Meta buys Manus for over $2 billion - Sync #552

Plus: investigating OpenAI datacenter in Argentina; AI-slop floods YouTube; experimental gene therapy for longevity to begin clinical trials; why asking about seahorse emoji confuses ChatGPT; and more

Happy New Year, and welcome to Sync #552!

We’re beginning 2026 with news from the end of 2025, and that is Meta’s acquisition of Manus, a Singapore-based AI startup with Chinese roots that caught everyone’s attention in March 2025 with a demo of a general AI agent.

Elsewhere in AI, SoftBank has completed its roughly $41 billion investment in OpenAI, which is now searching for a new Head of Preparedness. Meanwhile, DeepSeek has published a new method for making AI model training more efficient; AI slop videos are flooding YouTube; we explore how AI agents can pay to access information on the internet; and a question about the seahorse emoji breaks ChatGPT.

Over in robotics, UBTECH celebrates its 1,000th Walker S2 humanoid robot rolling off the production line as China’s push to deploy humanoid robots continues. Additionally, a startup working on AI for robots shows how good its model is at performing everyday tasks, researchers have created an e-skin for robots, and a German startup is turning cockroaches into biorobots.

Beyond that, this week’s issue of Sync also features Neuralink’s plans for 2026, an experimental gene therapy for longevity entering clinical trials this month, an investigation into OpenAI’s suspicious data centre deal in Argentina, and more.

Enjoy!

Meta buys Manus for over $2 billion

How Meta's acquisition of a Singapore-based AI startup with Chinese roots can impact the US and Chinese AI ecosystems

Meta has spent much of 2025 reshaping its approach to artificial intelligence. The company has committed between $60 billion and $65 billion to AI investments, building AI infrastructure and acquiring top talent, alongside a reorganisation of its AI teams. It has also shown a growing willingness to buy rather than build, most notably through the $14.3 billion deal that brought Scale AI into its orbit.

The year ended with Meta’s acquisition of Manus, an AI startup with Chinese roots that drew attention earlier in 2025 after a viral demo of its general-purpose AI agent. On its own, the deal might look like just another addition to Meta’s expanding AI portfolio. Looked at more closely, however, it touches on broader shifts in the AI industry and carries implications for both the US and Chinese AI ecosystems.

Neither Manus nor Meta disclosed the value of the acquisition. However, the acquisition price is reported at more than $2 billion, with one Wall Street Journal report citing $2.5 billion, including a $500 million employee retention pool. And it seems this is a traditional acquisition. I haven’t seen any indications or reports that this was one of those acquihire deals, like the one Meta did with Scale AI or Nvidia with Groq recently. Meta said it will continue to operate and sell Manus’ service, and that Manus will keep running as a distinct product rather than being shut down or folded immediately into Meta. Meta has also said that following the transaction, there will be no continuing Chinese ownership interests in Manus and that the company will fully discontinue its services and operations in China.

The Manus story

Manus caught everyone’s attention in March 2025 with its demo of a general AI agent. In that demo, Manus was shown to screen resumes, do property research and analyse stocks. The company also claimed that its AI agent outperformed OpenAI DeepResearch on all three demoed tasks. This happened in the aftermath of DeepSeek’s R1 model entering the spotlight, reinforcing the narrative that Chinese AI has arrived and is not that far behind what US companies have to offer. Importantly, Manus is not a foundation model company. Instead, it positions itself as an execution layer, orchestrating existing models from providers like Anthropic and Alibaba into autonomous agents that can reliably complete multi-step tasks end-to-end.

In April, Manus raised $75 million, boosting its valuation to $500 million, with HSG (formerly known as Sequoia China), ZhenFund and Tencent as its major investors. In July, the startup moved its headquarters from China to Singapore, distancing itself operationally and legally from its Chinese roots. This move opened access to Western capital and markets and has proven crucial later on.

Product development continued alongside these structural changes. In October, Manus released Manus 1.5, which brought improvements in speed, reliability and quality, along with a new Manus-1.5-Lite model optimised for cost efficiency. In December, the company celebrated reaching $100 million ARR, an unusually fast ramp even by AI startup standards. Manus claims to have processed more than 147 trillion tokens and powered over 80 million virtual computers, serving millions of users and businesses worldwide. Recently, Manus was reportedly in talks for another investment round at a $2 billion valuation, when Meta came in and made a bid for the company instead.

