. The cloud built the digital economy. INTELLIGENT INFRASTRUCTURE will build the physical one, connecting the world to intelligence and enabling the AI economy. The INTELLIGENT INFRASTRUCTURE Executive Order is coming soon, now that the other EOs have been released. Intelligent Infrastructure Economic Zones bridge the Civil, Market, and Defence industries. The era of isolated autonomy is ending. We are entering the era of Autonomy 2.0 - orchestrated mobility and integration into our communities. Autonomy Institute https://lnkd.in/giGieZXi

2

Energy consumption is one of the biggest issues that needs to be addressed as AI adoption accelerates. Interesting article on how 2 companies are working together on the problem. Christopher Cubiss

12

Nvidia is reportedly building a $3B robotaxi fleet  Nvidia is reportedly investing $3 billion to create its own internal robotaxi fleet for US operations, a project separate from its existing partnerships with other major car manufacturers.

1

Reasoning is the next frontier for autonomous vehicles. Matt Cragun, Director of Product Management for Autonomous Vehicles and Simulation at NVIDIA, demonstrates how the #Alpamayo family of open models, simulation tools, and datasets enables AVs to perceive, reason, and act safely in complex scenarios. #nvidia #autonomousvehicles #autonomousdriving #ai #artificialintelligence Learn more → https://bit.ly/4pCcT21

11

1 Comment

Watch how the NVIDIA Omniverse DSX Blueprint provides an open, comprehensive #framework for designing and operating large-scale AI factories. It unifies simulation, live data, and AI-driven orchestration to boost throughput, resilience, and energy efficiency. #nvidia #ai #artificialintelligence #nvidiadsx

12

1 Comment

McObject has released eXtremeDB/rt 2.0, the next evolution of its HARD real-time database system. Now supporting persistent storage with deterministic flash management and advanced deadline scheduling, version 2.0 ensures predictable performance even under extreme workload and timing constraints. Ideal for aerospace, industrial, and safety-critical environments where timing precision truly matters. Learn more: https://t.ly/y6lf4 #extremedb #RealTimeData #FlashMemory #missioncritical #dbms

3

Cool story - Starburst has achieved a significant milestone by becoming the first open hybrid lakehouse platform optimized for NVIDIA’s new inference compute platform. This advancement allows customers to perform real-time AI and analytics on governed, federated data at unprecedented speeds. For those interested, Jensen Huang's keynote can be viewed here: https://lnkd.in/gb-6pStQ. Benefits for Starburst Customers include: - **Faster time to AI value**: Live, governed enterprise data supports inference workloads, RAG pipelines, and agentic AI in real time, eliminating the ETL bottleneck that often delays AI projects. - **Lower total cost of ownership**: Vera’s efficient architecture, combined with Starburst’s federated model that avoids costly data duplication, leads to significant savings in both compute and storage. - **Predictable performance at scale**: Deterministic throughput for concurrent analytics and AI workloads is maintained, even with complex, multi-source datasets. - **Built-in enterprise governance**: Fine-grained access controls and policy enforcement accompany every query, ensuring that AI applications utilize only authorized and compliant data as organizations expand into regulated environments. Dion Harris, senior director of HPC, Cloud, and AI Infrastructure at NVIDIA, emphasized, “The future of enterprise AI relies on securing instant insights from data that is often distributed across complex hybrid environments. Our collaboration with Starburst, optimizing their Trino-powered platform for NVIDIA Vera CPU, will deliver a foundational solution for real-time, federated data processing.”

36

1 Comment