The AI Daily Brief: Artificial Intelligence News and Analysis - What Manus and Groq Acquisitions Tell Us About AI

Episode Date: January 3, 2026

Two blockbuster deals over the holidays quietly marked the real start of the AI agent era, revealing where competition is actually heading in 2026. This episode breaks down why Meta’s acquisition of Manus signals a shift toward agents as distribution, not features, and why Nvidia’s $20B Groq deal is really about owning the future of inference as workloads fragment and latency becomes decisive. Together, these moves show how the battle is moving from models and benchmarks to agents, infrastructure, and the interfaces people refuse to leave. In the headlines: xAI’s massive compute expansion, OpenAI’s renewed push on voice and devices, SoftBank’s infrastructure spree, Brookfield’s AI cloud ambitions, and Claude Code reaching the point of writing all of its own code. Brought to you by:KPMG – Discover how AI is transforming possibility into reality. Tune into the new KPMG 'You Can with AI' podcast and unlock insights that will inform smarter decisions inside your enterprise. Listen now and start shaping your future with every episode. ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://www.kpmg.us/AIpodcasts⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Blitzy.com - Go to ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://blitzy.com/⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ to build enterprise software in days, not months Robots & Pencils - Cloud-native AI solutions that power results ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://robotsandpencils.com/⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠The Agent Readiness Audit from Superintelligent - Go to ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://besuper.ai/ ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠to request your company's agent readiness score.The AI Daily Brief helps you understand the most important news and discussions in AI. Subscribe to the podcast version of The AI Daily Brief wherever you listen: https://pod.link/1680633614Interested in sponsoring the show? [e-mail]

Today on the AI Daily Brief, what two massive acquisitions tell us about the state of AI competition. Before that in the headlines, Claude Code is now writing 100% of ClaudeCode Code. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. All right, friends, welcome to 2026, quick announcements before we dive in. First of all, thank you to today's sponsors, KPMG, Zen Coder, and Superintelligent. To get an ad-free version of the show, which is, of course, just $3 a month, go to Patreon.com slash AI Daily Brief, or you can subscribe on Apple Podcasts. If you're interested in sponsoring the show, send us a note at Sponsors at AIDilybrief.A.I.

And a couple of other quick housekeeping things. First of all, for those of you who missed the results of our AI-R-OI benchmarking survey, you can find more information about that at AIDBIntel.com. You can also sign up to join our AI tracking panel. We're going to be moving to do a lot more original research this year, which hopefully gives everyone access to much better benchmarks around how AI adoption and performance is going. And so I would love for you to contribute to that tracking panel.

Again, that's AIDB Intel.com. Lastly, you might have heard me talk about this in our New Year's episode, but to help everyone kick off the year with some practical AI skills upgrades, we've got a free 10-week program that's basically a set of weekend projects that will give you exposure to a lot of different aspects of AI. We've had a pretty phenomenal response so far with more than 700 people signing up to participate. And so I think we're going to spin up a whole community around it.

To get all the information about that, go to AIDB New Year. And with that, let's get into today's episode. Today's most interesting story to me actually isn't news. It's about ClaudeCodeCode, but there was so much news that's happened over the last week or so as we've been in holiday episodes that I got to rip through a bunch of stories before we get into that, starting with XAI continuing to double down on compute with the purchase of a third building to expand their facilities outside of Memphis. Now, if you guys were listening closely when I talked about GROC as part of my 2026 prediction,

I said that basically they were going to have to do something to break out from the back of the pack. I was not, however, pessimistic about their ability to do so. And the thing I pointed out as the most likely contender for how they could start to do that is basically taking advantage of more access to compute through Elon's fundraising and operational prowess, and already we have a story that points in that direction. The information reports that XAI has purchased a large warehouse in southern Mississippi, just over the border and a few miles south of their existing data centers. At the moment, XAI has one data center operational.

That is their Colossus supercluster, which was built rapidly in 2024. After rolling expansions, it now has around 230,000 GPUs operational in a single coherent training cluster, making it the largest in the world. Alongside Colossus in the same industrial park, the Colossus 2 data center is still under construction. In July, Elon Musk said the goal is to install 550,000 Blackwell GPUs and that the first deliveries were underway. XAI now says that they have 450,000 GPS.

