Microsoft denies cutting AI sales targets - Sync #549
Plus: GPT-5.2; OpenAI + Disney; Nvidia H200 chip exports to China approved; Trump signs an executive order on AI; Waymo crosses 450,000 weekly paid rides; Google to launch AI glasses in 2026; and more
Hello and welcome to Sync #549!
This week, we take a closer look at recent reports claiming that multiple divisions at Microsoft have lowered sales growth targets for certain AI products.
Elsewhere in AI, OpenAI responded to Gemini 3 Pro and Claude Opus 4.5 with GPT-5.2 and announced a partnership with Disney. Meanwhile, President Trump signed an executive order targeting state AI laws and advancing a national regulatory framework, the Department of Commerce approved sales of Nvidia H200 chips to China, Google launched Gemini 3 Deep Think, and the ARC Prize wrapped up the second year of its ARC-AGI challenge.
Over in robotics, Waymo crossed 450,000 weekly paid rides while its robotaxis began behaving like New York cabbies. We also have news of Mercado Libre announcing the deployment of humanoid robots from Agility Robotics and bio-hybrid robots made from food waste.
This week’s issue of Sync also features what a former Neuralink co-founder is building, an interview with a veteran of the semiconductor industry about recent RAM shortages, smart glasses from Google and Alibaba, and more!
Enjoy!
Microsoft denies cutting AI sales targets
When ChatGPT kicked off the AI revolution, Microsoft looked well-placed to benefit from the surge in interest, thanks to its close partnership with OpenAI. The company had exclusive access to, at the time, the best model in the world, and was offering it through Azure for developers. Additionally, Microsoft had big plans to infuse all its products with Copilots, AI features and AI tools, to the point where the company started to talk about entering its “Copilot Era.”
Three years later, the landscape looks very different. OpenAI’s lead has evaporated as rivals such as Google and Anthropic roll out models that match or outperform its offerings, with open-weight Chinese models close behind. The Microsoft–OpenAI relationship has also become less exclusive, with OpenAI diversifying its infrastructure and building new AI data centres with Oracle rather than relying solely on Microsoft.
Against this backdrop, Microsoft has found itself forced to respond to reports that its own AI ambitions may be facing headwinds. Earlier this week, the company denied a report from The Information claiming that several internal divisions had lowered sales growth targets for certain AI products after sales teams struggled to meet goals in the fiscal year that ended in June.
The report focused on Microsoft’s cloud-computing unit, Azure, a key pillar of its AI strategy. According to The Information, sales staff struggled to meet aggressive targets for tools such as Foundry, used to build AI applications, prompting lower growth expectations for the current fiscal year.
Microsoft rejected the claims, saying the report “inaccurately combines the concepts of growth and sales quotas” and insisting that aggregate AI sales quotas had not been reduced. The denial helped Microsoft shares recover from early losses, though the episode highlighted investor sensitivity to any sign that AI demand may be weaker than expected. Reuters said it could not independently verify The Information’s report.
The dispute has fuelled wider doubts about whether enterprise adoption of AI is keeping pace with investment. While technology companies have committed billions of dollars to infrastructure, many customers remain in the experimental phase rather than scaling deployments. Some have already pulled back. The Information reported that private equity firm Carlyle reduced spending on Microsoft’s Copilot Studio after encountering integration problems, a claim Microsoft declined to comment on.
Analysts say such frictions are common at an early stage. Critics, however, argue Microsoft’s challenges go beyond timing. They point to rushed rollouts and uneven product quality, particularly as competitors such as Google push more tightly integrated AI systems. Market data suggests Google’s Gemini is gaining share faster than Microsoft’s Copilot, even as ChatGPT remains the overall market leader.
All this comes despite Microsoft’s willingness to spend heavily on AI. The company reported a record capital expenditure of nearly $35 billion in its fiscal first quarter and warned that spending would rise further this year. That spending has coincided with strong growth at Azure, where revenue grew 40% in the July–September quarter, outpacing expectations. Microsoft has also said it expects to remain short on AI capacity until at least June 2026.
For now, Microsoft maintains that its AI strategy is on track. But as Google, Anthropic and others refine their offerings, early-mover advantage is no longer enough. The next phase of the AI cycle will be decided by who delivers products that customers are willing to use at scale.
