AI darlings race to IPO - Sync #556

Plus: Project Genie; the Adolescence of Technology; SpaceX, Tesla, and xAI in talks to merge; Nvidia's latest investments; Helix 02; Qwen3-Max-Thinking; Kimi K2.5; and more!

Hello and welcome to Sync #556!

This week, we take a closer look at emerging reports that OpenAI and Anthropic are preparing for IPOs, potentially as soon as this year.

Elsewhere in AI, Google is opening access to Project Genie, while OpenAI launches Prism — a free, AI-powered workspace designed for scientists. Meanwhile, Nvidia invests in CoreWeave and plans to invest further in OpenAI, even as reports emerge that its $100 billion partnership with OpenAI is on ice.

Over in robotics, Figure unveils Helix 02, its new model for controlling humanoid robots; iRobot emerges from Chapter 11 as a restructured Picea US subsidiary; and Nvidia, Mercedes-Benz, and Uber make progress on their robotaxi plans.

Beyond that, this week’s issue of Sync also includes new models from Qwen and Kimi, Dario Amodei’s The Adolescence of Technology essay, reports of a possible SpaceX, Tesla, and xAI merger, a hard look at humanoid robots and AI hype—and more!

Enjoy!

AI darlings race to IPO

For much of the past decade, Silicon Valley’s most valuable technology companies delayed public listings for as long as possible. Deep private capital markets allowed startups to scale without the scrutiny or short-term pressures of public investors. That, however, is now changing.

OpenAI and Anthropic are both laying the groundwork for initial public offerings. OpenAI is targeting the fourth quarter of this year, according to the Wall Street Journal, whilst Anthropic has engaged Wilson Sonsini Goodrich & Rosati, the law firm behind the Google and LinkedIn IPOs, to prepare for a potential listing as early as 2026.

At stake is not simply a race to market, but a test of whether the AI business model can withstand public scrutiny.

The economics of frontier AI

Unlike earlier generations of software startups, leading AI companies face exceptional capital demands. Training and running advanced models requires specialised chips, long-term cloud contracts, and dedicated data centre infrastructure. OpenAI plans to spend $115 billion by 2029. Anthropic recently announced a $50 billion infrastructure build-out with data centres in Texas and New York. Both companies are losing billions annually, with Anthropic expecting to break even in 2028—two years earlier than OpenAI.

Private funding has so far absorbed these costs, but each successive round has grown larger and more concentrated among a small group of strategic investors. OpenAI is now seeking more than $100 billion at a valuation of $830 billion. Meanwhile, Anthropic is negotiating a round that could value it above $300 billion. At this scale, private capital begins to strain.

Revenue, meanwhile, is growing fast. OpenAI hit $13 billion last year and expects to triple that in 2026. Anthropic reached an annual run rate of $8–10 billion by late 2025 and is projecting revenues approaching $26 billion this year. Yet unlike traditional software, AI margins remain tightly linked to compute costs. Revenue growth does not automatically translate into operating leverage—and unlike Facebook and Google at the time of their IPOs, today’s AI companies are losing substantial sums and must continually raise more.

Public markets may be the only option to raise capital at the scale these companies require.

Preparing for the public markets

OpenAI has taken the most visible steps toward a listing. In 2025, the company completed its transition from a nonprofit to a for-profit entity—a necessary prerequisite for any offering. It has since expanded its finance leadership and held informal discussions with investment banks. Sam Altman has acknowledged that public markets may be unavoidable, whilst expressing little enthusiasm for the role.

Anthropic has pursued a quieter but similar path, making a string of finance hires, including Krishna Rao, a former Airbnb executive who played a key role in that company’s 2020 IPO. Executives stress that no decision has been taken on timing. Yet OpenAI's leadership has privately expressed concerns that Anthropic could beat them to market. Whichever company lists first stands to capture investors eager for exposure to the generative AI wave.

What would listing mean

If OpenAI or Anthropic proceed, the consequences would extend beyond their own finances. Public companies must disclose information that remains opaque today: infrastructure obligations, cloud contracts, and the relationship between performance improvements and spending. This transparency would offer policymakers, competitors, and customers a clearer picture of how AI systems are financed and priced.

Just as earlier IPOs brought social media and cloud computing into the mainstream, AI listings could mark the sector’s transition from experimental technology to permanent infrastructure.

Whether 2026 becomes the year of the mega IPO depends on market conditions, company readiness, and investor appetite for loss-making businesses with extraordinary ambitions. But the preparations suggest that the era of AI as a purely private experiment may be coming to an end.

