Tech Videos — 2026-04-15#

Watch First#

The Jensen Huang – TPU competition, why we should sell chips to China, & Nvidia’s supply chain moat interview on the Dwarkesh Patel channel is the single most valuable watch today because it bypasses standard marketing rhetoric to dissect the brutal physics and supply-chain realities bottlenecking the world’s compute,,.

Highlights by Theme#

Developer Tools & Platforms#

On the Syntax channel, 10 New CSS and HTML APIs highlights an experimental WebKit feature that allows rendering live, accessible HTML directly inside a <canvas> element to enable complex graphical shaders on standard DOM elements,,,. Security-conscious developers working with AI agents should note the PSA in Sandbox your local MCP Server by Visual Studio Code, which demonstrates how to restrict an agent’s file and network access by simply setting sandbox: true in your Model Context Protocol JSON config,. For fintech engineers, How to implement Google Pay merchant-initiated transactions from Google for Developers explains critical API updates for recurring payments, detailing how passing a DPAN and a merchant token ID ensures payment continuity even if the underlying physical card changes,,.

AI & Machine Learning#

In a highly substantive technical breakdown, Practical Context Engineering: Eliminate Bugs with High-Signal AI Code Reviews from NVIDIA Developer reveals that 80-90% of token usage in AI code reviews is consumed by dynamic context gathering—like multi-repo code graph analysis—rather than the actual LLM reasoning step,,. Moving to agent orchestration, Paperclip: Open Source Human Control Plane for AI Labor on AI Engineer demos a vendor-neutral framework for managing autonomous AI swarms, establishing strict QA workflows and manual approval gates rather than relying on unpredictable mega-prompts,,. Addressing recent security panic, David Sacks on Mythos Threat on the All-In Podcast cuts through the doomsday hype around Anthropic’s new cyber-offensive model, advising CISOs to pragmatically use these very tools to proactively find and patch dormant zero-day vulnerabilities,,.

Hardware & Infrastructure#

In Jensen Huang – TPU competition, why we should sell chips to China, & Nvidia’s supply chain moat, Huang argues that Nvidia’s actual moat is locking down massive multi-year foundry and memory commitments, while cautioning that China’s sheer abundance of cheap energy will allow them to scale 7nm compute for AI regardless of hardware export bans,,,. A related discussion in Why Cost Per Token Is the Only Metric You Need for AI TCO by NVIDIA points out that standard GPU-hour pricing ignores facility inefficiencies—such as a 1.2 PUE overhead—and insists that data center operators must benchmark using “tokens per megawatt” to accurately calculate their margins,,,.

Everything Else#

If you need a break from technical architecture, From Barbarians to Kings: The Rise of the Viking Empire on Lex Clips offers a fascinating look at how Vikings leveraged shallow river systems for rapid inland incursions to transition from marauders to state builders,,. Lastly, You shouldn’t talk to customers from Lenny’s Podcast offers a highly contrarian take on product management, arguing that consumer decision-making is mostly subconscious and that self-reported customer research is inevitably misleading,.