CPUs operational across their facilities. The third facility is still at its earliest stages, but Elon Musk is clearly setting his sites on dominating training compute. Confirming the reports earlier this week he posted, XAI has bought a third building called Macro Harder. We'll take XAI's training compute to almost two gigawatts. Now, Musk is seemingly referring to plans to build an AI-first Microsoft replacement called Macro Hard, as opposed to Microsoft, get it, but he also might just really enjoy the joke.

So far, none of the hyperscalers have completed even a one-gigawatt data center, but many, including OpenAI, are racing towards this milestone in 2026. Alongside their third data center, XAI is also making progress in constructing their own natural gas power plant in the surrounding area. This will be one of the first power plants built specifically to power AI infrastructure. Next up, some model news. OpenAI is renewing their focus on audio models, seemingly in preparation to release their first consumer device. Once again, according to the information, OpenAI has consolidated engineering, product, and research teams to overhaul their audio models. The report stated a new audio model to drive voice mode is expected to be released in the first quarter of this year.

Citing sources with knowledge of the project, the information wrote that the model will, quote, sound more natural and emotive and provide more accurate in-depth answers. It will reportedly handle interruption more easily and can even speak over the user when appropriate, something current generation voice models can't do. Now, the assumption is, of course, that the model is a key part of OpenAI's Johnny I've designed consumer device, which is expected to arrive in about a year. And even if the form factor is still a little uncertain, it's pretty clear that Sam Altman and Johnny I believe a voice-only interface is the correct move. We also continue to get reports at various levels of verification around what OpenAI has planned for their device.

One recent report suggested that it's a pen-shaped device, although there also might be multiple form factors. One interesting sub-detail is that, according to Satrini analysts, you can, while the device was originally expected to be contract manufactured by China's luck share, due to strategic considerations around a non-China supply chain, OpenAI has shifted course and is now looking for ways to manufacture it outside of China. Speaking of non-China AI supply chains, Nvidia has closed their deal to invest $5 billion into Intel. The deal was struck back in September with Nvidia securing a price of 2328 per share in a private placement. At the time, that was a slight discount to the market price, but Intel stock is now up

50% since the deal was announced, making the deal even better for Nvidia. Nvidia will now own a roughly 4% stake in Intel, and more importantly, will have a vested interest in supporting a revival in their foundry business. AI chipmaking is capacity constrained at the moment, so the ability to bring new fabs online in the U.S. is key to Nvidia expanding their production. For Intel, the deal is viewed as a major financial lifeline for a company that's been facing a severe capital restriction. Staying on deal for a moment, SoftBank is stepping up their AI investments with a new $4 billion deal to acquire Digital Bridge. Digital Bridge is a private equity firm heavily involved in data center funding. The all-cash deal will see SoftBank acquire the entire firm, paying a 15% premium to their public market valuation from Monday's announcement. SoftBank CEO Masayoshi Sun said in a statement, as AI transforms industries worldwide, we need more compute, connectivity, power, and scalable infrastructure.

Digital Bridge is a leader in digital infrastructure and this acquisition will strengthen the foundation for next generation AI data centers. Now, the acquisition is clearly part and parcel of SoftBank's larger AI buildout. The firm partnered on OpenAI's Project Stargate at the beginning of last year. then over the summer, a string of reports suggested that funding was an issue. Now, SoftBank will have an in-house private equity partner to ensure a pipeline of funding to their AI projects. Digital Bridge currently has around $108 billion in infrastructure deals on their books, which include cellular towers and fiber optic networks as well as AI data centers. Digital Bridge will still fund their projects through outside investors,

meaning that SoftBank could have greater access to capital. Separately, SoftBank confirmed on New Year's Eve that they completed their $40 billion investment in OpenAI. A final payment of $22.5 billion was due by the end of the year, but reports suggested it was far from a smooth process. SoftBank sold their $5.8 billion stake in Nvidia and $4.8 billion in T-Mobile to fund the deal. On top of that in mid-December, Reuters reported that SoftBank was tapping margin loans against their armstock in a last-minute scramble to come up with the cash. SoftBank doesn't lack assets but was liquidity constrained after the government shut down

