Top Stories
Jensen Huang’s GTC Keynote: Vera Rubin, Kyber, and $1 Trillion in Orders
Nvidia CEO Jensen Huang delivered his keynote yesterday at GTC 2026, projecting $1 trillion in orders for Blackwell and Vera Rubin systems through 2027. The numbers reflect the scale of AI infrastructure investment: computing demand has increased by a million times in recent years.
The headline hardware announcement was Vera Rubin, a full-stack computing platform comprising seven chips, five rack-scale systems, and one supercomputer built for agentic AI. Nvidia claims it delivers 10x more inference throughput per watt and one-tenth the cost per token compared to Blackwell.
Huang showed off Kyber, the next-generation rack architecture that integrates 144 GPUs in vertical compute trays for higher density and lower latency. That design arrives with Vera Rubin Ultra in 2027. Looking further ahead, Nvidia previewed the Feynman architecture, which will include a new CPU called Rosa (named after Rosalind Franklin).
Perhaps the most interesting reveal: the Nvidia Groq 3 LPU (Language Processing Unit), the company’s first chip from Groq after acquiring most of the startup for $20 billion in December. Huang claimed the Groq LPX rack can increase tokens-per-watt performance of Rubin GPUs by 35 times.
On software, Nvidia announced DLSS 5 for gaming (using 3D-guided neural rendering for real-time 4K), and NemoClaw, an open source stack for secure autonomous agent deployment.
Sources: CNBC, NVIDIA Blog, StorageReview
Meta Plans 16,000 Layoffs as AI Spending Doubles
Meta is preparing to cut up to 20% of its workforce - roughly 16,000 employees - to offset ballooning AI infrastructure costs. This would be the company’s largest layoff round since the 2022-2023 “year of efficiency” cuts that eliminated over 21,000 positions.
The math is stark: Meta’s capital expenditures are projected to hit $135 billion in 2026, nearly double last year’s $72 billion. Most of that goes to data centers, Nvidia GPUs, and custom AI chips. Senior leaders have reportedly been told to prepare for operating with reduced headcount.
Meta’s stock climbed nearly 3% on the news, signaling that investors approve of cost-cutting to fund AI ambitions. A spokesperson called it “speculative reporting about theoretical approaches,” though multiple sources confirm preparations are underway.
The company continues to struggle with its AI models. Its next-generation “Avocado” model has been delayed to May 2026, and leadership has reportedly discussed licensing Gemini from Google as a stopgap.
Sources: CNBC, TechCrunch
March 2026 Tech Layoffs Hit 45,000 - AI Cited in 9,200 Cuts
The AI-driven restructuring of the tech industry continues. March 2026 layoffs have already reached 45,000 across 168 tech companies, with over 9,200 positions explicitly eliminated due to AI and automation.
Block cut 4,000 jobs (40% of its workforce), WiseTech Global eliminated 2,000, Livspace cut 1,000, and eBay dropped 800. If this pace continues, 2026 could see 264,730 total tech layoffs, surpassing 2025’s 245,000.
Executives are now openly citing AI as the reason. The narrative: generative AI and LLMs are making software engineering more productive, making traditional approaches to writing and maintaining code “increasingly obsolete.” Whether that’s genuine productivity gain or AI-washing to justify cost cuts remains debated.
Consulting firm Challenger Gray & Christmas reports AI has been cited in over 12,000 U.S. job cuts so far in 2026.
Sources: TechNode Global, Invezz
Quick Hits
-
AWS deploying 1 million Nvidia GPUs: Amazon Web Services announced it will deploy over 1 million NVIDIA GPUs across AWS regions starting this year, deepening its partnership with Nvidia. NVIDIA Blog
-
OpenAI releases gpt-oss, first open weights since GPT-2: OpenAI’s gpt-oss-120b is now available for fine-tuning and on-premises deployment. Partners include Snowflake, Orange, and AI Sweden. The model matches o4-mini on AIME, MMLU, TauBench, and HealthBench. LLM Stats
-
EU AI Act high-risk rules may be delayed again: The European Commission’s Digital Omnibus package proposes delaying high-risk AI rules by over a year beyond the August 2026 effective date. The Commission already missed its February deadline for publishing guidelines. Blank Rome
-
AlphaFold adds 1.7M protein complex predictions: DeepMind expanded its protein structure database with 1.7 million new predicted protein complexes, announced at GTC. NVIDIA Blog
Worth Watching
GTC continues through Thursday with 1,000 sessions and 2,000 speakers. The partnership announcements suggest Nvidia is positioning itself as the infrastructure layer for everything: cloud (AWS, Azure), automotive (BYD, Hyundai, Nissan, Uber), industrial robotics (ABB, Universal Robots, KUKA), and healthcare.
The layoffs narrative is shifting. Companies are now explicitly attributing cuts to AI productivity gains rather than market conditions. Whether AI is genuinely replacing workers or providing cover for cost cuts that would have happened anyway is a question worth tracking closely.
Meta’s model troubles persist. The Avocado delay and rumored Gemini licensing discussions suggest Meta is falling further behind in the foundation model race, even as it spends more than anyone on infrastructure. The 16,000 layoffs are partly about funding that gap.