AI: The Most Important News of the Week (September 29 – October 5, 2025)

Top 5 AI stories: leading researchers leave Big Tech for Periodic Labs, Sora 2 with synchronized audio, record investments, and new governance frameworks.

Every Monday, we select and analyze the 5 most significant news stories from the world of artificial intelligence. Not just a simple summary, but a critical reading of the developments that are truly changing the industry. Without hype, without unnecessary technical jargon.

Why 5 stories? Because they are enough to stay updated without being overwhelmed by information.

1. The Brain Drain: AI Researchers Choose Real Science

More than 20 of the top researchers from OpenAI, Google DeepMind, Meta, and Apple have simultaneously left their multi-million dollar positions to found Periodic Labs, a startup aiming to "automate scientific discovery" instead of producing ever more sophisticated chatbots. A talent migration that redefines the priorities of artificial intelligence.

🔍 What happened: Periodic Labs, founded by Liam Fedus (former VP of Research at OpenAI and co-creator of ChatGPT) and Ekin Dogus Cubuk (former head of DeepMind's materials team), has raised $300 million in seed funding from investors of the caliber of Andreessen Horowitz, Nvidia, Jeff Bezos, and Eric Schmidt. The goal is to build "AI scientists" – autonomous systems capable of conducting physical experiments through robotic labs, analyzing results, and iterating without human intervention.

💡 Why it matters:
This is not just another startup: it represents an implicit critique of the direction commercial AI is taking. Researchers are voting with their feet, choosing concrete scientific progress over the race-to-AGI and "supersized chatbots." Periodic Labs plans to generate fresh data from the physical world – something LLMs cannot do – to overcome the problem of internet data exhaustion. The initial focus on room-temperature superconductors could have a greater impact than a thousand new versions of ChatGPT. This exodus signals that the best talents believe the true future of AI is not in making computers talk better, but in making them think scientifically.

🎯 Our take: It's interesting that this migration is happening just as Big Tech announces billion-dollar investments in data centers and AGI. Periodic Labs represents a contrarian bet: instead of infinite parameter scaling, they choose deep integration with the physical world. If successful, we might look back at 2025 as the year AI stopped being just software. As we already explored in our article on AI-driven startups, new ventures are completely redefining the sector.

Source: New York Times | TechCrunch

2. Sora 2 and the AI Video War: When Everyone Can Be a Deepfake

OpenAI has launched Sora 2 with synchronized audio and a "cameo" feature that allows users to insert their own face into AI-generated videos. Meanwhile, Anthropic has released Claude Sonnet 4.5 for enterprise coding, while Meta has presented Vibes, its social response to AI videos. The competition is shifting from text generation to multimodal generation.

🔍 What happened: Sora 2 brings significant improvements in realistic physics (a basketball missing the hoop now bounces correctly), synchronized audio with dialogue and sound effects, and clips up to 20 seconds in 1080p. OpenAI also launched a TikTok-style social app for sharing and remixing AI videos. Anthropic responded with Claude Sonnet 4.5, which shows superior performance in coding and using agentic tools. Meta introduced Vibes in its Meta AI app.

💡 Why it matters:
The battle has shifted from textual AI to multimodal AI, with profound implications for content creators, marketers, and… truth. Sora 2's cameo feature democratizes personalized deepfakes, although OpenAI requires verified consent. The social approach by OpenAI and Meta signals that AI video is no longer just a professional tool but a consumer feature. Claude Sonnet 4.5 demonstrates that the competition isn't just about who generates the most realistic videos, but about who offers the most versatile tools for different use cases. The simultaneity of these launches suggests that Big Tech has understood that 2025 will be the year of AI video, just as 2023 was the year of AI text.

🎯 Our take: OpenAI calls Sora 2 "the GPT-3.5 moment for video" – and they might be right about the scale, but not about the type of impact. GPT-3.5 democratized AI writing. Sora 2 could democratize visual manipulation. The difference is not small: generated text can be fact-checked, generated video cannot. The social feature is brilliant from a business perspective but concerning from a social one. In our article on artistic deepfakes we had already highlighted how the boundary between digital art and reality manipulation is dangerously dissolving.

Source: TheAITrack | TechRadar

3. Nvidia Bets $500 Million on OpenAI While Oracle Expands Infrastructure

In a week of mega-deals, Nvidia invested record sums in OpenAI as part of a broader strategy to consolidate control over global AI infrastructure. Simultaneously, OpenAI announced a partnership with Oracle to massively expand computational capacity.

🔍 What happened: Nvidia participated in OpenAI's investment round, contributing approximately $500 million, strengthening its position as both a hardware provider and a strategic investor in major AI companies. OpenAI simultaneously announced an infrastructure expansion with Oracle that will multiply the computational capacity available for training and inference of advanced models. In the USA, new government policies were presented to regulate AI chip exports and infrastructure investments.

💡 Why it matters:
Nvidia is consolidating a unique position: it is simultaneously the seller of GPUs, the investor in the companies that buy them, and a competitor through its cloud services. This vertical integration creates concerning power dynamics where a single player controls hardware, capital, and distribution. The OpenAI-Oracle alliance signals that even giants need partnerships to scale the infrastructure required for next-generation models. The new US policies demonstrate that governments are finally understanding that AI infrastructure is as strategic as energy or telecommunications infrastructure. Control of chips becomes geopolitical control.

🎯 Our take: Nvidia is playing chess while others are playing checkers. Investing in OpenAI while supplying the GPUs creates formidable lock-in that no competitor can easily replicate. But this concentration of power could attract antitrust interventions that would fragment the market. As we analyzed in our article on AI on a Leash, reflections on machine control become increasingly urgent when a few players control the entire infrastructure.

