Google is winning the AI race - Sync #516
Plus: humanoid robots ran a half-marathon in China; a fresh wave of drama at OpenAI; lab-grown human teeth and lab-grown chocolate; the impact of tariffs on the computer industry; and more!
Hello and welcome to Sync #516!
For this week’s main story, we will examine how Google quietly gained the upper hand in the AI race and what the biggest threat to its recently gained leadership position is.
Apart from that, we have a packed issue of Sync this week, which includes humanoid robots running a half-marathon in China and a fresh wave of drama at OpenAI. Meanwhile, DeepMind workers plan to unionise and Elon Musk is in early talks to raise about $20 billion in funding for XAI Holdings.
Over in robotics, Waymo reports 250,000 paid robotaxi trips per week in the US, Tesla begins ride-hailing tests in Austin and the Bay Area, and Elon Musk reveals that Tesla's Optimus humanoid robots are affected by China's export curbs on rare earths.
Additionally, in this week’s issue of Sync, we also have lab-grown human teeth and lab-grown chocolate, as well as a report on longevity clinics. We will finish with an article exploring why the chip industry is struggling to attract the next generation, and how the recent US tariffs are impacting the computer industry and how they could make computing much less accessible.
Enjoy!
Google is winning the AI race
Google had all the pieces needed to lead the AI revolution. It had the talent and knowledge—after all, the Transformer model, the foundation of today’s large language models, was created inside Google’s research labs. It had access to nearly the entire internet’s information, across text, video, and audio. And it had access to the required computing power to crunch through massive training runs at scale.
Yet despite these advantages, it wasn’t Google that kicked off the AI explosion. That honour went to OpenAI and its small experiment, ChatGPT, launched in late 2022. ChatGPT captured the world’s imagination and set off the generative AI gold rush.
Google was caught off guard. It scrambled to respond with Bard—a rushed, underwhelming product that only fueled narratives that Google was slow, bureaucratic, and no longer capable of leading the next wave of tech innovation.
Meanwhile, OpenAI and Microsoft raced ahead, integrating AI tools into search engines, office suites, and developer platforms. Google’s rivals grabbed mindshare and early enterprise traction. Public trust wavered, and the sense that Google had missed its moment only deepened.
But then the tech giant woke up.
Two years later, the story looks very different. Google's stumble may have cost it the early headlines, but with its full-stack approach, new leadership focus, and massive distribution power, it’s now better positioned than anyone else to dominate the next phase of AI.
Google enters the Gemini Era
Following the Bard stumble, Google merged its Brain team and DeepMind into a single AI lab named Google DeepMind. Leadership across Google Cloud and Google Research doubled down on execution, not just pure research.
The key moment in Google's pivot came with a laser-focused strategy built around a new family of AI models—Gemini. At Google I/O 2023, Sundar Pichai announced the company was entering the "Gemini Era" and outlined a vision in which every Google app and service—from Search and Gmail to YouTube and Workspace—would eventually be infused with AI capabilities.
At the core of this strategy was the Gemini family of models. While the first version didn’t immediately take the performance crown from OpenAI, Gemini consistently improved with each release, and the gap quickly narrowed.
Gemini 2.5 Pro is arguably Google's crowning achievement to date in AI. Its technical advantages are clear: a one-million-token context window (with plans to double it), far exceeding GPT-4 and Claude’s limits. It can process text, images, audio, video, and code—all within a single, unified model with reasoning capabilities.
Moreover, Gemini 2.5 Pro performs extremely well on benchmarks, often matching or exceeding competitors. But topping benchmarks is only part of the story. Gemini delivers this performance at a significantly lower cost, as perfectly illustrated by benchmark results from Aider.
This combination—performance plus efficiency—gives Google a major competitive advantage, especially in attracting enterprise customers. These customers are looking to process massive amounts of data and deploy large-scale AI agent systems. With Gemini 2.5 Pro, Google can deliver comparable—or better—performance than its competitors, but at a significantly lower cost, making it a far more practical solution for businesses.
In addition, Google offers deep integration across its own tools and apps, further simplifying AI workflows. This tight ecosystem makes introducing AI into the enterprise not only easier but also more cost-effective, compared to other options.
