AI and the Past
Gen AI as a concept has seen a large shift in sentiment over the past year; initially the imperfections and the glitches that came with models such as ChatGPT 4o, Claude 4 Sonnet, Gemini 1.5 Pro etc. implied doubt on its future. However, with new models coming such as Claude Opus, GPT 5, Gemini 2.5 pro and Grok 4 there has been a significant jump in the perceived intelligence and reliability of these new-gen models. These models enable agentic workflows, seamless integration, and diversification beyond text to full video and audio production, delivering real-world value for both consumers and enterprises. Agentic AI is leading this race, providing real world solutions to a plethora of problems providing never seen before levels of automation and efficiency in workflows, AI and its adoption is at breakneck speed, as evidenced by a single customer purchasing $10.7 billion worth of NVIDIA chips in one quarter.
Meta’s slow but potent entry into the AI world
While Meta initially seemed to lag behind OpenAI, Google, and Anthropic in the AI race, this was actually strategic patience. Meta released Llama as a capable but not industry- leading model, popular among developers for its open-source nature and customization abilities, but lacking the intelligence of competitors’ flagship offerings. Rather than rushing to compete when AI reliability was still questionable, Meta was waiting for the technology to mature. Now that AI has reached current reliability standards, Meta has completely reversed course, investing heavily with massive talent acquisition and infrastructure spending to challenge the leaders.
Meta’s AI Spends and Mega Poaching
Over the past 6 months, Meta has aggressively recruited top AI talent from leading companies and established its Super-Intelligence Lab with massive investment commitments. Their AI spending has surged from $39.2B in FY2024 to a projected $66- 72B in FY2025—a roughly $30B increase year-over-year. This funds AI servers and multi- gigawatt data centres, targeting 350,000 Nvidia H100 chips by end-2024. Zuckerberg has committed “hundreds of billions” over multiple years for titan clusters, starting with Prometheus in 2026 and the scalable Hyperion facility supporting up to 5 gigawatts.
| NAME | COUNTRY OF ORIGIN | SOURCE COMPANY | ESTIMATED COMPENSATION | FORMER ROLE/TEAM |
| RUOMING PANG | China | Apple | >$200M (multi-year) | Head of Foundation Models Team |
| ALEXANDR WANG | USA (Chinese desc.) | Scale AI | $14.3B co. deal, exec pay | Co-founder & CEO |
| TRAPIT BANSAL | India | OpenAI | ~$100M (multi-year) | Researcher, o-series reasoning models |
| JACK RAE | UK | Google DeepMind | Not disclosed | Principal Researcher (pre-training) |
| SHENGJIA ZHAO | China | OpenAI | Not disclosed | Researcher, GPT-4 contributor |
| JIAHUI YU | China | OpenAI/Google | Not disclosed | Researcher, DeepMind alumnus |
| LUCAS BEYER | Belgium/Germany | OpenAI/Google | Major offer, $100M rumours | ViT (Vision Transformer) co-creator |
| JOHAN SCHALKWYK | South Africa | PlayAI (prev. Google) | Not disclosed | Software Fellow, PlayAI Lead |
| YUANZHI LI | China | OpenAI | Not disclosed | Researcher |
| ANTON BAKHTIN | Russia | Anthropic | Not disclosed | Researcher |
| NAT FRIEDMAN | USA | GitHub/Microsoft | Not disclosed | Former CEO of GitHub, Superintelligence Lab Lead |
| DANIEL GROSS | Israel | Cue/Apple | Not disclosed | Founder, Superintelligence Lab Leadership |
Poaching from the best at unbelievable prices
Base salary ranges between $10M-$100M+ signing bonuses and large RSU grants. Top hires get $100M-$300M total packages.
Where They Hire From
- OpenAI: ~40% of AI team hires
- Google DeepMind: ~20%
- Apple: Key leaders like Ruoming Pang
- As well as talent from Anthropic, Scale AI, GitHub
The $14.3B Scale AI deal (49% stake) landed CEO Alexandr Wang as Meta’s Chief AI Officer running their Superintelligence Lab. They are also buying whole startups like PlayAI to get entire teams at once.
What does this mean for India?
Reliance is partnering with Meta in a ₹855 crore joint venture to build business AI solutions using Meta’s Llama models. The 70/30 partnership will create AI services for Indian companies across sales, customer support, and IT operations. By combining Llama’s flexible, customizable technology with Reliance’s infrastructure at Jamnagar— including data centres, renewable energy, and Jio’s network—they can offer AI solutions at lower costs than global cloud providers. This timing works well as Meta heavily invests in AI talent and technology, meaning the partnership will benefit from improving Llama capabilities while giving Indian businesses faster setup, lower costs, and local data hosting. For Reliance this is most likely a strategic answer to perplexity entering the Indian market through airtel and Open AI offering its India tailored plan at ₹399 a month. Regards,
Team Sowilo
