AI continued to dominate the news this week, no more so than at Nvidia’s annual GTC conference in San Jose, which arguably kicked off the real start of the next era of artificial intelligence.

“The conference for the Era of AI,” as the company called it, was jam-packed — so much so that CEO Jensen Huang (pictured) had to move his signature keynote to the nearby sports and concert venue SAP Center. The expo hall was gridlocked even by the third day, and one session, Huang’s fireside chat with seven of the eight authors of a seminal 2017 paper that launched generative AI into the tech mainstream, had to be capped at 1,800 people. Not least, Huang himself emerged as a major tech star, followed around for selfies by attendees and even, embarrassingly, some journalists.

Much more on that below, but that wasn’t even the only AI news, as new models from Apple, Elon Musk’s x.AI and more were released and new funding rounds continued for startups — even as earlier startups such as Inflection AI and Stability AI look to be struggling.

Meantime, it’s possible the initial public offering window is opening. Intel-backed Astera Labs and the social media site Reddit both saw their stock pop in their IPOs.

Not least, the Department of Justice’s hammer came down on Apple as it sued the iPhone maker on antitrust grounds.

This and other news, including the upshot of GTC, the rise of Broadcom as an AI powerhouse, and who’s going to create the leading AI factories, were discussed and analyzed in John Furrier’s and Dave Vellante’s weekly podcast theCUBE Pod, available now on YouTube. And catch Vellante’s weekly deep tech dive, Breaking Analysis, coming Saturday.

Here are the news highlights from this week:

Nvidia launches the real AI era

All the news and analysis from Nvidia’s GTC conference

First, some key points that stood out to me from our three days of onsite coverage in San Jose:

• GTC signaled nothing less than the emergence of a new computer industry centered on AI — what Jensen Huang calls generation, versus mostly retrieval, of information — along with all the new infrastructure and software required to make it happen — what Huang calls “AI factories.” Including, of course, all those big honking GPUs. “In the future, almost all of our computing experience will be generative,” Huang said in a press briefing. “We’re at the beginning of this platform shift to AI.” As Furrier put it, “The old world is pre-recorded. You’ve got the news, videos, everything is prerecorded and that means you’ve got to fetch it. Generative AI is the seeds of innovation, where data is seeds.”
• Tokens are emerging as a key component of the AI age — and a key point of competitive advantage. These are the basic units of data processed by AI large language models and perhaps, as Huang suggested, even movements of robots. “Jensen’s setting the table for a new infrastructure for AI and an operating system,” says Furrier. “And it’s built for tokens,” Vellante adds.
• 200 billion. That’s the number of transistors the Blackwell chip has. Two. Hundred. Billion. And even crazier, it’s not enough for powering coming AI models. “We need bigger GPUs,” Huang said.
• Despite the lower power requirements of Nvidia’s Blackwell chip, power will only become a bigger issue going forward. Whether it’s new algorithms and entirely new thinking such as the biology-inspired models from Sakana or more adept use of lower-power computing resources or new (and existing) chips that use much less power, a lot of effort is going into lowering AI power requirements. For example, analyst Sarbjeet Johal suggests that the ability of systems to switch between GPUs and CPUs as appropriate for the particular task will become important in coming years. After all, you don’t need a 70 billion-parameter model to add 2+2 — and still wonder if it produces the right answer.
• Artificial general intelligence is coming — just not the kind you think. Huang said AGI is coming in five years, but he defined it more narrowly than the doomers: “If we specified AGI to be something very specific, a set of tests where a software program can do very well — or maybe 8% better than most people — I believe we will get there within 5 years,” he told a press gathering. When one reporter asked about the notion of AGI exterminating humans — wondering if Huang could be seen as an AI Oppenheimer — Huang stared at her in a long pause. “Oppenheimer was building a bomb,” he replied slowly. “We’re not doing that.”
• Hallucinations will be solved. At least according to Huang, who says retrieval-augmented generation or RAG — essentially, requiring the chatbot to look up the answer before spitting out its result — should solve the problem. Given the architecture of today’s chatbots, though, it seems unlikely to be a complete solution. Then again, a lot of search results suck too.
• The march of the AI giants is beginning to crush the also-rans. Not just Nvidia but Microsoft famously credited with the policy of  “embrace, extend, extinguish”) with its gutting of Inflection AI’s leadership, as well as troubles at Stability AI and reported “tepid” revenue at Cohere. As fast as the rise of generative AI has been, it could be that the reckoning comes sooner than expected. Whoever wins, though, Nvidia benefits.
• Still, all this attention can be treacherous. Let’s see, market cap doubled to $2 trillion in the last nine months, and Ashton Kutcher, Trevor Noah, Kendrick Lamar showed up to the geekfest. Too early to call a top to Nvidia’s fortunes — way too early — but remember Silicon Graphics’ rise and fall?
• Indeed, AI is changing really fast, so today’s winning technologies may not be tomorrow’s. At a gathering of seven of the eight authors of the now-famous  “Attention Is All You Need” paper from 2017 that explained the Transformers architecture that’s the basis for most generative AI efforts today, co-author Aidan Gomez from Cohere said it’s already perhaps reaching its limits. “All of us here hope it gets succeeded by something that will carry us to a new plateau of performance,” he said. Kanjun Qui, co-founder and CEO of Imbue, even declared in a chat with Bryan Catanzaro, Nvidia’s VP of applied deep learning research: “Foundation models will be boring in the future.”
• AIs that can reason may be the next frontier. That’s what OpenAI Chief Operating Officer Brad Lightcap told Nvidia VP of Enterprise Computing Manuvir Das in a press chat he’s most excited about. “The way we see these systems evolve is a sort of reasoning agent,” he said. Models’ reasoning capabilities need improving and they need a way to work through multistep problems and take action.
• But enterprises are still in the nascent stages of using generative AI. “We don’t really do sales, we do therapy,” Lightcap said of customers talking to OpenAI. “‘Could AI fix all these things for me?’ Usually we have to talk them off the edge, get them some water.”

