AI Weekly Roundup – March 27, 2026

This week may go down as one of the most consequential in recent AI history. From NVIDIA's CEO declaring that AGI has arrived, to OpenAI shuttering its video platform, to TSMC facing a helium supply crisis that could disrupt the global chip supply chain — the pace of change across artificial intelligence, semiconductors, and autonomous vehicles is staggering. Here is everything that happened, with primary sources.

1. Jensen Huang Declares AGI Has Arrived

In a wide-ranging conversation on the Lex Fridman Podcast (episode #494, March 23), NVIDIA CEO Jensen Huang made a statement that reverberated across the tech world: AGI, artificial general intelligence, is here. Huang framed AGI as the ability to operate autonomously — not just to answer questions, but to build and manage complex enterprises without constant human direction.

The claim challenges the prevailing narrative that AGI is a decade or more away. Coming from the head of the world's most valuable company — one that sits at the centre of the entire AI compute stack — the declaration carries particular weight. Huang also discussed the enormous demand for NVIDIA hardware and predicted further repricing across the semiconductor supply chain.

2. OpenAI Shuts Down Sora Video Platform

OpenAI confirmed this week that it is shutting down Sora, its AI-powered video creation app. Launched in September 2025, Sora peaked at 3.3 million downloads in November before declining steadily to 1.1 million by February 2026. A reported partnership with Disney for character-integrated content did not proceed.

The decision reflects a broader pattern of aggressive cost control at OpenAI, as the company redirects compute resources toward its core research mission. The move also acknowledges a competitive reality: xAI's Grok Imagine has rapidly overtaken Sora's capabilities in both image and video generation.

3. Grok Imagine Surges as xAI's Creative Engine

xAI's Grok Imagine platform has had a breakout quarter. After launching its API in January 2026 with text-to-video, image-to-video, and video editing capabilities at $0.05 per second, the platform released version 1.0 in February with 10-second videos at 720p resolution. A major March update added stylised templates and expanded creative tools. The numbers speak for themselves: 1.245 billion videos were generated in January alone.

4. OpenAI's Rapid-Fire Pivots Raise Questions

Sora is only the latest casualty. In recent months, OpenAI has cancelled or indefinitely postponed a string of major initiatives: the expansion of its flagship Stargate data centre in Abilene, Texas (a financing dispute with Oracle); a planned advertising revenue model; in-app shopping with direct checkout; and its first hardware device — a listening wearable that reportedly drew personnel from Apple. The Stargate cancellation alone involved a planned 4.5 GW expansion that is now being partially absorbed by Meta, which placed a $150 million deposit with Crusoe Energy.

The volume of pivots in a compressed timeframe raises strategic questions. When priorities change every few months, execution and team alignment become extremely difficult — even for a company of OpenAI's scale.

5. Anthropic's Claude Takes Control of Your Desktop

Anthropic released a research preview of agentic desktop control for Claude, allowing the AI to directly control a Mac's mouse, keyboard, and screen. Available to Claude Pro and Max subscribers, the feature pairs with Dispatch — a companion tool launched the previous week that lets users assign tasks from an iPhone. Claude now leads several agentic AI benchmarks, particularly in coding, and Anthropic reports that roughly half of its revenue comes from enterprise customers.

The Pentagon blacklist remains a concern for the company, though its commercial trajectory appears strong in the near term.

6. 27 Software Companies Flag AI Agents as Existential Risk

A record 27 US-listed software companies identified AI agents as a material risk to their business in Q1 2026 regulatory filings. Companies including Figma, HubSpot, ServiceNow, and Workday have added or strengthened language around the threat that autonomous AI agents pose to traditional enterprise software. Across the S&P 500, 448 companies mentioned AI in their most recent 10-K filings, up from 359 in 2023.

The fear is not abstract. Developers have demonstrated single-session rebuilds of complex enterprise software like HR systems and CRM platforms using agentic AI tools from Anthropic, Google, and xAI — a capability that threatens the recurring-revenue models underpinning most SaaS businesses.

7. Google's TurboQuant Slashes Inference Costs

Google Research published TurboQuant, a technique that compresses large language model KV caches to just 3 bits with no measurable accuracy loss. The result: up to 8x faster attention on H100 GPUs and a dramatic reduction in the memory required for AI inference. The method uses two algorithms — PolarQuant and QJL — and requires zero additional training. The paper will be presented at ICLR 2026.

Memory chip stocks took a hit on the news, as the breakthrough could significantly reduce the amount of high-bandwidth memory required by AI data centres.

8. Gemini Gets Personal Across Google Apps

Google is rolling out a "Personal Intelligence" feature that connects Gemini to Gmail, Photos, YouTube, Search, Maps, and other Google services. Initially available to AI Pro and Ultra subscribers in the US, it is expanding to free-tier users. The feature links personal accounts only (not Workspace or enterprise) and does not use personal data for model training.

9. Google Maps Gets Conversational AI

Google Maps now includes "Ask Maps," a Gemini-powered conversational search feature. Users can ask natural-language questions and receive personalised answers drawing from reviews, search history, and saved places. The feature is rolling out on Android and iOS in the US and India, with desktop support planned.

