PRODUCTS

Google unveils Gemini 3 in “path toward AGI”

On Tuesday, Google took what it calls “another big step on the path toward AGI” and unveiled Gemini 3, the latest version of its flagship frontier model.  

The model family is Google’s most advanced release to date, demonstrating state-of-the-art reasoning, vision and spatial understanding. The model outperforms previous Gemini generations and competitors’ models across major AI benchmarks, including academic reasoning, multimodal understanding, agentic tool use and long context performance. 

Along with debuting the model itself, Google announced a slew of other features powered by Gemini 3, including:

• Google Antigravity, an AI coding platform that uses agents as an “active partner” rather than a tool;

• Gemini 3 Deep Think, a research-focused version of the model featuring enhanced reasoning, Gemini 3 Pro, combining reasoning with multimodal capabilities and bringing a “new level of depth and nuance to every interaction,” both available in preview before going to Google AI Ultra subscribers; 

• And AI Mode is available in Search using Gemini 3, bringing new generative UI experiences such as immersive layouts and interactive tools. 

Google also touted considerable monthly user numbers, claiming 2 billion AI Overviews users, 650 million Gemini app users, and 13 million developers using its generative models. 

“Ultimately, this is a key component of what we see as an eventual AGI system,” Demis Hassabis, CEO of Google DeepMind, told journalist Alex Heath in an interview for the Sources newsletter. “It needs to do everything really well.” 

Gemini 3 follows an influx of model releases from major providers this fall, including OpenAI’s GPT-5.1, xAI’s Grok 4.1, and Anthropic's Claude Sonnet 4.5. But as these models get bigger and better in the pursuit of AGI, so are the concerns that the market is nearing an inflection point, as rumblings of an AI bubble grow louder. 

It’s getting to a point where tech leaders themselves can’t ignore it: Google CEO Sundar Pichai told the BBC in an interview published Tuesday that, though the impact of AI will be profound, similar to the adoption of the Internet despite the dotcom crash in the late 90s, the astonishing trillion-dollar investment in the market has “elements of irrationality.” 

A market retraction would likely come before any hint that we are even remotely close to the AGI that these companies are desperately vying for. Whether these companies will be able to pick up the pieces and continue their efforts in the fallout is yet to be determined. Though Pichai said he was confident that Google could survive a burst, “I think no company is going to be immune, including us,” he told the BBC.

TOGETHER WITH WARP

Warp’s Big Agents Update Catapults It to #1 on Terminal-Bench

Warp just launched it's biggest Agents update yet, propelling it to #1 on Terminal-Bench and #5 on SWE-bench Verified ahead of Claude Code and Codex CLI.

Warp's #1 Agent now includes:

✅ Full Terminal Use: agents can interact with long-running commands like running servers, debuggers, and more

✅ Supercharged Planning: Agents are steerable -- refer to and edit plans as the agent works

✅ Full lifecycle: planning → coding → deployment and more

Warp is emerging as the platform of choice for developers, trusted by 700K+ users and 56% of the Fortune 500.

Try Warp’s Agents for yourself →

BIG TECH

OpenAI may be weighed down by “cost to serve”

OpenAI just keeps spinning more plates.

The AI firm on Tuesday announced a multiyear partnership with Intuit, the financial firm behind TurboTax, QuickBooks, Credit Karma and Mailchimp, to give the company access to OpenAI’s APIs and expose ChatGPT users to Intuit products. The deal will generate more than $100 million for OpenAI over an unspecified period.

These kinds of enterprise deals highlight OpenAI’s horizontal approach to the AI market: The company seeks to have its hands in everything, from browsers to consumer apps to enterprise workflows to finance. But in doing so, it’s spending a lot of cash.

• The company has committed to shelling out more than $1 trillion on developing AI infrastructure over the next eight years through a series of deals with the likes of Oracle, SoftBank, Nvidia, Amazon and more. 

• These deals may already be dragging down OpenAI’s partners. According to the Financial Times, Oracle might already be underwater following its $300 billion deal with the AI firm.

• And the costs of running the models they already have is getting steeper, too: according to data reported last week by tech blogger Ed Zitron, OpenAI’s inference costs from the first through third quarter hit $8.67 billion, with revenues of $4.33 billion during that period (though Altman has claimed revenues are “well above” $13 billion and expected annualized revenue run rate of $20 billion by the end of the year.) 