OK, on to our coverage:

Check out editorial interviews from Furrier, Vellante and others on theCUBE onsite this past week.

Blackwell: Nvidia’s GPU architecture to power new generation of 1T-parameter generative AI models

Keynote analysis from Nvidia GTC 2024: Spearheading AI and accelerated computing innovation

Nvidia GTC 2024 day two analysis: Generative AI emerges as the new seed of innovation

Nvidia releases Blackwell platform to go back to the future, extends partnership with AWS for scale

Nvidia’s new microservices APIs promise to speed up AI development

Nvidia unveils Project GR00T AI foundation model for humanoid robots

Nvidia announces APIs for Omniverse Cloud to power digital twins for software tools

Nvidia’s newest cloud service promises to accelerate quantum computing simulations

Nvidia’s latest generative AI model LATTE3D can create 3D images and shapes in seconds

Dell expands infrastructure portfolio with new Nvidia-powered AI platforms

HPE debuts its Nvidia GPU-powered on-premises supercomputer for generative AI

ServiceNow injects more generative AI capabilities into its workflow platform

CrowdStrike and Nvidia form strategic partnership to enhance cybersecurity with AI

Kinetica ramps up RAG for generative AI, empowering enterprises with real-time operational data

Dataloop and Nvidia collaborate to enhance AI application development for businesses

Meteomatics collaborates with Nvidia to enhance its hyper-local weather forecasts with AI

In other AI news

Report: Nvidia could pay up to $1B to acquire AI infrastructure startup Run:ai

Report: GPU cloud operator CoreWeave seeking new funding at $16B valuation

Databricks acquires text dataset management startup Lilac

Report: Apple in talks with Google to use Google Gemini AI model on iPhone

Google fined $272M by French government over AI use of news content

United Nations gives green light to first resolution on artificial intelligence

Elon Musk’s xAI releases Grok-1 architecture, while Apple advances multimodal AI research

Japanese startup Sakana releases AI models created through ‘evolutionary’ processes

