When it comes to web browsing, the familiar interface of Chrome or Safari has long dominated the mobile landscape. But for users seeking a new kind of intelligence at their fingertips, Perplexity is betting that AI can do more than answer questions—it can rethink the way we navigate the web.

On Thursday, the AI search startup launched its Comet browser for Android, expanding capabilities first introduced on desktop in July, and signaling a growing push to bring advanced AI tools directly into users’ hands.

The Comet browser on Android carries over much of the desktop version’s functionality. Users can set Perplexity as their default search engine, ask questions about open tabs, and use voice commands to have the assistant summarize information across all tabs. Beyond search, the AI assistant can research and shop on behalf of users, providing an experience that blends the familiar convenience of a browser with proactive task management. An in-built ad blocker further enhances usability, making the browsing experience smoother and less cluttered.

Perplexity is not stopping there. In the coming weeks, the company plans to introduce a suite of new features, including a conversational agent capable of searching across multiple sites, shortcuts that allow the assistant to take quick actions, and a fully functional password manager. These enhancements aim to make Comet a true AI-powered productivity companion, rather than just a search tool.

Android was chosen as the first mobile platform due to high demand from carriers and original equipment manufacturers, who have requested Comet for integration on their devices. While Perplexity previously partnered with Motorola to preload its AI assistant app, it remains unclear whether the new browser will receive similar treatment. An iOS version is also in development, though no timeline has been confirmed.

Perplexity’s move into mobile positions the company among an increasingly crowded field of AI browser competitors. OpenAI, Opera, and The Browser Company—now owned by Atlassian—have all developed AI-enabled browsers, though most remain desktop-focused. The Browser Company’s Arc Search was released for mobile last year, but development shifted to a new browser, Dia, which currently lacks a mobile version. This leaves Comet with an opportunity to define what AI-assisted browsing looks like on handheld devices.

Security, however, remains a critical concern. In October, Perplexity acknowledged the risks associated with AI agents in browsers, warning that new forms of AI-driven attacks could require a fundamental rethink of online security. Analysts have noted that while AI assistance can dramatically increase convenience, it may also expose users to new vulnerabilities if safeguards are insufficient. Striking the right balance between functionality and safety will be crucial for Perplexity’s success.

Incumbent browsers like Chrome and Safari dominate the global market, and dislodging them will require more than AI novelty, making the stakes high. Comet must deliver speed, reliability, privacy, and utility, proving that an AI-centric browsing experience can be both seamless and secure. The Comet browser is expected to reshape how people conduct searches, manage information, and interact with intelligent agents in daily life.

Android users, for now, have the first taste of this vision. Perplexity is betting that the combination of AI-driven search, productivity features, and proactive assistance will redefine mobile browsing. The company sees it not just as a tool for retrieving information, but as a platform that anticipates and supports the user at every step.

When Google DeepMind quietly brought in Aaron Saunders earlier this month, it wasn’t just another executive hire. It signaled an escalation. Saunders, the engineer who helped give the world its most famous back-flipping, crate-hauling, and occasionally dance-choreographed robots during his years at Boston Dynamics, is now DeepMind’s vice president of hardware engineering. And his arrival suggests the company is no longer content with producing only world-class AI models. It wants machines—real, physical machines—that can use them.

according to Wired, the hire folds directly into CEO Demis Hassabis’ sweeping ambition: to turn Gemini, DeepMind’s flagship model, into something like a robotics equivalent of Android. A base AI system that manufacturers everywhere could load into their humanoids, quadrupeds, warehouse bots, or whatever new chassis emerges next.

“You can sort of think of it as a bit like an Android play […] We want to build an AI system, a Gemini base, that can work almost out-of-the-box, across any body configuration,” Hassabis told WIRED. “Obviously humanoids, but nonhumanoids too.”

For Saunders, this is familiar terrain. Before becoming Boston Dynamics’ CTO in 2021, he led some of the company’s most complex engineering efforts, starting with an amphibious six-legged prototype and rising to VP of engineering in 2018. At Boston Dynamics—now majority-owned by Hyundai Motor Company after passing through SoftBank and, earlier, Alphabet itself—he helped steward the hardware behind the company’s signature creations: four-legged dog-sized robots and humanoids capable of highly precise and visually stunning acrobatics.

DeepMind, for its part, has spent years publishing robotics research rooted in large-scale reinforcement learning, imitation learning, and multimodal perception. But the industry’s recent surge of interest in humanoids has shifted that research from an academic project into something commercial. Hassabis said he’s convinced the field is nearing a tipping point. AI-powered robotics, he said, “is going to have its breakthrough moment in the next couple of years, if I was to predict.”

Outside DeepMind’s walls, the momentum has been unmistakable. In the U.S., startups such as Agility Robotics, Figure AI, 1x, and even Tesla have been racing toward humanoid production. Elon Musk said recently he intends to build a million Tesla Optimus robots over the next decade. In China, meanwhile, companies like Unitree have surged ahead on cost and scale; Unitree, based in Hangzhou, has already overtaken Boston Dynamics as the largest supplier of four-legged robots for industries ranging from construction to manufacturing.