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

---

🧠 Artificial Intelligence

China gives nod to ByteDance, Alibaba and Tencent to buy Nvidia’s H200 chips
China has approved ByteDance, Alibaba and Tencent to buy more than 400,000 of Nvidia’s H200 AI chips. As Reuters reports, the approvals come with conditions that are still unclear, and some companies say they may be too strict. The decision reflects Beijing’s aim to meet strong demand for advanced AI chips while continuing to encourage the use of Chinese-made alternatives.

Project Genie

Google DeepMind is opening access to Project Genie, an experimental AI tool that generates short, interactive game-like worlds from text prompts or images. It is powered by DeepMind’s Genie 3 world model, the Nano Banana Pro image-generation model, and Gemini. The system can maintain limited world consistency and interaction, but it remains a research prototype with notable constraints, including a 60-second session limit, inconsistent realism, basic physics, and strict content guardrails. Project Genie is currently available only to Google AI Ultra subscribers in the United States, and is intended primarily as a testing ground for DeepMind’s world-model research rather than a finished product.

The Adolescence of Technology
Dario Amodei's latest essay caused some stir in the AI community by suggesting that AI is going through an unstable “adolescent” stage, where its abilities are growing faster than society’s ability to control and guide it. The essay’s key themes are responsibility, caution, and urgency: he warns against both panic and ignoring the risks, pointing out that powerful AI could be misused, concentrate power, or disrupt society, while also offering major benefits. The overall message is that governments, companies, and the public must act quickly to build better rules, norms, and oversight so AI can develop safely before the consequences become too large to manage.

Introducing Prism
OpenAI launches Prism, a free AI-powered workspace designed to help scientists write and collaborate more easily. Built around GPT-5.2, Prism combines drafting, LaTeX editing, research support, and real-time teamwork in one online space. It is available now to anyone with a ChatGPT personal account and aims to make everyday research work simpler and more efficient.

Inside OpenAI’s in-house data agent
OpenAI’s in-house data agent is an internal AI tool that helps employees explore and analyse very large and complex datasets using everyday language. It is built using OpenAI’s core models and tools, and it reduces the time needed to find the right data, run queries, and produce useful insights. The agent uses strong contextual understanding, remembers previous questions, and follows existing access permissions, allowing it to give accurate, clear, and secure answers while working more like a helpful teammate than a traditional data tool.

Microsoft announces powerful new chip for AI inference
Microsoft has introduced the Maia 200, a new in-house AI chip designed to improve the speed, efficiency and cost of running AI models. Building on the Maia 100, the chip delivers significantly higher inference performance and aims to reduce power use while supporting increasingly large models. The chip is part of a wider trend of big tech companies creating their own hardware to rely less on Nvidia. Microsoft says Maia is already being used for its AI tools, including Copilot, with access now opening to developers and researchers.

Nvidia, Others in Talks for OpenAI Funding, Information Says
According to a report from The Information, OpenAI is in early talks to raise up to $100 billion at a valuation of about $750-830 billion. Nvidia is considering an investment of up to $30 billion. Microsoft may invest less than $10 billion, while Amazon could commit more than $20 billion. The planned new investments into OpenAI are in addition to the $30 billion that SoftBank is planning to commit, according to the report. Meanwhile, the Wall Street Journal reports that Amazon might invest up to $50 billion in OpenAI.

The $100 Billion Megadeal Between OpenAI and Nvidia Is on Ice
The Wall Street Journal reports that Nvidia’s plan to invest up to $100 billion in OpenAI and provide vast amounts of computing power has stalled after internal doubts emerged and negotiations failed to progress. The original deal was non-binding, and both companies are now reconsidering their partnership, possibly replacing it with a smaller investment. The delay is a setback for OpenAI as it tries to secure enough computing power for future growth, and it has also increased concerns about the cost and risk of its large-scale deals. Jensen Huang responded to the report by calling it “nonsense.”

Nvidia invests $2B to help debt-ridden CoreWeave add 5GW of AI compute
Nvidia has invested $2 billion in CoreWeave to accelerate the expansion of its AI data centre capacity to more than 5 gigawatts by 2030, deepening an existing partnership between the two companies. The deal shows strong support for CoreWeave, even as the company faces scrutiny over its large debt. For Nvidia, it is part of a wider effort to keep driving growth in the fast-moving AI industry.

Anthropic reportedly upped its latest raise to $20B
Anthropic is seeking to raise $20 billion in venture capital—double its original target—amid strong investor demand, according to the Financial Times. The funding round, expected to close soon, would value the company at around $350 billion.

Elon Musk’s SpaceX, Tesla, and xAI in talks to merge, according to reports
Rumours have emerged that three of Musk’s companies—SpaceX, xAI, and Tesla—are in early discussions over a possible merger, with reports suggesting one or more could ultimately be folded into SpaceX. Several scenarios are being considered, including a SpaceX–Tesla or SpaceX–xAI combination, potentially ahead of a planned SpaceX IPO. While no public statements have been made, recent corporate filings and significant cross-investments point to Musk’s broader aim of consolidating resources across his businesses.