If you enjoy this post, please click the ❤️ button and share it.
🦾 More than a human
After Neuralink, Max Hodak is building something even wilder
This article explains how Max Hodak, former Neuralink co-founder, is building Science Corp., a company developing BCI technology that begins with restoring vision for blind patients. It shows how Science Corp.’s long-term strategy moves from medical implants to gene therapy and lab-grown neurons that could integrate with the brain. The ultimate goal is not just treating disease but understanding and engineering consciousness itself, possibly allowing minds to merge, expand or even exist outside the body in the future.
🧠 Artificial Intelligence
Introducing GPT-5.2
OpenAI has released GPT-5.2, its most powerful AI model so far, aimed at helping professionals work faster and more accurately. The new model shows improvements in tasks such as coding, analysing data, handling long documents, and understanding images, and in some tests performs as well as or better than human experts. An in-depth article on GPT-5.2 is coming soon, so please stay tuned!
Department of Commerce approves Nvidia H200 chip exports to China
The US will allow Nvidia to sell older H200 AI chips to approved buyers in China, taking a 25% cut of the sales. Nvidia supports the decision, but it goes against efforts in Congress to block advanced chip exports over security concerns. The move follows shifting US policies on chip restrictions, during which China has begun relying more on its own technology.
Trump signs an executive order targeting state AI laws and advancing a national regulatory framework
President Trump has signed an executive order directing the federal government to curb and challenge state-level AI regulations that the administration says hinder innovation, while moving toward a single, minimally burdensome national AI framework. The order creates a Justice Department task force to sue states over conflicting AI laws, requires a federal review of existing state regulations, allows restrictions on certain federal funds for non-compliant states, and instructs agencies to develop federal standards and legislative proposals that would pre-empt state AI rules deemed inconsistent with US policy.
The Walt Disney Company and OpenAI reach landmark agreement to bring beloved characters from across Disney’s brands to Sora
Disney and OpenAI have signed a three-year partnership that will let fans create and share short AI-generated videos and images using well-known characters from Disney, Pixar, Marvel and Star Wars, with selected content appearing on Disney+. As part of the deal, Disney will also become a major OpenAI customer, invest $1 billion in the company, and work with OpenAI to develop new experiences. Both sides commit to using AI responsibly and protecting creators’ rights.
ChatGPT’s user growth has slowed, report finds
Sensor Tower data shows that while ChatGPT is still the leading AI chatbot worldwide, its growth is slowing down. Google’s Gemini is gaining users much faster, helped by Nano Banana, its new image tool, and its integration into Android devices. Other apps like Perplexity and Claude are also growing quickly. Although ChatGPT remains on top, rising competition suggests the race between AI chatbots is far from over.
OpenAI Staffer Quits, Alleging Company’s Economic Research Is Drifting Into AI Advocacy
OpenAI has allegedly become more cautious about publishing research on the potential economic harms of AI, such as job losses, as it deepens corporate and government partnerships and seeks to protect its public image, WIRED reports. Sources say the shift has frustrated researchers and driven recent departures, amid concerns that the company is prioritising favourable narratives and policy strategy over independent analysis, contrasting with rivals like Anthropic, which openly warns about disruptive labour impacts.
ARC Prize 2025 Results & Analysis
This post wraps Year 2 of the ARC Prize, an open research competition built around the ARC-AGI benchmark to track progress toward AGI. Although the Grand Prize remains unclaimed, there was clear progress: all winning work was open-source, new systems reached higher scores, commercial models were improved using refinement loops, and tiny models trained at test time showed new ways to learn. ARC-AGI-3 will launch in 2026 to push AI toward more interactive and general reasoning, and to address concerns about current benchmarks being outgrown.
Traversing the Frontier of Superintelligence
Poetiq reports that its AI system has set a new state of the art on the ARC-AGI-1 and ARC-AGI-2 benchmarks, surpassing all previous models. The system scored 65.32% on ARC-AGI-2, making it the first model to outperform the average human test taker. It is also cheaper to run than the previous leader, Gemini 3 Deep Think, which achieved 46.1% at a cost of $77.16 per task, compared with Poetiq’s $30.57 per task. Poetiq’s method uses a multi-step, self-improving and self-auditing process that uses other models, such as GPT-5.1 and Gemini 3 Pro, to solve complex tasks requiring advanced reasoning.