Google's full-stack advantage
Google’s AI advantage doesn’t stop at the model level.
It’s the only major AI player that also owns the infrastructure stack end-to-end. The recently announced Ironwood TPUs—the seventh generation of Google’s custom chips—offer massive performance and energy efficiency gains over conventional GPUs. Additionally, Google’s full control over its hardware and software opens opportunities to optimise software and hardware together, bringing even more gains in cost and performance. This enables Google to deliver “intelligence per dollar” advantages at a scale others can’t easily replicate.
Google’s Vertex AI platform, BigQuery’s AI-native upgrades, and tight Workspace integrations create a seamless ecosystem for enterprises. Instead of stitching together models from one company, cloud from another, and productivity apps from yet another, Google offers it all under one roof.
That vertical integration translates into lower costs, faster innovation, and a better user experience—key factors for winning enterprise deals.
Google's distribution power
If raw app usage were the only metric, OpenAI would still be leading. According to documents revealed in Google’s antitrust trial, Gemini has reached 350 million active monthly users as of March 2025. The same documents estimate that ChatGPT has around 600 million monthly active users, almost twice as many as Gemini.
But that view misses the bigger picture. Google isn’t just relying on Gemini as a standalone app—it's weaving Gemini deeply into its core products.
AI Overviews in Search now reach 1.5 billion users monthly, as revealed in Alphabet’s Q1 2025 results. Android distribution deals ensure Gemini will be accessible to billions of smartphone users by default. Chrome, Gmail, YouTube—all massive apps with millions of users each—are gradually integrating Gemini features. Workspace, which powers daily workflows for millions of workers worldwide, is also getting AI features to make workers more productive.
Google’s enormous reach gives it a distribution advantage that few companies can match. With Gemini embedded across its ecosystem, Google has a rare opportunity: to deliver AI features at a scale almost no other player can achieve.
Google's traction in open models
Google is also gaining momentum in open-source AI, thanks to Gemma—its family of small, open models. While OpenAI remains a closed platform, Google's move into open models positions it to win favour among academics, startups, and smaller enterprises that prefer models they can run and fine-tune themselves.
Gemma models are highly capable for their size and are designed to fit within a single GPU, making them an attractive choice for researchers, developers, and hobbyists alike. Although Gemma itself may not yet be a direct revenue driver for Google, it is helping the company build valuable goodwill within the AI community.
By contrast, OpenAI faces growing criticism for its closed approach—a sore point made even sharper by the irony of "open" being part of its name.
Google is winning the AI race… for now
Google’s journey in AI is a reminder that success isn’t just about having the best research or the earliest start. It’s about execution, infrastructure, distribution, and patience.
Two years ago, it looked like Google had missed the AI moment and lost the leading position. Today, it looks like the tech giant has finally woken up. Now that it’s firing on all cylinders—models, chips, cloud, and reach—Google is setting the pace everyone else has to match.
Google is winning the AI race today. But the question is: how long can it hold that lead?
The biggest threat may not be OpenAI, Anthropic, or some unexpected challenger releasing a model that leaps ahead of Gemini 2.5 Pro. Instead, it could be the US government.
Google is currently defending itself in a major antitrust trial brought by the Department of Justice. If the case doesn’t go in Google's favour, the company could be forced to split up, selling off key assets like Chrome. OpenAI's head of product suggested that OpenAI might be interested in buying Chrome if it comes to market.
A court-ordered break-up could dramatically weaken Google’s distribution power—the very foundation of its AI strategy today. Depending on whether the court orders further break-ups, it might affect other pillars of Google’s AI strategy.
For now, Google is winning. But the next chapter of the AI race may be decided not in the lab, but in the courtroom.
If you enjoy this post, please click the ❤️ button or share it.
Do you like my work? Consider becoming a paying subscriber to support it
For those who prefer to make a one-off donation, you can 'buy me a coffee' via Ko-fi. Every coffee bought is a generous support towards the work put into this newsletter.
Your support, in any form, is deeply appreciated and goes a long way in keeping this newsletter alive and thriving.