Source: AI Architects

4. Meta Launches Anti-Regulation Super PAC: AI-Powered Lobbying

While launching new AI products, Meta has activated a political super PAC to counter proposed AI regulations in the US and Europe. A move that raises questions about how much power tech companies should have in determining the rules that govern them.

🔍 What happened: Meta created and funded a super PAC focused on lobbying campaigns against AI regulations at the federal and state level in the US, and against the European AI Act. The PAC is supporting political candidates favorable to a "light touch" regulatory approach and funding think tanks that produce research against restrictions. Simultaneously, Google DeepMind announced the new Robotics-ER 1.5 models for advanced physical robotics.

💡 Why it matters:
Big Tech is moving from passive defense to active political offense. Meta, burned by social media and privacy regulations, wants to prevent the same from happening with AI. The timing is not coincidental: as the European AI Act comes into force and the US considers federal regulations, companies are investing hundreds of millions in lobbying to dilute or block the rules. This raises fundamental questions about the relationship between technological innovation and democratic governance. DeepMind's Robotics-ER 1.5 models demonstrate that AI is becoming increasingly physical and embodied, making the regulatory debate even more urgent.

🎯 Our take: The paradox is stark: the same companies that publicly claim to want "responsible" AI regulations are spending fortunes to block them. Meta is right on one point – poorly designed regulations could stifle innovation. But the idea that tech companies should self-determine the rules is as problematic as letting banks regulate themselves. As we highlighted in our deep dive on regulating artificial intelligence, the question "who decides the rules of the game?" becomes increasingly critical as AI permeates every aspect of society.

Source: AI Architects | CGS Daily Digest

5. Enterprise AI Trends: 5G Monetization and Pharma AI Boom

New market data shows how enterprise AI is consolidating in specific sectors, with particular growth in 5G-AI integration for telecommunications and in the pharma AI boom for drug discovery. Blacksmith raises $10 million for AI-powered continuous integration.

🔍 What happened: Market reports highlight explosive growth in enterprise AI, with a particular focus on: integration between 5G networks and AI for network optimization and predictive services; pharmaceutical AI accelerating drug discovery with investments exceeding $3 billion in Q3 2025; startups like Blacksmith raising significant capital to democratize enterprise AI tools previously accessible only to large corporations. AGI (Artificial General Intelligence) trends show convergence between academic research and commercial applications.

💡 Why it matters:
AI is finally moving out of the "impressive demo" phase and into the "measurable ROI" phase. The 5G-AI integration is particularly significant because telcos have massive infrastructures to optimize and margins under pressure – AI thus becomes an economic necessity, not an experiment. Pharmaceutical AI is showing concrete results: molecules discovered in months instead of years, accelerated clinical trials, reduced R&D costs. Blacksmith and similar companies are democratizing capabilities that were the monopoly of Big Tech, lowering the barrier to entry for mid-sized companies. AGI trends suggest that the convergence between research and business is accelerating, with implications for development timelines.

🎯 Our take: These signs of enterprise AI market maturation are positive but create new risks. When AI becomes "mission critical" for telecommunications or pharmaceuticals, failures have amplified consequences. A bug in a network optimization algorithm can cause blackouts. An error in AI drug discovery can cost lives. Robust technical governance is needed, not just commercial hype. In our article on Medical AI, we have already explored how the future of digital health requires reliability standards completely different from consumer entertainment.

Source: Artificial Intelligence News | CGS Daily Digest

📊 What these developments tell us

This week in artificial intelligence tells us five stories that converge towards a fundamental crossroads: AI for what?

Science reclaims AI: The exodus to Periodic Labs is not a simple talent migration – it's a vote of no confidence in AI-as-entertainment. The best minds are saying: enough with chatbots, we want discoveries. Artificial intelligence returns to its roots: a tool to expand human knowledge, not to replace human conversation.

Video becomes a weapon (or a toy): Sora 2 democratizes visual manipulation with the same ease with which TikTok democratized dances. OpenAI and Meta's social approach transforms deepfakes from a geopolitical threat to a consumer feature. It's unclear whether this makes the problem more manageable or simply more pervasive. Perhaps both.

Infrastructure becomes geopolitical: Nvidia investing in OpenAI while supplying it with GPUs is not business – it's imperial strategy. Control of AI chips becomes the oil of the 21st century, and we are witnessing the birth of new, vertically integrated technological empires. US policies show that governments have understood this, albeit belatedly.

Lobbying enters the algorithmic era: Meta funding a super PAC against regulation marks the beginning of a new phase: Big Tech no longer defends itself, it attacks. The paradox is stark – the same companies that publicly call for "responsible regulation" spend fortunes to block it. Democracy meets algorithmic power, and it's not an equal meeting.

Enterprise AI Comes of Age: 5G-AI, Pharmaceutical AI, Continuous Integration AI – artificial intelligence is moving out of the demo phase and into the mission-critical one. This is positive for ROI and adoption, but it also means that failures will have amplified consequences. A bug in network AI can cause blackouts. An error in pharmaceutical AI can cost lives.

The Paradox of the Week? The brightest researchers are abandoning projects aimed at replicating human intelligence to build systems that do things humans cannot. Perhaps AGI is not the goal – it's the distraction.

The Inconvenient Truth: We are witnessing the bifurcation of artificial intelligence. On one side, spectacle AI: viral videos, increasingly human-like chatbots, algorithmic social feeds. On the other, tool AI: scientific discovery, industrial optimization, concrete problem-solving. Big Tech invests billions in the former. The best researchers are betting on the latter. Who will be right will determine whether AI is remembered as the technology that entertained us or the one that helped us progress.

💬 Your Opinion Matters: Which of these news stories do you think will have the greatest impact? Write to us or share on social media.