This week in AI, Elon Musk’s xAI raised $20 billion and secured a major Department of Defense contract, while Apple partnered with Google to use Gemini as the foundation for its next-gen AI features, sidelining OpenAI. At Davos, AGI leaders predicted rapid progress toward artificial general intelligence, and new research and startups like DeepSeek’s Engram and Kilo Code signaled ongoing innovation in AI architecture and developer tools.
This week saw major developments in the artificial intelligence industry, with Elon Musk’s xAI closing a record-breaking $20 billion Series E funding round. This round values xAI at $230 billion, putting it in the same league as OpenAI and Anthropic. The funds will be used to expand xAI’s Colossus supercomputers in Memphis, Tennessee, which now host over a million H100 GPU equivalents. Despite facing regulatory investigations in multiple countries due to Grok’s generation of inappropriate deepfakes, xAI secured a Department of Defense contract, making Grok the DoD’s AI agent platform. This demonstrates that investors are willing to look past early-stage safety issues, betting on the long-term value of AI.
At the World Economic Forum in Davos, two leading figures in AGI—Dario Amodei of Anthropic and Demis Hassabis of Google DeepMind—shared a stage to discuss the future of artificial general intelligence. Both agreed that AGI is approaching faster than most people realize and that entry-level white-collar jobs are at significant risk. Amodei predicted AGI could arrive as soon as 2026 or 2027, driven by AI’s ability to write its own code, while Hassabis was more conservative, estimating a 50% chance by the end of the decade. Hassabis argued that even if AI can do 95% of a job, the remaining 5%—which only humans can do—becomes even more valuable, suggesting that AI will augment rather than fully replace human workers in the near term.
Another major story was Apple’s announcement of a multi-year partnership with Google, making Google’s Gemini models and cloud technology the foundation for Apple’s next-generation AI features. This is a significant setback for OpenAI, which lost out on a billion-dollar annual deal. Google is reportedly building a custom 1.2 trillion parameter Gemini model specifically for Apple, far surpassing Apple’s own AI capabilities. This move positions Gemini as the default AI on both Android and iOS, increasing pressure on OpenAI to find new distribution channels, possibly through hardware collaborations.
In research news, DeepSeek introduced Engram, a new conditional memory architecture designed to address the limitations of transformer models in knowledge retrieval. Traditional transformers require multiple layers of attention for simple lookups, making them inefficient. Engram uses hash functions to quickly retrieve short sequences from a large embedding table, filtering them through a gating mechanism for context relevance. This approach significantly improves token efficiency and could pave the way for models with more robust factual memory, marking a notable breakthrough in AI architecture.
Finally, Kilo Code, founded by GitLab co-founder Sid Sijbrandij and CEO Scott, launched its app builder after a rapid development sprint and $8 million in seed funding. Unlike competitors like Replit and Lobe, Kilo Code targets professional engineers with an open-source, engineering-friendly platform akin to VS Code. The company’s fast-paced development and focus on flexibility, reliability, and integration with existing tools set it apart in the increasingly crowded AI coding market. As the market matures, Kilo Code’s bet is that engineers will prefer specialized tools over general-purpose solutions, and it remains to be seen if there’s room for another major player alongside Cursor and Claude Code.