delayed the IPO of a portfolio company called Pay Pay Pay, which was expected to net $20 billion for them. With the deal now closed, SoftBank owns roughly 11% of OpenAI and seems to be eager for more AI deal-making. Meanwhile, Canadian asset management giant Brookfield is spinning off their own cloud business to take advantage of the AI boom. The information reports the new business will be tied to Brookfield's AI Infrastructure Fund, which was launched in November. The fund will have a cap of $100 billion, but currently has $10 billion in commitments from investors including Nvidia and the Kuwait Investment Authority. The fund is currently developing data centers in France, Qatar, and Sweden. Overall, the idea is to lower the cost of AI infrastructure by leveraging Brookfield's scale and vertical integration. The firm has over a trillion dollars in assets under management, including a heavy and

emphasis on energy and real estate. writes Reuters, a cloud business would allow the company to control inputs of the AI value chain in a way inaccessible to pureplay cloud providers. Finally, what I said was most interesting to me in today's headlines is Claude Code writing 100% of Claude Code Code. The rapid growth of AI coding was, of course, one of the key inflection points for 2025, and some of the creators of the technology are astounded at how far it's come. Claude Code creator Boris Charney posted over the holiday break,

A year ago, Claude struggled to generate bash commands without escaping issues. It worked for seconds or minutes at a time. Fast forward to today. In the last 30 days, I landed 259 PRs, 497 commits, 40,000 lines added, 38,000 lines removed. Every single line was written by Claude Code in Opus 4.5. Claude consistently runs for minutes, hours, and days at a time. Software engineering is changing, and we're entering a new period in coding history, and we're still just getting started.

Now, the comments caused some to do a double-take, asking Chirney if he really meant that he hadn't manually written code in the last month. He responded, correct, in the last 30 days, 100% of my contributions to Claude Code were written by Claude Cod Cod. Ethan Malick wrote, in retrospect, the articles mocking Dario Amade's prediction of 90% of code being written by AI by September seemed to be very misguided. He seems to have only been off by a couple months, if that. And indeed, less than a year after Andre Carpathie coined the term vibe coding, Claude Code is now good enough. to write Claude Code. Speaking of Carpathy, he went viral over the holidays for a take on the rapid advancement in AI coding. He wrote, I've never felt this much behind as a programmer. The

profession is being dramatically refactored as the bits contributed by the programmer are increasingly sparse in between. I have a sense that I could be 10x more powerful if I just properly string together what has become available over the last year. And a failure to claim the boost feels decidedly like skill issue. There's a new programmable layer of abstraction to master, in addition to the usual layers below, involving agents, subagents, their prompts, context, memory, modes, permissions, tools, plugins, skills, hooks, MCP, LSP, LSP, slash commands, workflows, IDE integrations, and a need to build an all-encompassing mental model for strength and pitfalls of fundamentally stochastic, failable, unintelligible, and changing entities,

suddenly intermingled with what used to be good old-fashioned engineering. Clearly, some powerful alien tool was handed around except it comes with no manual, and everyone has to figure out how to hold it and operate it, while the resulting magnitude 9 earthquake is rocking the profession. Andre ends with the most salient advice for the moment, roll up your sleeves to not fall behind. That will, of course, be one of the key themes of the AI Daily Brief this year. For now, that is going to do it for today's headlines.

Next up, the main episode. All right, let's talk about the signal versus the noise in Enterprise AI. The challenge right now isn't just about what's possible, it's about what's practical. That's the entire focus of the You Can With AI podcast I host for KPMG. Season one, cut through the hype to focus on deployment and responsible scaling. Season two goes a level deeper. We're bringing together panels of AI builders, clients, and KPMG leaders to debate the strategic questions that will define what's next for AI in the enterprise.

Six episodes packed with frameworks you can actually use. Find you can with AI wherever you get your podcasts. Subscribe now so you don't miss the new season. If you're using AI to code, ask yourself, are you building software or are you just playing prompt roulette? We know that unstructured prompting works at first, but eventually it leads to AI slop and technical debt. Enter Zenflow. Zenflow takes you from vibe coding to AI-first engineering. It's the first AI orchestration layer that brings discipline to the chaos. It transforms freeform prompting into spec-driven workflows and multi-agent verification, where agents actually cross-check each other to prevent

drift. You can even command a fleet of parallel agents to implement features in fixed bugs simultaneously. We've seen teams accelerate delivery 2x to 10x. Stop gambling with prompts. Start orchestrating your AI. Turn raw speed into reliable production-grade output at zenflow.3. Today's episode is brought to you by my company Superintelligent. In 2026, one of the key themes in enterprise AI, if not the key theme, is going to be how good is the infrastructure into which you are putting AI in agents? Superintelligence agent readiness audits are specifically designed to help you figure out, one, where and how AI and agents can maximize business impact for you, and two, what you need to do to set up your organization to be best able to leverage those new gains.