Hassabis admitted Unitree’s progress has impressed him. But the true prize for DeepMind isn’t hardware supremacy—it’s cognitive supremacy.

“I’m most interested in the [AI] brain part of it,” he said, emphasizing Gemini’s multimodal structure.

The model’s ability to see, plan, parse language, and reason across different modalities, he argued, makes it especially suited to controlling robotic systems that need to operate in unpredictable physical environments.

That’s where Saunders’ arrival becomes strategically important. His deep, practical understanding of what real robots can and cannot do—how they behave, where they fail, and what they need at the hardware level—gives DeepMind a bridge from system-level AI to embodied intelligence. If Gemini is to become the Android of robotics, it needs precisely the sort of hardware awareness Saunders spent decades refining.

And the timing is advantageous as the components and expertise required to build legged machines have grown dramatically more accessible in recent years. Startups that once would have needed tens of millions in research grants to assemble a basic prototype can now buy actuators, sensors, and control modules off the shelf. The hardware ceiling is dropping while the software ceiling rises—an alignment that DeepMind sees as an opening.

The company, by hiring Saunders, is believed to be signaling that the next frontier for Gemini isn’t just better reasoning or smoother multimodal alignment, but embodiment. A brain that can live in many bodies, across many environments, and adapt on the fly.

Nothing in robotics ever moves as fast as the hype cycles surrounding it. But DeepMind’s latest move suggests the company sees the window opening—and intends to step through it with both hardware sophistication and AI horsepower in hand.

Brookfield Asset Management on Wednesday unveiled one of the largest financing initiatives yet for artificial intelligence infrastructure — a sweeping $100 billion program developed in partnership with Nvidia and the Kuwait Investment Authority (KIA).

The announcement lands at a moment when demand for computing power, energy supply, and physical space for AI workloads is rising at a pace that global investors now describe as unprecedented.

The backbone of the initiative is the newly launched Brookfield Artificial Intelligence Infrastructure Fund, which targets $10 billion in equity commitments. Brookfield said the fund has already secured half of that amount, with $5 billion pledged by Brookfield itself, Nvidia, and KIA. With co-investor capital and financing layered on top, the fund is expected to acquire up to $100 billion worth of AI infrastructure assets over time — a portfolio spanning energy facilities, land, hyperscale data centers, and compute capacity.

Brookfield, a heavyweight in global infrastructure and alternative investments, noted that it has already deployed more than €100 billion across the AI value chain, from digital infrastructure to renewable power and semiconductor manufacturing. The new program pushes the company deeper into the heart of an arms race among tech firms, chipmakers, and cloud providers, all scrambling to secure the physical backbone needed for next-generation AI systems.

A rapid surge in AI adoption is intensifying the scramble for these resources. Companies across industries are racing to lock in access to the high-performance computing clusters that train and run large models, and to secure the electricity required to power them. This has created a boom in data center construction, land acquisition, and long-term energy contracting, turning infrastructure — not just chips — into a defining bottleneck for AI development.

Sikander Rashid, Brookfield’s head of AI infrastructure, called the moment “one of the largest infrastructure buildouts in history” and warned that the world will need an estimated $7 trillion in capital over the next decade to keep pace with AI demand. That figure underlines the energy-intensive nature of modern AI models. Training and running them requires enormous volumes of power, far outstripping the capacity built during previous waves of digital expansion.

The fund’s first commitments are already underway. They include a $5 billion framework agreement with Bloom Energy to deploy up to 1 gigawatt of behind-the-meter power solutions for data centers and AI factories — a move aimed at supplementing strained power grids and reducing exposure to energy shortages. Behind-the-meter systems allow operators to generate power on-site, which has become critical for facilities that cannot afford downtime or an unstable supply.

Brookfield has also been moving aggressively in Europe. Earlier this year, it announced plans to invest up to 95 billion Swedish crowns (about $10 billion) in an AI data center campus in Sweden, tapping into the region’s reliable energy supply and cooling advantages. In France, the company committed €20 billion to AI-related projects, including facilities that support cloud providers and advanced compute clusters.

The partnership with Nvidia marks another sign of the chipmaker’s widening role not only as a supplier of GPUs but as a strategic investor shaping the broader AI ecosystem. As the world’s largest designer of AI chips, Nvidia has been pushing into data centers, networking technologies, and infrastructure development to ensure that the companies training large models have the capacity required for its hardware.

The involvement of the Kuwait Investment Authority highlights the growing interest among sovereign wealth funds in the long-term economics of AI infrastructure. These investors, including funds from the Middle East, Europe, and Asia, have been pouring billions into data centers, energy assets, and semiconductor supply chains, viewing AI as a decades-long growth frontier.

Brookfield’s new program amplifies the investment wave sweeping across the sector, underscoring that the AI era is no longer defined solely by algorithmic breakthroughs or frontier models. It is increasingly shaped by an industrial-scale buildout of land, power, and compute.