🦾 More than a human
Longevity clinics around the world are selling unproven treatments
As longevity and the pursuit of extended healthspan enter the mainstream, a wave of high-end clinics has emerged worldwide, offering treatments aimed at slowing ageing. A recent survey of 82 such clinics reveals a fragmented and inconsistent industry, often blending traditional medical tools with unproven therapies like stem-cell injections and off-label prescriptions. Most clients are Gen Xers seeking both health and aesthetic improvements, yet the field still lacks clear standards and scientific backing. Despite annual costs ranging from $10,000 to $150,000, only a minority of these clinics are turning a profit.
Parkinson’s Patients Say Their Symptoms Eased After Receiving Millions of New Brain Cells
Scientists have made a breakthrough in treating Parkinson’s disease by safely transplanting lab-grown dopamine neurons into patients’ brains, resulting in higher dopamine levels and easing symptoms for up to 18 months without serious side effects. Unlike earlier controversial methods using foetal tissue, the new approach used "off-the-shelf" donor stem cells and offers a more scalable solution. Patients showed improved movement, sleep, and daily functioning. Though larger trials are still needed to confirm long-term safety and effectiveness, experts say the results mark a significant step towards regenerative therapies for Parkinson’s.
Lab-grown teeth might become an alternative to fillings following research breakthrough
Scientists from the UK have created lab-grown teeth grown from a patient’s own cells. Researchers successfully recreated the tooth development process in the lab by developing a special material that allows cells to communicate gradually. This method could one day enable adults to regenerate their own teeth, transforming dental care with stronger, longer-lasting, and biologically compatible alternatives to fillings and implants.
🔮 Future visions
Generative Ghosts: Anticipating Benefits and Risks of AI Afterlives
In this paper, a pair of researchers introduce the concept of "generative ghosts"—AI-powered, evolving digital replicas of deceased individuals. As advances in AI make such technology increasingly feasible, the authors explore the emotional, ethical, and societal impacts of creating digital afterlives, highlighting both potential benefits, such as legacy preservation and serious risks such as mental health harms, identity theft, and cultural disruption. They call for careful design, strong policy frameworks, and further interdisciplinary research to ensure these technologies are used responsibly.
🧠 Artificial Intelligence
Inside OpenAI's Controversial Plan to Abandon its Nonprofit Roots
This article examines OpenAI’s controversial plan to convert into a fully for-profit company and highlights how critics view the move as a betrayal of its original nonprofit mission to ensure AI benefits humanity. It discusses OpenAI’s creation of a new philanthropic commission, which many see as a PR tactic to appease regulators amid lawsuits, government investigations, and growing nonprofit opposition. The piece also details concerns about conflicts of interest, leadership shakeups, and the significant legal and financial challenges OpenAI faces as it pushes forward with the restructuring.
Ex-OpenAI staffers urge states not to approve ChatGPT maker’s restructuring effort
A group of former OpenAI employees, Nobel laureates, and civil society organisations have urged attorneys general in California and Delaware to block OpenAI’s plan to restructure into a for-profit company, warning it would undermine its original nonprofit mission to benefit humanity. In a letter also sent to OpenAI’s board, they argued the shift would eliminate crucial governance safeguards. OpenAI defended the move, claiming it would strengthen its nonprofit role, while critics, including ex-employees and AI experts, fear the change could prioritise profits over safety in developing powerful AI technologies.
Elon Musk’s XAI Holdings Is in Discussions to Raise $20 Billion
Elon Musk’s XAI Holdings, a merger of his AI venture, xAI, and social media platform X (formerly Twitter), is in early talks to raise about $20 billion in funding, which could potentially value the company at over $120 billion. If completed, it would be the second-largest startup funding round ever, only behind OpenAI’s $40 billion financing earlier this year.
Top OpenAI Catastrophic Risk Official Steps Down Abruptly
Joaquin Quiñonero Candela has quietly stepped down from his role as the head of OpenAI’s Preparedness team, tasked with managing catastrophic AI risks. This move follows a wave of high-profile exits from OpenAI's safety and alignment teams, raising concerns about the company’s commitment to AI safety. While OpenAI has introduced new governance structures such as the Safety Advisory Group, critics—including former employees—claim the organisation is increasingly sidelining safety in favour of rapid product development, with recent model releases lacking promised transparency and safety evaluations.