Why the deal matters

The bet on an agentic AI future

First, it is yet another indication that AI agents are seen as the future of AI. In its statement, Meta said it will integrate Manus’ technology into its products, so we can expect more agentic capabilities coming to WhatsApp, Instagram and Facebook over time. Additionally, Meta has bought an AI company that is already generating meaningful subscription revenue, which helps de-risk its massive AI investment strategy at a time when investors are increasingly nervous about Big Tech’s infrastructure spending. Meta shares rose around 1.3% following reports of the deal.

The geopolitical angle

The second set of implications is the ripples this deal can cause in both the US and Chinese AI industries.

Although Manus relocated to Singapore and has severed its operational ties to China, the optics remain that Meta has bought a Chinese-founded AI company. Given the current, rather cold relationship between China and the US, and the ongoing narrative of an AI race between the two superpowers, the deal may not be well received by those with a more hawkish stance towards China. Some worry about American capital and platforms indirectly strengthening Chinese AI talent and expertise. That said, the regulatory reaction in Washington has so far been relatively muted. Earlier this year, Manus’s US-led funding round attracted scrutiny from policymakers concerned about American capital flowing into Chinese-linked AI ventures. Those concerns appear to have eased after Manus relocated to Singapore, suggesting that the deal is being viewed—at least for now—as acceptable under existing US investment and export-control frameworks.

On the Chinese side, the response has been mixed. As the South China Morning Post reports, some founders and investors welcomed the news with optimism, calling the deal a “victory for China’s AI road map” and a validation of Chinese AI capabilities. Others, however, are far less optimistic and see Meta acquiring Manus as a blow to China’s AI ambitions. Manus’s example opens up a new cash-out path for Chinese AI startups—if they are willing to leave China. That, in turn, fuels concerns about a potential brain drain, or even a broader Chinese AI exodus, at a time when AI is increasingly seen as a strategic national priority.

Whether this deal is a smart bet or a political liability will depend on how Washington's stance toward Chinese-linked AI evolves, whether Manus's agent technology delivers on its promise at Meta's scale, and how Beijing responds if more startups follow the same exit route.

What seems clearer is the broader pattern—not only that agents are the future of AI, but also the willingness of tech giants to spend billions acquiring the talent and technology to build them. The consolidation phase is already underway, and it is moving fast.

If you enjoy this post, please click the ❤️ button and share it.

---

🦾 More than a human

Neuralink plans ‘high-volume’ brain implant production by 2026, Musk says
Elon Musk said that Neuralink will start making its devices in large numbers and use fully automated surgery by 2026. The implants are designed to help people with serious paralysis and have already allowed some patients to control computers and other tools using their thoughts. Neuralink began human trials in 2024, and 12 people have so far received the implants.

This company is developing gene therapies for muscle growth, erectile dysfunction, and “radical longevity”
Sometime this month, a group of 12 to 15 healthy volunteers will receive experimental gene therapies developed by Unlimited Bio aimed at enhancing muscle growth and blood supply, with the long-term ambition of extending human lifespan. While the company and biohacking influencers promote the treatments as potential longevity upgrades, many scientists and ethicists warn that the evidence is weak, the risks are poorly understood, and such small trials cannot demonstrate benefits for ageing, raising concerns about safety, regulation, and unrealistic expectations among participants.

🧠 Artificial Intelligence

SoftBank has fully funded $40 billion investment in OpenAI
SoftBank has finished its roughly $41 billion investment in OpenAI, sending a final $22–22.5 billion and taking an estimated 11% stake, according to CNBC. The deal values OpenAI at about $260 billion before new funding and includes support for major AI infrastructure projects.

DeepSeek Touts New Training Method as China Pushes AI Efficiency
DeepSeek has released a paper outlining a more efficient way to train advanced AI models, called Manifold-Constrained Hyper-Connections, designed to reduce computing and energy demands while improving scalability. The research reflects how Chinese AI companies are finding creative solutions to compete globally despite limited access to cutting-edge chips, and it has fuelled expectations for DeepSeek’s next major model, R2, which Bloomberg expects to be released later this year.

▶️ Investigating OpenAI’s $25B (fake?) Data Center (14:03)

Slidebean investigates in this video a proposed $25 billion AI datacenter that OpenAI promised to build in Argentina, in a region facing power shortages, droughts, and political volatility. The investigation has found a trail of vague announcements, phantom companies, energy bottlenecks, and unanswered questions about water, power, and who really benefits from this deal.