Block, Anthropic, and OpenAI Launch the Agentic AI Foundation
Block, Anthropic and OpenAI have joined with other tech companies to form the Agentic AI Foundation (AAIF), which aims to develop open, interoperable AI systems that can act independently. The AAIF will support community-driven projects, set industry standards and help prevent a closed, proprietary AI landscape. Initial contributions include Block’s goose framework, Anthropic’s Model Context Protocol and OpenAI’s AGENTS.md, backed by major supporters such as Microsoft, Google and AWS.
Meta acquires AI device startup Limitless
Meta has bought Limitless, the AI startup that made a pendant to record conversations. Limitless will stop selling its devices, remove subscription fees, and wind down some features, but will support current users for a year and let them export or delete their data. The company says competition from bigger tech firms made it hard to survive alone and will now help Meta develop AI wearables.
Jony Ive’s OpenAI Device Barred From Using ‘io’ Name
A US appeals court has upheld a temporary order stopping OpenAI and Jony Ive’s new hardware venture from using the name “io” for products similar to those planned by AI audio startup iyO. The court said the names could easily be confused and that iyO might be harmed because OpenAI is much better known. The case will continue, with a full hearing set for April 2026.
The AI Backlash Is Here: Why Public Patience With Tech Giants Is Running Out
AI promised to cure cancer and make our lives better, but instead, we are being flooded with AI-generated slop, deepfakes and ads no one asked for. As billions pour into tech projects that don’t seem to solve real problems, the industry looks increasingly like a massive bubble, and, as this article notes, the public backlash is already here.
Build with Gemini Deep Research
Google has announced a new, more powerful version of its Gemini Deep Research agent, now available to developers through the Interactions API. Powered by Gemini 3 Pro, the agent is designed to carry out complex, multi-step web research with high accuracy and lower costs, and has achieved state-of-the-art results across several industry benchmarks. Google is also open-sourcing DeepSearchQA, a new benchmark to measure how effectively AI agents perform real-world, multi-step research tasks.
Devstral2 and Mistral Vibe CLI
Mistral, a French AI startup, released Devstral2, its next-generation open source coding model family. Devstral2 is available in two sizes: Devstral 2 (123B) and Devstral Small 2 (24B). According to benchmark results provided by Mistral, Devstral2 reaches the same level of performance on SWE-bench Verified as DeepSeek-V3.2, Gemini 3 Pro, GPT-5.1 Codex Max and Claude Sonnet 4.5. Additionally, Mistral’s new models are 5x and 28x smaller than DeepSeek V3.2 and up to 7x more cost-efficient than Claude Sonnet at real-world tasks. Devstral2 is available on HuggingFace and for free via Mistral API. Alongside new models, Mistral also introduced Mistral Vibe, a native CLI built for Devstral that enables end-to-end code automation.
AWS Trainium3 Deep Dive | A Potential Challenger Approaching
At this year’s re:Invent conference, AWS announced, amongst other things, its new Trainium3 AI accelerators. In this article, SemiAnalysis explains how they are designed to compete with Nvidia by focusing on lower cost, faster deployment and open-source software. It outlines Trainium3’s new hardware architecture for large-scale AI training, especially for huge MoE models, and briefly looks ahead to Trainium4 as part of AWS’s long-term plan to build its own high-performance AI chips.
🤖 Robotics
Waymo crosses 450,000 weekly paid rides as Alphabet robotaxi unit widens lead on Tesla
Alphabet’s robotaxi company Waymo now provides more than 450,000 paid rides each week in the US, almost twice as many as in April, according to a letter from investor Tiger Global seen by CNBC.
Mercado Libre and Agility Robotics Announce Commercial Agreement to Deploy Humanoid Robots
Agility Robotics has announced a commercial agreement with Mercado Libre, a leading commerce and fintech ecosystem in Latin America, to integrate its Digit humanoid robots. Mercado Libre will use Digit at a fulfilment centre in Texas to handle repetitive warehouse tasks, with the goal of improving efficiency and reducing physical strain on workers. The company may later expand the technology to other sites in Latin America. It is not yet clear how many robots are being deployed or whether this is a pilot program or a full deployment. Digit is already being used for similar warehouse work by major firms such as GXO, Schaeffler, and Amazon.
