Open-weight models are catching up to the leaders, but safety is lagging
A report by SaferAI shows that Z.ai's open-weight model, GLM-5.2, is approaching the frontier in capabilities. However, it practically lacks key safety mechanisms. This revives the debate over whether powerful open models can evolve faster than risk management institutions.
In my view, this trend highlights a fundamental market paradox. The commercialization of AI requires openness for developers — it lowers the cost of adoption. But when an open model matches the power of flagship solutions, the risks of uncontrolled use multiply. The gap between capabilities and protection is becoming a critical vulnerability.
Anthropic and Volta: $10 billion for compute power
Anthropic has signed a $10 billion contract with AI cloud startup Volta. This is part of the company's aggressive strategy to expand its cloud partnerships. Deal volumes of this scale confirm that the demand for compute exceeds supply, and labs are willing to pay top dollar for access to GPUs.
This contract is an indicator of where real value is being created in the AI industry. Model developers are competing at the algorithm level, but the physical infrastructure remains the bottleneck. Cloud providers specializing in AI are becoming the main beneficiaries of the boom, monetizing a fundamental resource deficit.
Nvidia is capturing AI safety standards
Nvidia is leading the creation of the Open Secure AI Alliance. In just a week, the alliance has united over 120 companies and already presented its first proposals for protection against AI agents. Nvidia isn't just selling chips; it's shaping the architecture of trust in technology.
In parallel, the company released the open-weight Alpamayo 2 Super model for robotaxis and autonomous vehicles. The model is designed to handle rare and complex driving situations that require a deep understanding of context rather than simple object recognition. Nvidia is building an ecosystem where its safety standards and models become the foundation for the entire autonomous driving industry.
AI monetization in media: Spotify's strategy
Spotify is expanding its AI remixes and covers project, securing support from Merlin, which represents over 30,000 independent labels. Universal Music Group had previously joined the initiative. The tool will allow fans to create AI-generated covers of music from participating artists.
This is a rare example of a working B2C (business-to-consumer) monetization model for generative AI. The platform turns AI from a copyright threat into an engagement tool and a source of additional revenue. Spotify's strategy shows how proper licensing removes legal barriers and unlocks new revenue streams.
Robotaxis shift from testing to scaling
Waymo has opened access to its robotaxi service in Dallas to everyone, dropping waitlists. This is another step in the company's campaign to scale self-driving technologies in the US, UK, and Europe.
Lifting restrictions indicates that technological readiness is moving into a commercial phase. Companies are no longer testing hypotheses — they are capturing market share. Scaling in new metropolises directly depends on infrastructure and regulatory approvals, making leaders like Waymo the primary contenders for a regional monopoly.
AI agents in coding: A budget crisis
AI agents for writing code are rapidly burning through corporate budgets. At Kilo Code, engineers manually write or read code only about 1% of the time — agents do the rest. This is forcing teams to rethink their approaches: which systems to trust AI with, who is responsible for errors, and how to manage multi-agent systems.
The issue of API costs is becoming critical. Autonomous agents make thousands of calls to optimize code, leading to unpredictable expenses. In my view, the next step for the market will be the emergence of specialized models optimized not for quality, but for the economic efficiency of performing routine tasks.
Bottom line
Infrastructure and standardization are becoming the new battlefields in the AI market. While labs compete on model power, compute providers and industry standard creators are concentrating the bulk of the profits. Expect consolidation among cloud providers and increased pressure on agent developers to reduce operating costs.
Sources
- TechCrunch: Open-weight AI models are catching up
- TechCrunch: Anthropic signs $10B deal with Volta
- TechCrunch: Nvidia leads Open Secure AI Alliance
- NVIDIA Blog: Alpamayo 2 Super for Autonomous Vehicles
- TechCrunch: Spotify expands AI remix with Merlin
- TechCrunch: Waymo opens up robotaxi service in Dallas
- VentureBeat: AI coding agents are blowing through budgets