Alphabet, Nvidia invest in OpenAI co-founder Sutskever's SSI, source says
Alphabet and Nvidia have reportedly invested in Safe Superintelligence (SSI), a new AI startup co-founded by former OpenAI scientist Ilya Sutskever, which has quickly reached a $32 billion valuation.
OpenAI seeks to make its upcoming ‘open’ AI model best-in-class
OpenAI is reportedly preparing to release its first open language models since GPT-2 by early summer. Led by VP of Research Aidan Clark, this new model will be optimised for high-end consumer hardware and may feature toggleable reasoning functions, aiming to outperform other open reasoning models with a highly permissive licence.
Google’s DeepMind UK team reportedly seeks to unionize
Financial Times reports that around 300 DeepMind employees in London are moving to unionise with the Communication Workers Union, citing anger over Google’s removal of a pledge not to use AI for weapons or surveillance and its $1.2 billion cloud deal with the Israeli military.
Generate videos in Gemini and Whisk with Veo 2
Google has launched Veo 2, an AI video generation tool, for Gemini Advanced users, allowing them to create high-quality, 8-second videos from text prompts. Additionally, the Whisk Animate feature in Google Labs now enables users to turn text and image prompts into animated clips using the same technology. Google has also ensured that videos generated using Veo 2 and Whisk will include built-in safeguards such as SynthID watermarks and policy-based content filtering.
The business of the AI labs
This excellent post explores the uncertain economics of AI labs such as OpenAI and Anthropic, questioning whether they can build sustainable businesses amid fierce competition, rapid knowledge diffusion, and commoditisation pressures from hardware providers, cloud platforms, and enterprise users. The piece analyses the strategic dynamics within the AI ecosystem and outlines potential paths forward for labs to protect margins and capture value—ranging from vertical integration and regulatory moats to building marketplaces and AI agents that function like remote workers.
Hugging Face: Tracking Energy Use in ChatUI
Modern AI models are notorious for their energy use, but none of the leading AI companies will disclose how much energy generating tokens consume. Hugging Face wants to change that and has introduced an energy tracker in its ChatUI. Whenever you make a query, the UI shows how much energy was used to generate an answer in watt-hours (Wh) and as a percentage of a phone charge. Hugging Face hopes that information about the energy used to generate an answer could one day be as visible as nutrition labels on food.
Teaching machines the language of biology: Scaling large language models for next-generation single-cell analysis
Google researchers have introduced C2S-Scale, a family of open-source large language models built on Google's Gemma models and designed to transform complex single-cell RNA sequencing data into natural language. By converting each cell’s gene expression into readable “cell sentences,” C2S-Scale enables researchers to chat with biological data, predict cellular responses to treatments, and rapidly interpret experiments, paving the way for realistic "virtual cells" and faster, more scalable biomedical discoveries. Models and resources are available on platforms such as HuggingFace and GitHub.
If you're enjoying the insights and perspectives shared in the Humanity Redefined newsletter, why not spread the word?
🤖 Robotics
China pits humanoid robots against humans in half-marathon for first time
Twenty-one humanoid robots, varied in size and design, ran alongside thousands of humans in Beijing’s Yizhuang half-marathon, marking the first time robots have competed over a 21 km course with people. Despite technical hiccups, the winning robot, Tiangong Ultra, finished in 2 hours 40 minutes, aided by advanced algorithms and battery swaps. The full hour-long livestream of the event is available on YouTube.
π0.5: a VLA with Open-World Generalization
Physical Intelligence introduces π0.5, a new vision-language-action (VLA) model that enables robots to perform tasks in unfamiliar, real-world environments. Unlike earlier models confined to familiar settings, π0.5 is co-trained on diverse data sources—including robotic demonstrations, web data, and verbal instructions—to build both physical skills and task understanding. In experiments, it successfully completed complex cleaning tasks in homes it had never seen before, demonstrating flexibility and resilience. While promising, Physical Intelligence acknowledges that π0.5 is still a work in progress, with further improvements in learning and generalisation needed to make robots even more adaptable and capable.
Waymo reports 250,000 paid robotaxi rides per week in U.S.