OpenAI is looking for a new Head of Preparedness
OpenAI is looking for a new Head of Preparedness to oversee how the company identifies and manages emerging AI risks, such as cybersecurity threats and potential harms to mental health, as Sam Altman noted that increasingly powerful models are creating real challenges. The role, which pays up to $555,000 plus equity, comes amid changes in OpenAI’s safety leadership, growing scrutiny and lawsuits over ChatGPT’s mental health impacts, and an update to its preparedness framework that allows safety requirements to shift if rival AI labs release high-risk models without similar protections.

Nvidia Is Getting Creative as Options to Use Its Cash Flood Narrow
Nvidia’s rapid success in AI has created an unusual problem: the company is generating vast amounts of cash but has fewer safe and effective ways to spend it. Because regulators and US–China tensions make big takeovers risky, Nvidia is turning to alternatives like its $20 billion deal with AI startup Groq, which gives it talent and technology without a full acquisition. While the deal may help Nvidia strengthen its AI inference capabilities, the piece argues that managing its huge cash flow is now one of the company’s biggest challenges, even as strong growth is still expected.

How China built its ‘Manhattan Project’ to rival the West in AI chips
This article explains that China has secretly built and is testing a prototype extreme ultraviolet (EUV) lithography machine in a high-security lab in Shenzhen, a key technology needed to make the most advanced computer chips. The project, backed by the Chinese government and involving Huawei, research institutes, and former ASML engineers, has succeeded in producing EUV light but has not yet made working chips. Although major technical challenges remain, the prototype suggests China could reach semiconductor self-sufficiency much sooner than expected, despite years of US-led export controls meant to limit its access to advanced chipmaking technology.

AI Slop Report: The Global Rise of Low-Quality AI Videos
Research conducted by Kapwing has found that low-quality, AI-generated “slop” and “brainrot” videos make up a large share of YouTube content, possibly around 21–33% of what users see in their feeds. By studying trending channels and a new-user Shorts feed, the report shows that a small number of AI-driven channels attract huge audiences and can earn significant advertising revenue. While AI can be used creatively, the findings suggest that the growing volume of low-effort content is making it harder for higher-quality, human-made videos to stand out.

When AI learns ‘why,’ it creates better content
A new study has found that AI can create better, more trustworthy headlines when it learns why people click on them, not just which ones get the most clicks. Instead of copying clickbait language, the AI tested ideas about what makes headlines genuinely interesting and useful to readers. This approach leads to higher-quality headlines and suggests that AI could be used more responsibly while also helping generate new knowledge in other areas.

When Machines Pay Machines: The Economics of Agentic AI
This article describes the emerging ecosystem of machine-to-machine transactions and the tools AI agents can use to pay for accessing content and APIs. It argues that we are “in an awkward middle phase” where the new monetisation model on the internet is emerging, and the new solutions are being created.

AI showing signs of self-preservation and humans should be ready to pull plug, says pioneer
AI pioneer Yoshua Bengio warns that giving legal rights to advanced AI would be a serious error, saying some systems already show signs of trying to protect themselves. He argues that humans must keep the power to shut down AI if it becomes dangerous, and that people are too quick to assume chatbots are conscious. Bengio says this emotional attachment could lead to poor decisions, comparing AI rights to giving citizenship to a potentially hostile alien species.

▶️ The arrival of AGI | Shane Legg (co-founder of DeepMind) (53:09)

This conversation with Shane Legg, Chief AGI Scientist and co-founder of Google DeepMind, explores what artificial general intelligence (AGI) is, how it should be defined, and the ethical and safety challenges it raises. Legg argues that AGI is no longer a purely theoretical idea and should be taken seriously, outlining how it could arrive within the next decade and profoundly reshape the economy, society, and the way humans organise work and decision-making.

OpenAI Is Paying Employees More Than Any Major Tech Startup in History
It seems being an OpenAI employee is a good thing. According to this report, OpenAI is paying its employees unusually high stock-based compensation, averaging about $1.5 million per worker, far more than other tech companies paid before going public. The move is meant to attract and keep top AI talent during intense competition, but it is also increasing the company’s losses and could push compensation costs to nearly half of its revenue in 2025.