Mixed reports aside, OpenAI has committed itself to spending far more than it’s earning, tying its fate to other tech giants in the process, in lofty hopes of developing large swathes of AI infrastructure. The costs may overwhelm them before these projects can even be completed, Scott Bickley, advisory fellow at Info-Tech Research Group, told The Deep View. 

With fears of an AI bubble reaching a fever pitch, valuations and expectations of AI may face a harsh course correction, Bickley said, leading to data center buildouts being “vastly curtailed.” Even before constraints like supply chain and energy constraints come to the fore, “I think just the cost to serve will overwhelm that economic model before anything else,” he said. 

Rather than trying to be everything to everyone (and spending billions in the process), OpenAI might be better off focusing on where it’s deriving the most value, such as enterprise deals. (Clearly, it’s working for rival Anthropic.) 

TOGETHER WITH BLAND

There’s an AI company without churn?

In San Francisco, Bland AI is providing services to enterprises that have yielded them 127% Net Revenue Retention.

They’re offering conversational AI agents that handle inbound calls, SMS, and webchat for large businesses, and apparently, it’s pretty good.

If you want to see for yourself, you can get a free custom agent built for your business today.

STARTUPS

Lambda raise signals neocloud hot streak

Neoclouds just keep winning. 

On Tuesday, AI cloud company Lambda announced a $1.5 billion Series G funding round led by TWG Global. The deal comes just weeks after Lambda reached a “multibillion-dollar” agreement with Microsoft to deploy tens of thousands of Nvidia GPUs. 

“Lambda sits at the intersection of AI’s explosive growth and the infrastructure gap,” TWG’s Thomas Tull told The Wall Street Journal. 

Lambda’s funding marks another milestone for neocloud providers as the AI boom drives up demand for AI-specific hardware. According to Synergy Research Group, neocloud revenues are projected to exceed $23 billion this year. 

• CoreWeave has signed deal after deal with companies like Meta, OpenAI and Nvidia after going public in March. 

• In October, Nscale announced an IPO and a partnership with Microsoft worth $14 billion, just two weeks after announcing two funding rounds within days of each other. 

• Nebius scored a $19.4 billion partnership with Microsoft in September and a $3 billion deal with Meta last week. 

• And Lambda in September scored a deal with Nvidia worth $1.5 billion to lease back its GPUs from the startup. 

Though these deals might look like trouble for conventional cloud hyperscalers like Amazon, Microsoft and Google, the landscape more likely resembles “co-opetition” rather than sheer rivalry, Trevor Morgan, chief operating officer of OpenDrives, told The Deep View. 

“I speculate they're in a ‘watch and wait’ period while they try to maintain their current business in what's kind of an uncertain economic climate,” said Trevor. 

So what happens to these neoclouds amid a market retraction? That’s going to depend on how much these neoclouds are relying on a “pure AI play,” and whether they have enough substance to withstand turbulence, said Morgan, and which is the most balanced, realistic and diversified as far as customers and growth trajectories.

 “Even if a bubble bursts, that doesn't mean that that's bad for a player who strikes that balance perfectly well,” said Morgan. “Hopefully they're planning for that.” 

LINKS

• Google opens data center in Winschoten to power Netherlands' AI

• Meta’s Vibes AI slop app reportedly has 2 million daily active users

• Databricks reportedly is in talks to raise funds at a valuation of over $130 billion 

• Meta's chief revenue officer, John Hegeman, is leaving the company

• Waymo launches autonomous driving in five new cities in Florida and Texas

• PyTorch creator Soumith Chintala joins Thinking Machines Lab

• Microsoft to add Copilot, AI agents to Windows

• Microsoft Agent Factory and 365: Debuted at Microsoft Ignite, the company now offers agent building and orchestration platforms.

• Ramp Sheets: An AI spreadsheet editor by fintech firm Ramp for finance teams.

• Replit Design: An AI-powered design tool that creates professional-grade work. 

• Deep Research Tulu: An open-source end-to-end tool for deep research 

• Poe Group Chats: Quora’s AI app now allows users to collaborate with other users on more than 200 AI models

• Google DeepMind: Senior Research Engineer, Responsibility

• Perplexity: AI Researcher

• Cisco: AI Researcher

• AMD: Applied ML researcher, Generative AI - Advanced Graphics Programs

GAMES

A QUICK POLL BEFORE YOU GO