If you want to truly take advantage of how AI and agents can not only enhance productivity, but actually fundamentally change outcomes in measurable ways in your business this year, go to B-Supert.aI. Welcome back to the AI Daily Brief. Today we are discussing what two major acquisitions tell us about the state of AI competition. All right, friends, we are back with the first main episode of the AI Daily Brief of 2026. and you might have heard a few days ago, me drop my two episodes set about my AI predictions for the next year. Before the proverbial ink was dry on that episode, one or kind of maybe

even two of them had already start to come to pass. I'm talking, of course, about the prediction that the first leading crop of generalist AI agent companies, specifically Jen Spark and Manus, were going to be massive acquisition targets for the big hyperscalers and labs in 2026. The logic was not about any sort of short-term need from Gen Spark and Manus, Both of those companies were doing extremely well, seeing their revenue grow incredibly quickly, presumably having access to lots and lots of private capital. But at the same time, knowing that they were in a space that was going to be directly in the line of sight for all of the big labs.

As the companies who are pushing the first generation of actually performant general purpose agents, they were in many ways softening the ground for the sort of interfaces and experiences that are presumably going to become a key part of what those major labs and current chatbots ultimately offer. Ultimately, my bet was and is that despite those companies racing to nine figures in ARR in just a number of months, they're still going to be staring down the barrel of competition so intense that I think it will make sense for them, from a strategic perspective, to get acquired by one of those partners. And obviously, I think from the perspective of the acquirers, getting all of that lived experience around how people are actually interacting with

agents and what for, is going to be worth effectively whatever price they pay for it. As it turns out, the first company to go was Manus. Just before the end of the year, news broke that Mark Zuckerberg's meta would be buying Manus for more than $2 billion. The former scale leader, now Metis chief AI officer Alexander Wang tweeted, excited to announce that Manus has joined meta to help us build amazing AI products. The Manus team in Singapore are world class at exploring the capability overhang of today's models to scaffold powerful agents. Now, by way of background on Manis, You might remember that at the beginning of 2025, a number of people thought that Manus's launch was sort of the Deepseek Moment 2.0. What I mean by that is that in January, when Deepseek released

their R1 model and their companion chatbot app to go with it, it really awoke people to the potential of Chinese labs as major competitors. A couple months later in March, Manus's first general-purpose agent launch went completely viral, although it was nearly impossible to get an invite code. Building on that momentum, Manus raised money in April at a $500 million valuation with the round being led by benchmark, an investment that was somewhat controversial because of Manus's Chinese origins. Now, nine months on from that, Manus has proved that they were not just a hyped-up launch. In December, the company claimed a 125 million revenue run rate, and going from zero to 100 million in eight months, by some estimates, makes them the fastest growing startup of that scale in history.

Now, it's very clear that in spite of all that, Manus' Chinese roots continue to loom large. over the deal. Manus was originally launched out of offices in Beijing and Wuhan to a largely western user base, and the company quickly relocated to Singapore to distance themselves from the U.S.-China AI conflict. Meta went to great length to get ahead of the issue, providing a statement that said there will be no continuing Chinese ownership interest in Manus AI following the transaction, and Manus will discontinue its services and operations in China. Still, Manus's CEO is a Chinese national and will now take a prominent AI role at one of the largest U.S. tech companies. From the Chinese perspective, the acquisition is a huge validation of the Chinese AI startup ecosystem. Li Jing,

the founder of a Chinese startup incubator, told Bloomberg, this is truly an exhilarating event, a big era that belongs to China's startup founders. Entrepreneur Huang Dongshu said, it's the best gift for the start of 2026. This is among the most significant news in recent times, a real boost for startup founders of Chinese ethnicity, especially those building businesses overseas. Tony Pang, the writer of the Recode China AI newsletter, suggested that Manus has created a new playbook for Chinese founded startups, writing, This isn't just another normal acquisition story. It's a blueprint for how a new generation of Chinese entrepreneurs can build world-class

AI products, win over global capital and tech companies, and execute a clean exit. It's also a microscope through which we can observe the latest dynamics of U.S.-China AI competition, where talent and technology flows across borders, even as geopolitical walls rise higher. Po Zhao wrote, China trains AI users but exports AI founders. Manus just became the latest proof. In another tweet, he wrote, The question everyone in Chinese tech is asking, what if Manus had stayed instead of relocating to Singapore?