Tesla to invest $2B in Elon Musk’s xAI
Tesla has confirmed it invested $2 billion in xAI, even though shareholders had earlier voted against the move. The company says the investment supports its plan to use AI in real-world products such as self-driving vehicles and humanoid robots, and will deepen collaboration between the two firms while helping to speed up its AI development.

Apple acquires Israeli audio AI startup Q.ai
Apple has acquired Q.ai, an Israeli startup working on AI technology for audio, for reportedly $1.6 billion. Apple did not say how it will use Q.ai’s technology but said the startup has worked on new applications of machine learning to help devices understand whispered speech and to enhance audio in challenging environments. Q.ai last year filed a patent application to use “facial skin micromovements” to detect words mouthed or spoken, identify a person and assess their emotions, heart rate, respiration rate and other indicators.

Zuckerberg teases agentic commerce tools and major AI rollout in 2026
Mark Zuckerberg says new Meta’s AI tools and models will start reaching users in the coming months after the company rebuilt its AI programme. He said Meta will focus on AI-powered shopping tools that help people find products based on their interests and activity. Meta believes its access to personal data will give it an advantage over competitors, and the company is increasing spending on AI infrastructure, even as some investors remain unsure how this investment will lead to profits.

Yahoo is adding generative AI to its search engine
Yahoo joins the AI-powered search club by launching Yahoo Scout. Currently in beta and built on Anthropic’s Claude, designed to synthesise information from the web and Yahoo’s own content with clear source links. Alongside this, Yahoo is rolling out an AI intelligence platform across its products, adding features such as email summaries, news takeaways, sports analysis, and personalised shopping and finance insights.

Yann LeCun’s new venture is a contrarian bet against large language models
Yann LeCun discusses in this interview his vision for the future of AI, not based on large language models but on world models. He also explains the importance of open models, why his new company, Advanced Machine Intelligence (AMI), is headquartered in Paris, what academia should be focusing its research beyond LLMs, and teases a potential announcement sometime in February.

Meet ‘Amelia’: the AI-generated British schoolgirl who is a far-right social media star
The British far-right has a new social media star. Amelia is a fictional schoolgirl who first appeared in a UK government-funded game designed to prevent extremism, but she has since been taken over online and turned into a far-right meme. The article explains how extremists have used AI tools to remake and share her image to spread racist messages, reach wider audiences and even make money through cryptocurrency, showing how quickly online projects can be misused on social media.

Researchers Are Using A.I. to Decode the Human Genome
AlphaGenome is a new AI system created by Google DeepMind to help scientists understand how changes in DNA affect how genes behave. It can predict whether mutations might wrongly turn genes on or off, which is important for studying diseases like cancer. Researchers say it is a powerful and advanced research tool, but it is not as groundbreaking as AlphaFold. Its predictions have limits and cannot yet be used in medical care, so scientists still need to confirm its results with laboratory experiments.

Legal AI giant Harvey acquires Hexus as competition heats up in legal tech
Harvey, a fast-growing legal AI company, has bought Hexus, a small startup that builds tools for product demos and guides. The Hexus team is joining Harvey, and its founder, Sakshi Pratap, will help lead engineering work to improve products for in-house legal teams. The move comes as Harvey continues to grow quickly, with a high company valuation, strong investor backing, and a large number of law firm clients around the world.

Kimi K2.5: Visual Agentic Intelligence
Kimi has released K2.5, which it describes as the most powerful open-source model to date—and it delivers on that promise. In terms of performance, Artificial Analysis placed Kimi K2.5 fourth overall, just behind GPT-5.2, Claude Opus 4.5, and Gemini 3 Pro, while ranking it as the highest-performing open model. Moreover, it is more efficient and cheaper to run than the top proprietary models. The main innovation in K2.5 is the introduction of an agent swarm, with up to 100 sub-agents executing parallel workflows across as many as 1,500 tool calls. If you want to learn more about Kimi K2.5, Caleb Writes Code has an excellent video on the subject.

Pushing Qwen3-Max-Thinking Beyond its Limits
Alibaba’s Qwen team has released Qwen3-Max-Thinking, its latest flagship reasoning model. The Chinese lab says that, according to its internal benchmarks, Qwen3-Max-Thinking demonstrates performance comparable to leading models such as GPT-5.2 Thinking, Claude Opus 4.5, and Gemini 3 Pro. However, independent benchmarks from Artificial Analysis place the new model roughly on the level of Grok 4, MiniMax-M2.1, DeepSeek V3.2, GLM-4.7, and Kimi K2.5. Qwen3-Max-Thinking is a proprietary model, as Alibaba has not released the weights. It is currently available through Qwen Chat and via the first-party API on Alibaba Cloud.