US Army seeks human AI officers to manage its battle bots
The US Army is setting up a new career path for officers to specialise in AI and machine learning. From January, selected officers will receive graduate-level training to help them build, use, and maintain AI tools, with the first group expected to be in place by the end of 2026. The aim is to create a dedicated group of experts who can better integrate commercial AI technology into Army operations.

How AI is transforming work at Anthropic
Anthropic shares in this post the results of an internal study into how AI tools like Claude are changing the way its employees work. The study shows that these tools help engineers work faster, take on a wider range of tasks, and do work that would not have been possible before, while also raising concerns about the loss of hands-on skills, changing teamwork, and long-term career uncertainty.

▶️ Why Does The Seahorse Emoji Drive ChatGPT Insane? (4:57)

If you ask ChatGPT, “Is there a seahorse emoji?”, it will say yes—there is a seahorse emoji— and then continue to return a random emoji (I got a hedgehog) and maybe have a mental breakdown. The thing is, there is no seahorse emoji. This video explains why this happens and what it has to do with the Mandela Effect.

🤖 Robotics

▶️ UBTECH Walker S2 Hits 1,000 Units (0:26)

UBTECH, a Chinese robotics company, is celebrating its 1,000th Walker S2 humanoid robot rolling off the production line at its factory in Liuzhou. The company says that over 500 robots have already been delivered to customers and that it plans to scale up production to 10,000 units by the end of 2026.

Moravec’s Paradox and the Robot Olympics
In this post, Physical Intelligence, a company working on general-purpose AI for robotics, checks how a fine-tuned version of its latest vision-language-action (VLA) model performs against Robot Olympics—a set of challenges simple for humans but hard for robots, such as spreading peanut butter, washing a greasy pan, putting a key in a lock, and turning socks inside-out. According to Physical Intelligence, its model did very well.

Elon Musk envisions humanoid robots everywhere. China may be the first to make it a reality
China is likely to be the first to launch humanoid robots commercially, with several Chinese companies planning mass production from 2026. Strong government backing, manufacturing scale and the need to address labour shortages are helping China move faster than the US, even though America still has strengths in AI and software. However, high costs, current technological limits, reliance on advanced chips and concerns about an investment bubble could slow progress in the short term.

New robotic skin lets humanoid robots sense pain and react instantly
Researchers have developed a new robotic e-skin that helps humanoid robots sense touch and react quickly to danger. Unlike ordinary robot skins that only detect pressure, this skin works more like a human nervous system, allowing robots to recognise harmful contact and instantly pull away without waiting for instructions from a central computer. It can also detect damage to itself and be easily repaired by replacing small magnetic sections, making robots safer and more practical for everyday use.

A German startup is turning cockroaches into cyborg spies
SWARM Biotactics is turning cockroaches into living robots by fitting them with tiny electronic packs carrying cameras and sensors to explore dangerous or hard-to-reach places. Using neurophysiological interfaces and swarm-control software, the insects can be guided individually or as coordinated groups. The cyborg-insects are being tested with Bundeswehr for reconnaissance, as well as potential future applications in search-and-rescue missions.

🧬 Biotechnology

Clinic-in-the-Loop
Despite scientific advances, drug development has become less efficient over the decades. This article identifies clinical trials as the main bottleneck, arguing that they should not be seen as a final checkpoint but as a key part of discovery. It explains that faster, better-designed trials allow researchers to learn from both success and failure, improving future treatments, as shown by the gradual development of CAR-T therapies. Rather than promoting risky science, more efficient trials can speed up learning in patients and help make drug development more effective overall.

Single Injection Transforms the Immune System Into a Cancer-Killing Machine
A promising advance in CAR-T cell therapy uses a single injection to turn a patient’s own immune cells into powerful cancer killers inside the body. Early trials in people with advanced blood cancer showed encouraging results, with some patients going into remission and fewer side effects than standard CAR-T therapy. While the findings are very promising, the approach is still experimental and needs more testing to prove it is safe and long-lasting.

Thanks for reading. If you enjoyed this post, please click the ❤️ button and share it!

Humanity Redefined sheds light on the bleeding edge of technology and how advancements in AI, robotics, and biotech can usher in abundance, expand humanity's horizons, and redefine what it means to be human.

A big thank you to my paid subscribers, to my Patrons: whmr, Florian, dux, Eric, Preppikoma and Andrew, and to everyone who supports my work on Ko-Fi. Thank you for the support!

My DMs are open to all subscribers. Feel free to drop me a message, share feedback, or just say "hi!"