The answer is uncomfortable but clear. In China's AI app market, big tech controls 70% of the top 20. White Dance launched 11 AI products in 2024 alone. When a startup's product goes viral, incumbents cloned in days. Manus relocated 40 core engineers to Singapore. The move was a survival decision. The Singapore relocation gave Manus something critical, defensible traction. That's what meta valued.

Now, holding aside the geopolitical dimension of this, the more interesting questions to me, frankly, are about the product itself and what it means for meta strategy. The product will continue to operate, with Manna's CEO Xiaohong stating, joining META allows us to build on a stronger, more sustainable foundation without changing how Manis works or how decisions are made. Tech analysts Rahad Jark wrote, Meta has just opened the floodgate for the AI agentic application layer. He goes on to argue that Manus is more than just an LLM wrapper.

Manus, unlike Chatchapit, he writes, was built to execute tasks rather than provide text answers. The goal is to assign it a high-level task so the agent can navigate different tasks autonomously to complete the job. The unique part is that instead of just talking about a problem, Manus writes a Python script on the fly to solve it, executes that script in a secure sandbox, and looks at the result. Now, in this way, it actually brings up another one of my predictions of meta-reentering the AI competition conversation in a big way this year. Basically, my argument was that if 2025 was a rebuilding year with the recruitment of the superintelligence team and the changes to how AI was organized internally, we were going to see in 26, the manifestation

of that strategy come to the four. Now, I don't think it's exactly clear what part of this whole pie meta is going to go after. But perhaps with this Manus acquisition, we're starting to get a picture of what that might look like. Rehard again continues, this best fits into Meta's WhatsApp as an assistant that they can offer both to consumers and businesses, and a strong play for their meta-rayband smart glasses where you need an autonomous agenic system to run those glasses. Ben Palladian writes, Manus wasn't a vibes higher, its capability overhang to scaffolding to real agents. This is how chatbots turn into labor. And I think some people's interpretation is that this is going to be meta moving more into the enterprise and getting work done side of things. But I'm not so sure.

I think First Mark's Matt Turk is a little closer when he writes, if you're Amazon, you need your Manus. If you're Shopify, you need your Manus. If your bookings, you need your Manus. If you're bookings, you need your Manus. If you're a big consumer and commerce brand and don't own a major LLM, you need to build or acquire an agent because consumer intent is going away from consumer apps. And so the point here is what I'm using Manus's general purpose capabilities for right now, i.e. building slide presentations and things like that, is probably not what Meta is interested in using Manus 4 in the future. To the extent that Matt is right and consumer intent is moving away from consumer apps, and we will increasingly in the future be deploying agents on our behalf

to do the things that we do now around e-commerce and interacting financially on the internet. This is a way for meta to build the next generation way that it's billions of users continue to use it as their starting point for everything that touches commerce on the internet. Sean Chahan writes, meta didn't pay $2 billion for Manus's technology. They paid for eight months of distribution proof. Open AI has better models, Anthropic has better reasoning. but neither owns the workflow where 3 billion people already live.

The agent war won't be won in benchmarks. It will be one in the apps users refuse to leave. Distribution is the new moat. Model quality is table stakes. I don't think we know exactly how it's going to play out yet. I don't even think that meta necessarily knows. I just think that they knew that general purpose agents are going to be an increasingly important part of not just the AI battle, but the internet landscape in general, and that by buying Manus for what is ultimately an incredibly cheap price, frankly, they were going to get a massive head start in this essential area. Now, the second big story of the break period was also an acquisition, and this one happened just before Christmas.

