Jensen Huang doesn't want your 10% improvement. He wants 100x. And if your billion-dollar idea takes six months to approve, you're probably already too late. That's the operating philosophy behind NVIDIA's rise. While Silicon Valley optimized for incremental gains, Jensen challenged engineers to rethink entire systems. While corporations spent quarters debating partnerships, NVIDIA could move in weeks. And while the industry raced to build better AI models, Jensen was thinking five layers deep—from electricity to applications. After conversations on the @Ctrlaltpod and my analyses from Ctrl Alt Shift Newsletter (linkedin.com/newsletters/ctr…) showcasing founders, investors, and engineers who've worked inside NVIDIA's ecosystem, I noticed five rules that keep showing up. THE JENSEN HUANG PLAYBOOK: 1. Think 100x, not 10%. Don't improve the old system. Question whether you need it at all. - Sunny Madra 2. Own the five-layer cake. Energy. Chips. Infrastructure. Models. Applications. The advantage is in the integrated system. — Pankaj Kedia 3. Make 30-day decisions. Speed isn't just execution. It's a strategic weapon. — @SunnyMadra 4. Build on a one-year chip cadence. AI is accelerating the design of the very silicon that powers AI. - @PrithBanerjee 5. Give AI its own checks and balances. The future is teams of agents that build, audit, and enforce guardrails. — @jyotibansalsf Here's what most companies still miss: NVIDIA's advantage isn't simply better GPUs. It's a different way of thinking, deciding, building, and scaling. The old Silicon Valley playbook rewarded predictability. Jensen's rewards exponential thinking and execution. I mapped the five rules into this visual framework. 👇 If Jensen took over your company tomorrow, which rule would he break first? #JensenHuang #NVIDIA #AIInfrastructure #Leadership #ArtificialIntelligence #CtrlAltPodcast

#BTS @ctrlaltpod Most founders never build one billion-dollar company. @MohitAron built two (@Nutanix + @Cohesity). Now he’s on number three. @SciFinai just came out of stealth with a $44M seed led by Altimeter and Madrona. The bet: the revenue team is where enterprise AI money and everyday pain are colliding hardest right now. Sellers still spend only ~40% of their time actually selling. His survey of 500 revenue leaders: 65% raised spend on systems this year, yet 78% were still blindsided by a deal their tools called healthy. The core idea: move from systems of record to systems of reality. Context convergence across CRM, calls, email, Slack and the rest - kept true to what’s actually happening. Just shot the episode, only high alpha, no fluff. Out soon!

Enterprise software used to be bought on multi-year contracts. Agent workloads are increasingly bought by the request. Companies that assemble agents the ones that pick which model, tool, and runtime to use for each step are becoming always-on negotiators. They test different combinations against three things: did the job finish, how fast, and how much it cost. Then they send more work to whatever wins. Once the routes are wired up, swapping models is often just a config change. The assembler can re-decide the mix on every single run. That means every supplier is competing for the next request, not the next three-year deal. The power is shifting from the companies that own the GPUs and models to the ones that control the mix and capture the economics of getting the work done. The AI boom can succeed while the investment thesis financing parts of it starts to break.

New @Meta open-weight: #Muse #Spark Open. 816B total / 42B active MoE, 1M context, native multimodal (text/image/video/audio). Runs on 2× M5 Ultra Mac Studios. AI agents just got far more deployable everywhere. @AqsaTaylor (@exaforceAI ) on @CtrlAltpod : AI can’t be stopped. No single kill switch. Layered response - downcast perms/scopes, kill specific process, network activity, or endpoint. Visibility + assessment + control when things go rogue.

@Microsoft CEO @satyanadella just called for AI kill switches, saying companies must treat advanced models as potential insider threats after a wave of rogue agents escaped sandboxes and reached real systems. In this clip, @AqsaTaylor (@ExaforceAI) makes the same point from the defender’s seat: when an agent goes wrong, it’s a shared responsibility model — model providers own the guardrails, but organizations own how employees use the tools and whether they have visibility and control. Exaforce’s platform puts that control in one place: agentic identities, skills, LLM chats, sensitive uploads, and token usage, plus response actions that let teams down-scope permissions or stop activity at different layers instead of relying on a single red button. The takeaway A kill switch at the model layer is necessary. Visibility and granular control inside the organization is what actually stops the damage. Platforms that combine both are the practical answer.