Alphabet’s CEO, Sundar Pichai, revealed that Waymo delivers over 250,000 paid robotaxi rides per week in the US. He also said Waymo is exploring business models and partnerships with Uber, automakers, and fleet operators, while also considering future personal ownership options. Waymo operates in San Francisco, Los Angeles, Phoenix, and Austin, and plans to launch in Atlanta this summer.
Tesla begins ‘FSD Supervised’ ride-hail tests with employees in Austin, Bay Area
Tesla has begun testing its autonomous ride-hailing service, which uses Tesla’s FSD Supervised system, with employees in Austin and the San Francisco Bay Area ahead of a planned public launch in June 2025. The service still requires a safety driver and is not fully autonomous. Tesla has completed over 1,500 trips and 15,000 miles (about 24,000 km) during internal testing, with plans to initially deploy 10 to 20 vehicles. While Tesla previously suggested a driverless launch in Austin, it is unclear whether the company will proceed without safety drivers. In California, Tesla remains limited to testing with a safety driver.
The Future of AI and Robotics Is Being Led by Amazon’s Next-Gen Warehouses
In this article, Amazon explains how the constant need to make its warehouse operations more efficient drives advancements in robotics and AI. The company is deploying systems such as robotic arms, autonomous mobile robots, and intelligent storage solutions to boost efficiency, safety, and productivity. By training AI models through real-world interactions, Amazon is creating smarter, more adaptive robots.
Musk says Tesla's Optimus humanoid robots affected by China's export curbs on rare earths
Elon Musk said China’s new export restrictions on rare earth magnets have delayed production of the Optimus humanoid robot, as China seeks assurances the materials will not be used for military purposes. Tesla is working with Beijing to secure an export licence, a process that could take weeks or months. Musk emphasised that the robots are not weapons and reaffirmed plans to produce thousands of units this year.
▶️ Protoclone in 4K | Synthetic Human with Artificial Muscles (0:59)
Protoclone is a unique robot, as it does not use traditional actuators for its movements. Instead, it uses hydraulics and artificial muscles, resulting in more natural (or creepy, depending on your perspective) movements of this synthetic human.
▶️ RIVR Partners with Evri to Deploy Autonomous Robots for Parcel Delivery (0:46)
In this video, RIVR, a Swiss robotics company solving the last-mile delivery problem with four-legged robots, announces a partnership with Evri, the UK’s largest dedicated parcel delivery company, and shares how it envisions robots delivering packages from a van to the door. A cool vision, but everyone in the UK would agree that Evri should focus on the quality of its services before thinking about using robots.
▶️ Authenticity in artwork with robotics (2:51)
Professor Patrícia Alves-Oliveira, director of Robot Studio at the University of Michigan, explores how robots and artists can cooperate—from the first brushstroke to the final piece—and how robots can serve as tools to better understand human creativity and cognition. Her research focuses on using gaze patterns and movement to trace artistic authorship, aiming to restore ownership to artists in the era of generative AI.
🧬 Biotechnology
The secrets of lab-grown chocolate
Climate change is threatening global chocolate supplies by damaging cocoa crops in West Africa, but scientists are offering hope with lab-grown chocolate. Using cellular agriculture, cocoa cells are cultivated in sugary water to create chocolate that could be healthier, free from contaminants, and quicker to produce than traditional methods. With cocoa prices soaring and ethical concerns mounting over deforestation and labour practices, lab-grown chocolate could provide a sustainable, unlimited alternative.
💡Tangents
Why the Chip Industry Is Struggling to Attract the Next Generation
The global semiconductor industry faces a major talent shortage. Countries are investing heavily in homegrown chip programs, but attracting new talent remains difficult due to outdated education models, misconceptions about pay, rigid career paths, poor documentation practices, and the high-stress culture of chip design. In this post,
▶️ The Death of Affordable Computing | Tariffs Impact & Investigation (2:58:23)
Global trade has been thrown into chaos after the US began the trade war against China and imposed new tariff policies almost every day. In this video, Gamers Nexus examines in depth, with actual numbers thanks to unprecedented behind-the-scenes access to PC manufacturers, the impact of those tariffs on the computer industry in and outside the US. It is quite a long documentary, but I highly recommend watching it if you are interested in how the US’s new tariff policies impact the computer industry and how they could make computing much less accessible.
Thanks for reading. If you enjoyed this post, please click the ❤️ button or 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!"