Well, technically, it's a licensing deal, but honestly, it's an acquisition. Let's be clear. I'm talking, of course, about Nvidia agreeing to a licensing deal, with the biggest air quotes you can possibly imagine with chipmaker Grock, paying them $20 billion for the use of their technology and the acquisition of several key executives. Grock, which is spelled with a cue, not to be mistaken, to Elon Musk's chatbot, GROC with a K, is a decade-old chip startup. The company was founded by former Google executive Jonathan Ross, who helped invent Google's TPU chip architecture. He took that knowledge to GROC and focused on producing high-speed inference chips.

Now, at this stage, GROC has carved out a small market share, largely producing chips for NeoCloud, servicing customers with specific latency needs. Their chips aren't necessarily better than NVIDIA's general-purpose GPUs, but they can be as much as 10 times faster at producing tokens during inference. Jonathan Ross is among the executives who will be joining NVIDIA, leaving GROC to continue as an independent company. That means, of course, that NVIDIA will now have the creator of the TPU in-house working on inference optimization. It's also not exactly clear how much of a company will be left over once the deal is closed, but despite initial concerns that this is going

to be another deal where the top executives get a major payday and the employees get left in a lurch, it appears that that actually won't be the case. Axios's Dan Primac tweeted, been a bunch of chatter about how GROC employees made out in the NVIDIA deal, made some calls to find out, in short, very, very well, even if not fully vested. Specifically, it sounds like around 90% of GROC employees are said to be joining NVIDIA and will be paid cash for all vested shares. Unvested shares will be paid out at the $20 billion valuation, but via NVIDIA stock that vests on its own schedule.

So what is this acquisition about? Some of the early chatter suggested it was simply about NVIDIA snuffing out the competition. and I don't think in this case that that's really accurate. At 20 billion, it's the largest acquisition in Nvidia's history and large enough to rank as a top 15 tech acquisition. It's roughly similar in size to the WhatsApp, Slack, and LinkedIn acquisitions. The sheer size of the deal has Wall Street concerned. Given that it was framed as a non-exclusive licensing agreement,

that raised a lot of red flags for investors who were already concerned about NVIDIA's valuation. Invidia's stock struggled over holiday trading sessions, suggesting that there isn't very much enthusiasm for the deal. Still, UBS nailed their colors to the mast and reiterated their buy rating for Nvidia just before the new year. They wrote that the deal, while coming at a substantial price tag, could, quote, bolster Nvidia's ability to service high-speed inference applications, an area where GPUs are not ideally suited because of all the off-chip high bandwidth memory.

This would also be one of the fastest growing parts of the inference market, and we see this is another pivot to offering ASIC-like architectures in addition to its mainstream GPU roadmap. Now, despite this being technical, it's worth unpacking just a little bit. NVIDIA's GPUs are reliant on high bandwidth memory, which is currently experiencing a price spike due to global memory shortage. GROX architecture, on the other hand, utilizes less costly S-RAM and allows Nvidia to offer a completely different product. Effectively, the more mature that AI gets, the more that different workloads have different types of needs, that can be optimized by different types of chips. The architecture of GROX chips is extremely

relevant for things like low-latency applications, i.e., the sort of general-purpose agent interactions we were talking about before with the Manus acquisition, where people don't want to be sitting around waiting for a response, they want to be interacting, as though the agent is actually an agent working on their behalf, as well as potentially being relevant for other types of applied AI contexts like edge devices running smaller models, and eventually lower power chips to put inside robots and embodied AI. There also is potentially a virtuous cycle. Here's GROC CEO, Jonathan Ross. InVITA will sell every single GPU they make for training. Right now, about 40% of their

you know, market is inference. If we were to deploy a lot of much lower cost inference chips, what you would see is that same number of GPUs would be sold, but the demand for training would increase because the more inference you have, the more training you need and vice versa. You could almost say we're one of the best things that ever happened to in video because they can make every single GPU that they were going to make and they can sell it for training, high margin, right, gets amortized across the deployment. And, you know, we'll take the low margin, high volume inference business off their hands and they won't have to sell either market. As SMGIT sums up, when Grok floods the market with cheap inference chips, everyone's going to need way more training to feed all that inference capacity.

It's a perfect cycle. More inference equals more training needed. Anyways, guys, for my money, those are the two biggest stories from the holiday period. But of course, we are just at the beginning of the year, and I expect a lot more to happen in very short order. For now, that is going to do it for this first episode of the AI Daily Brief of 2026. Appreciate you listening or watching as always. And until next time, peace. Thank you.