SoftBank, Nvidia looking to invest in Skild AI at $14 billion valuation
Reuters reports that SoftBank and Nvidia are in talks to invest over $1 billion in Skild AI, a deal that could value the robot software maker at about $14 billion—nearly triple its worth earlier this year. Skild develops universal AI systems to help robots operate more like humans. The deal reflects growing investor interest in advanced robotics, even though experts say widespread use may still take years.
Waymo’s Self-Driving Cars Are Suddenly Behaving Like New York Cabbies
Waymo’s self-driving cars, once very polite and slow, are now driving more like humans. They are becoming more aggressive (or “confidently assertive,” as Waymo describes the new behaviour), sometimes breaking small traffic rules to keep up with busy city driving. As this article notes, some passengers like the faster, more confident behaviour, while others feel it’s less safe and aren’t sure if they can trust the cars’ new approach.
▶️ Meet HMND 01 Alpha Bipedal (1:20)
Humanoid, a London-based robotics company, unveiled HMND 01 Alpha, its first bipedal humanoid robot. The company says that this 179 cm tall robot can carry up to 15 kg payloads with three hours of runtime. The robot was built in 5 months, and it is using end-to-end reasoning and skills powered by Nvidia and KinetIQ, its own VLA-based framework.
▶️ Full-Size Humanoid Robot Oli Walks Over Construction Debris (1:10)
LimX in this video shows how its full-size humanoid robot, Oli, steps through loose sand, protruding rocks, wobbling boards, and piles of scattered debris—all while maintaining its balance without falling or tripping.
Bio-hybrid robots turn food waste into functional machines
Researchers at EPFL have built robots using discarded langoustine shells, showing that natural exoskeletons can be strengthened with synthetic parts to make durable, flexible and sustainable machines. They created grippers, manipulators and a small swimming robot, and suggested that using recycled biological materials could help make robotics more environmentally friendly.
🧬 Biotechnology
One Dose of This Gene Editor Could Defeat a Host of Genetic Diseases Suffered by Millions
Scientists have developed a new gene-editing tool called PERT that targets a common cause of many rare genetic diseases—nonsense mutations, which prematurely stop cells from making complete, functional proteins. PERT teaches cells to ignore these mistakes and make complete proteins again, potentially with just one treatment. If it proves safe, this could lead to faster, more affordable treatments for many rare diseases that currently have no cure.
💡Tangents
▶️ PS6 & RTX 6090 Delayed by Shortages? RAM Executive Explains (1:50:42)
In this video, Tom from Moore’s Law Is Dead speaks with Dave Eggleston, a veteran of the semiconductor industry with 30 years of experience working with memory, about the recent RAM price hikes caused by AI companies buying as much RAM as they can. Eggleston provides an insider’s view into recent developments in the sector, including a massive contract OpenAI has signed with Samsung and SK hynix, how the memory industry works, and what we can expect to happen in the near future. I highly recommend listening to this conversation if you want a clearer picture of what is happening within the memory industry.
Google to launch first of its AI glasses in 2026
Google plans to launch its first AI-powered glasses in 2026, developed with partners including Samsung, Gentle Monster and Warby Parker. The company will offer audio-only models featuring the Gemini assistant, as well as versions with in-lens displays for navigation and translation.
Alibaba starts selling Quark AI glasses in China, enters global wearables race
Alibaba has released its new Quark AI glasses in China, which look like normal eyewear but use the company’s Qwen AI system. The glasses link with apps like Alipay and Taobao to offer tools such as translation and price checks. Alibaba hopes the product will help it compete for future users in the AI and e-commerce market, where other tech companies, such as Meta, Xiaomi and Baidu, are also launching similar smart glasses.
SpaceX to Pursue 2026 IPO Raising Far Above $30 Billion
SpaceX is planning an IPO as early as 2026 that could raise over $30 billion and value the company at about $1.5 trillion, potentially making it the biggest stock listing ever. The move is fuelled by the success of its Starlink internet service and development of its Starship rocket, with funds expected to support future space technology such as data centres. The company is also selling shares internally at a valuation above $800 billion as investor interest continues to grow.
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!"