10. NVIDIA Unveils Vera Rubin at GTC 2026

At its annual GTC conference on March 16, NVIDIA unveiled the Vera Rubin platform — a vertically integrated 7-chip system featuring the Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX-9, BlueField-4, Spectrum-6, and Groq 3 LPU. The platform promises 10x inference performance per watt over its predecessor generation, at one-tenth the cost per token. Jensen Huang disclosed that combined Blackwell and Vera Rubin orders are expected to reach $1 trillion through 2027.

NVIDIA also announced a major pivot toward inference workloads, aligning its AI factory infrastructure with the agentic AI shift that is driving demand across the industry. Energy efficiency is a central theme: each new generation of chips delivers substantially more compute per watt, easing the data centre energy bottleneck.

11. TSMC Faces Helium Supply Crisis

A geopolitical flashpoint is threatening the heart of the global chip supply chain. Qatar's Ras Laffan facility, which produces approximately 30% of the world's helium, went offline on March 2 following Iranian drone strikes. With the Strait of Hormuz effectively closed to Western shipping, TSMC — which manufactures roughly 90% of the world's advanced semiconductors — faces a critical shortage of helium (essential for chip fabrication cooling) and liquefied natural gas for electricity.

Reports suggest TSMC may have as few as 11 days of supply remaining. If the situation is not resolved, energy prices could spike, chip factories could halt production, and a global semiconductor shortage could return. There is currently no viable alternative supplier at the required scale.

12. ARM Launches First In-House AGI CPU

In a historic first, ARM Holdings announced its own processor — moving beyond chip design into manufacturing. The chip features 136 Neoverse V3 cores on TSMC 3nm process technology, with a 300W thermal design power (compared to 500W for competing x86 128-core processors). Meta is the lead customer, with additional commitments from OpenAI, Cloudflare, and SAP.

ARM projects $15 billion in revenue from AGI CPUs alone by 2031, with volume shipments expected by the end of 2026 and material revenue contribution from 2028. The stock surged from $120 to $160 in a single session on the announcement.

13. Tesla Robotaxi Expands Across the US

Tesla's unsupervised FSD-powered robotaxi service, which launched in Austin, Texas in January 2026, is expanding to Miami, Dallas, Phoenix, and Las Vegas. Job listings for safety operators and rapid response field operators have appeared in Minnesota, Washington, New York, Pennsylvania, and Massachusetts — a strong indicator that rollout to those regions is imminent. Tesla does not typically hire ahead of deployment timelines.

Population analysis suggests the planned rollout could cover approximately 39.2% of the US population in the first half of 2026, spanning major metro areas: the Bay Area (9 million), Los Angeles (18 million), and greater New York (20 million).

14. Tesla Optimus: Recruiting for the Biggest Product on Earth

Tesla released a new recruiting video for its Optimus humanoid robot programme, reiterating its claim that Optimus will be the single largest product in human history. The company is killing its luxury car line at the Fremont, California plant to make room for Optimus production, targeting one million units per year. Ground is expected to break within weeks on a new factory in Austin, Texas designed for 10 million units annually. Over 100 engineering positions are currently open, and a Gen 3 reveal is expected soon.

15. Grok Turns Teslas into AI Companions

Grok AI is now live in Tesla vehicles across Europe, Australia, and New Zealand, following a rollout via the 2026.2.6 software update. The integration allows drivers to have natural conversations with their vehicle — asking for restaurant recommendations, getting navigation assistance, and exploring topics hands-free. The feature requires Premium Connectivity or WiFi and is currently in beta. European deployment was designed to comply with GDPR, the Digital Services Act, and the EU AI Act.

16. America's First Robot Sports League

ProRL, America's first professional robotics sports league, will hold its inaugural Combine event on April 19 in Boston's Seaport District. Humanoid and quadruped robots will compete in speed races, obstacle courses, and precision challenges. A full 10- to 12-event season is planned for 2027. The concept follows China's lead, where similar robotics competition events drove massive public interest — Unitree reportedly saw a 9x increase in robot sales following its first sports event.

17. Bitcoin Miners Sell Holdings to Fund AI Pivot

Marathon Digital sold 15,133 Bitcoin (approximately $1.1 billion) over a three-week period in late February and early March 2026, using the proceeds to retire $1 billion in convertible debt at a 9% discount. The company is converting US mining facilities into AI-ready data centres in partnership with Starwood Capital, targeting 1 GW of near-term capacity and 2.5 GW potential.

Marathon is not alone. Multiple major mining operators — including Riot and Core Scientific — are pivoting infrastructure toward AI workloads. The logic is compelling: Bitcoin miners already have access to cheap energy and expertise in managing large-scale computing facilities. Swapping mining rigs for GPUs is a relatively straightforward infrastructure transition.

18. SpaceX IPO Filing Expected Imminently

Bloomberg reported on March 25 that SpaceX is preparing to file its S-1 prospectus, potentially as early as this week. The company is targeting a $75 billion raise at a $1.75 trillion valuation — which would make it the largest IPO in history. Morgan Stanley and Goldman Sachs are lead underwriters, with investor meetings planned after Easter and a target listing date of June 2026.

At a $1.75 trillion valuation, SpaceX would rank above every S&P 500 company except Apple, Microsoft, Alphabet, Amazon, and NVIDIA.