Mistral Vibe 2.0
Mistral is releasing Mistral Vibe 2.0, a major upgrade to its terminal-native coding agent, powered by the Devstral 2 model family. The new version brings custom subagents, multi-choice clarifications, slash-command skills, unified agent modes, and automatic updates.

Open Coding Agents: Fast, accessible coding agents that adapt to any repo
Allen Institute for AI (Ai2) has launched Open Coding Agents, beginning with its SERA (Soft-verified Efficient Repository Agents) models—a family of open-source coding agents that can be inexpensively fine-tuned to specific codebases for tasks such as code generation, review, debugging and maintenance. The approach lowers the barrier for small teams by offering fully open models, training data and code, along with straightforward integration with tools like Claude Code.

🤖 Robotics

Introducing Helix 02

Figure reveals Helix 02, a full-body humanoid control system that, according to the company, uses a unified neural architecture combining data from vision, touch, and proprioception sensors to control every joint and actuator in the robot. This is intended to enable human-like whole-body motion and long-horizon autonomy. The company says that with Helix 02, the robots can execute a four-minute, end-to-end autonomous task (like loading and unloading a dishwasher) with no resets, no teleoperation, and no human intervention, as well as reaching new levels of dexterity.

iRobot emerges from Chapter 11 as restructured Picea U.S. subsidiary
iRobot has emerged from bankruptcy after being acquired by Shenzhen Picea Robotics, giving the company a more stable financial base to focus on future products. The company will continue supporting its robot vacuums and operating from the United States as a privately owned business. To ease concerns about data privacy, iRobot has set up a separate US-based unit to protect customer data under the new ownership.

Nvidia, Mercedes-Benz Move Forward With Planned Robotaxis
Nvidia, Mercedes-Benz and Uber are advancing plans to launch robotaxi services in major cities, using Mercedes’ S-Class cars and Nvidia’s self-driving technology. The timeframe for the start of the service is still not publicly known.

▶️ Spot Cam 2 | Boston Dynamics (0:53)

Boston Dynamics launched Spot Cam 2, a new camera module that can be attached to Spot’s back. This second generation Spot Cam features a 4K pan-tilt-zoom camera with 25x optical zoom, an integrated radiometric thermal camera, and a 360 x 130º spherical camera. It also makes figuring out where Spot’s “face” is even harder.

Gartner predicts fewer than 20 companies will deploy humanoids at scale by 2028
Gartner says that although humanoid robots attract a lot of attention and investment, very few will be used in real workplaces before 2028. The technology is still expensive, limited, and hard to fit into existing operations, especially in busy supply chain and manufacturing settings. Instead, Gartner expects simpler, task-focused “polyfunctional” robots to be more practical and cost-effective for most companies.

▶️ The Hard Truth About Humanoid Robots and AI Hype (46:43)

Eric Danziger, the CEO of Invisible AI, shares the tougher side of robotics that shiny demos fail to show. In this conversation, he explains why demos are easy while deployment is hard—and why computer vision, not humanoid bodies, may be the real bottleneck in physical AI. He also discusses why manufacturing automation moves more slowly than software, how hype cycles distort expectations, and what “real robotics” actually means on the factory floor.

▶️ Oli Demonstrates the World’s First Scalable Autonomous Deployment (1:05)

LimX Dynamics has showcased what it claims is the world’s first practical autonomous deployment of humanoid robots, with 18 full-size Oli units independently emerging from shipping containers, walking, coordinating their movements, and performing a synchronised routine without human input. LimX highlights advances in multi-robot coordination powered by the company’s new COSA operating system, which integrates perception, reasoning, memory, and movement in a single framework.

🧬 Biotechnology

First AI-Generated Genomes: The Synthetic Biology Breakthrough of 2026
Scientists have successfully created the first genomes designed entirely by artificial intelligence. The AI was trained on huge amounts of DNA and used this knowledge to design new genetic material that worked inside living cells. Some of these AI-made viruses performed better than natural ones, especially in fighting drug-resistant bacteria. This breakthrough could speed up new medical treatments and biotechnology research, while also raising important safety and ethical concerns.

💡Tangents

Amazon to Shut Down All Amazon Go and Amazon Fresh Stores
Amazon is closing all Amazon Go and Amazon Fresh stores after deciding they did not offer a shopping experience that could grow successfully. The move affects 72 stores, some of which will become Whole Foods locations. The company will now focus on online same-day delivery and growing its Whole Foods business, including opening more than 100 new Whole Foods stores in the coming years.

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!"