AI Panic, Sovereign Pacts, and Robots With Skin

What happened
A financial recap from Sophic Capital notes that the tech and media rally became a sharp slide as investors started asking whether generative AI will disrupt the companies they hold. The sell‑off erased about $1 trillion in software‑stock value, even as AI firms raised eye‑popping sums (Anthropic closed a $30 billion round and is now valued at $380 billion, while Databricks raised $7 billion). Columnist Matt Shumer highlighted how advanced AI coding models let him describe an app and watch an AI agent build and test it in real time. Such capabilities fuel enthusiasm but also fear; he warned that many professional jobs could be automated and pointed out that this month’s AI‑driven sell‑off wiped out more than a trillion dollars of market capitalization. Amid the volatility, there were reports that U.S. authorities are exploring tariff carve‑outs for hyperscalers importing AI hardware.

Why it matters
The market turbulence underscores how quickly sentiment can swing when investors think AI might cannibalise existing business models. As cutting‑edge models handle more complex tasks, companies without defensible data or infrastructure are vulnerable. At the same time, fundraising and policy developments indicate that capital and governments are still betting heavily on AI’s long‑term upside.

What happened
At the Munich Security Conference, Canada and Germany signed a Joint Declaration of Intent on AI and launched a Sovereign Technology Alliance to expand secure compute capacity and accelerate responsible AI research. The alliance aims to coordinate funding and talent programs while ensuring that AI development aligns with democratic values. Canada’s AI minister Evan Solomon said the partnership recognizes that “AI is the future of economic strength and national security”; Germany’s digital minister Claudia Plattner stressed the need for guardrails so AI benefits society.

Why it matters
With AI hardware supply chains concentrated in a few countries, Canada and Germany are teaming up to build sovereign capacity and influence standards. Their public statements signal that democratic nations want to shape AI governance, not just reap economic gains. Expect similar alliances to emerge as nations race to secure compute and talent.

What happened
A Gartner‑inspired roundup warns that agentic AI—autonomous agents capable of executing tasks and adapting—creates a new attack surface. The report predicts organizations will need dynamic identity and access management policies, continuous authorization and advanced oversight to mitigate risks. Regulators are also expected to accelerate rulemaking, pushing companies to build cyber‑resilience amid geopolitical volatility and the looming arrival of quantum computing.

Why it matters
As organizations deploy autonomous agents to automate workflows, the security model has to evolve. Static controls and periodic reviews won’t catch prompt‑injection attacks or agents gone rogue. The takeaway: build layered defenses and adopt new governance frameworks before scaling up agentic systems.

What happened
Arista Networks’ earnings call revealed that its 2025 revenue surged to $9 billion, but CEO Jayshree Ullal said the company is battling a memory shortage and “horrendous” prices due to AI demand. Arista expects to double its AI‑networking revenue to $3.25 billion in 2026 and is positioning itself as a partner for a broader ecosystem that includes AMD, Anthropic, ARM and Broadcom—not just Nvidia.

Why it matters
The AI boom is straining supply chains not just for GPUs but for memory modules too. Hardware diversification is accelerating as networking providers and chipmakers work together to mitigate shortages and reduce dependency on Nvidia’s stack. Enterprises deploying agentic AI should prepare for ongoing supply constraints and consider alternative hardware.

What happened
ByteDance rolled out Doubao 2.0, a big leap toward the agentic era. The upgraded model boasts complex reasoning, multi‑step task execution comparable to OpenAI’s GPT 5.2 and Google’s Gemini 3 Pro while reducing costs by roughly an order of magnitude. Doubao’s user base has swelled to 155 million weekly active users, and ByteDance claims its cost advantage will let it serve tasks at scale.

Why it matters
The launch intensifies China’s homegrown competition in AI and signals that cost‑efficient agentic models are coming to the mass market. Doubao’s user metrics highlight a huge installed base, making it a platform to watch as ByteDance pushes beyond entertainment into productivity and services.

What happened
Chinese startup UniX AI unveiled Panther, a series of embodied robots designed for semi‑structured environments. Each robot features bionic dual arms with eight degrees of freedom, an omnidirectional chassis for sliding and rotating, a vertical lift of around 80 cm and a 48‑volt power system. The company demonstrated multi‑robot coordination in a kitchen, wine‑pouring with precise flow control, playing cards and even performing embroidery and golf swings. These feats are powered by UniX’s “Trinity” architecture: UniFlex for imitation learning, UniTouch for visuotactile fusion and UniCortex for decomposing long‑horizon tasks. UniX plans to commercialize the robots in eldercare facilities and commercial settings before expanding to homes.

Why it matters
The demonstrations highlight rapid progress in physical AI. Combining dexterous hardware with visuotactile perception and high‑level planning, the Panthers can execute complex tasks with minimal hard coding. For aging societies and labour‑starved industries, such robots could fill gaps—if they can be produced and deployed at scale.

What happened
DroidUp (Zhuoyide) just unveiled Moya, the world’s first biometric AI robot, at Shanghai’s Zhangjiang Robotics Valley. Moya sports warm, human-like skin, camera eyes for reading microexpressions, and a walking gait that’s 92% human-accurate. It’s built on the Walker 3 skeleton, already famous for its marathon performance.

Why it matters
Moya isn’t just a tech demo—it’s headed for real-world jobs in healthcare, education, and public service, from train stations to banks. With a late-2026 release and a $173,000 price tag, Moya signals that socially integrated, lifelike robots are moving from sci-fi to city streets. China’s rapid progress in humanoid robotics, showcased at the World AI Conference, cements its lead in embodied AI.

What’s next
Expect a wave of biomimetic robots in public spaces, raising new questions about trust, privacy, and the future of human-robot interaction.

What happened
India launched PARAM-2, a 17-billion-parameter multilingual model supporting all 22 official Indian languages, as part of the BharatGen initiative. Meta’s Llama 4 series is now open-weight, natively multimodal, and available via 25+ cloud partners. Meanwhile, Arcee AI’s Trinity (400B parameters) is now the largest open-source LLM from a U.S. company, directly challenging Meta’s dominance.

Why it matters
The generative AI arms race is going global and open. India’s public digital goods approach sets a new bar for sovereign AI infrastructure, while open-source behemoths like Trinity are lowering the barrier for startups and researchers. Security and safety are front and center, with Meta rolling out tools like Llama Guard and Prompt Guard .

What’s next
Expect a flood of local-language, open-source, and multimodal models—plus a new wave of creative and regulatory challenges as AI-generated content gets ever more realistic.

Market jitters, multi‑billion‑dollar alliances and a steady stream of new models and robots illustrate AI’s split personality—simultaneously racing ahead and confronting structural challenges. As countries build sovereign compute, companies experiment with embodied agents and investors fret about disruption, the path to the agentic future will be anything but smooth. Stay tuned for daily insights.