Stability AI launches new model that turns images into 3D videos

Foundry launches with $80M in funding to build an AI-optimized public cloud

Healthcare industry chatbot firm Hippocratic AI raises $53M at $500M valuation

Startup Borderless AI nets $27M to bring generative AI to global hiring practices

Balbix’s BX4 engine leverages Nvidia AI to transform cybersecurity risk management

Google DeepMind develops an AI with Liverpool Football Club to predict plays in the game

Snowflake documents huge growth in AI projects

ClearML debuts open-source Fractional GPU tool and new monitoring features

DHS introduces AI pilots to enhance public safety and immigration processes

Around the cloud and enterprise

Catch theCUBE’s coverage of KubeCon+CloudNativeCon here, and there’s more to come. For the TL;DR, here’s a closing overview: Cloud-native community celebrates Kubernetes’ ‘Linux moment’ and its growing role in enterprise AI And a dive by Intellyx’s Jason Bloomberg into significant startups leading the way: Platform engineering, API platforms and intelligent agents dominate innovation at KubeCon Paris

IPO winter ending? Astera Labs and Reddit surge in stock trading debuts Databricks, Arctic Wolf and a few others will be watching closely.

Intel wins $19.5B in CHIPS Act funding and loans for fab network expansion

Cisco completes its $28B acquisition of Splunk

Micron’s stock posts huge gain after crushing forecasts in its latest earnings results

Qualcomm’s new Snapdragon 7+ Gen 3 chip brings generative AI to midrange smartphones

Microsoft debuts enterprise-focused Surface Pro 10 and Surface Laptop 6

Redis acquires storage engine startup Speedb to enhance its open-source database

Google releases second developer preview for Android 15

Technology pioneer Mike Stonebraker raises $8.5M to launch DBOS and radically transform cloud computing

Confluent simplifies integration between Kafka stream processing and Iceberg storage

Sonatype debuts SBOM Manager to make enterprise software more transparent

IBM acquires Pliant to boost network IT automation capabilities

Jama Software to be acquired by Francisco Partners for $1.2B in significant exit for Portland-based company (from Geekwire)

Cyber beat

Exclusive: Dymium launches with platform that takes protection to the data

Data security startup BigID valued at $1B+ following $60M round

Cyber risk startup CyberSaint raises $21M for platform development and expansion

Newcomer BlueFlag Security raises $11.5M for developer-centric security platform

Researchers uncover unfixable vulnerability in Apple CPUs affecting cryptographic security

Deloitte introduces CyberSphere to enhance cyber operational efficiency with automation and AI

Darwinium introduces behavior identification to strengthen online transaction security

Elsewhere around tech

DOJ sues Apple over antitrust violations related to iPhones

Neuralink’s first brain implant patient says playing chess with his mind is like ‘using the force’

Supreme Court tackles controversial topic of Biden administration-big tech cooperation

Microsoft will reportedly debut new Qualcomm-powered Surface devices in May

Figure Markets raises $60M to build crypto ‘everything marketplace’

Morph raises $20M to build consumer-focused Ethereum scaling blockchain solution

Comings and goings

Microsoft appoints Inflection AI CEO Mustafa Suleyman to lead its consumer AI unit Despite speculation that this was a way for Microsoft to avoid antitrust scrutiny, I’m thinking this will do just the opposite.

And that’s not the only departure from a high-profile AI company: Three of Stable Diffusion’s original developers reportedly leave Stability AI

Paul Cormier, chairman and former CEO of Red Hat, is retiring. 

Skyhigh Security appoints former Snow Software CEO Vishal Rao as new CEO Meantime, Amazon Web Services hired former Skyhigh CEO Gee Rittenhouse to be VP of enterprise security.

Google shuffles search leadership: Liz Reid, who was heading up core search experiences, is now head of Search. Cheenu Venkatachary is new lead of Search quality and ranking, replacing Pandu Nayak, who becomes chief scientist of Search. Cathy Edwards is moving to the Long-term Bets team in Knowledge and Information (from Search Engine Land).

Former Goldman Sachs executive Stephanie Cohen joined Cloudflare as chief strategy officer.

Photos: Robert Hof/SiliconANGLE