Today's Highlights:

• Meta's $60B AMD Bet Is Infrastructure, Not Just a Deal. Meta's commitment to buy up to $60B in AI chips from AMD over five years — plus an option to acquire up to 10% of AMD — is one of the most consequential supply chain decisions in Big Tech history. It signals Meta is building AI infrastructure that won't depend on Nvidia alone and hedges against GPU supply volatility heading into a heavy inference scaling cycle. Tech Startups

• Databricks' Agent Bricks Are Now Production-Ready. Databricks made its Agent Bricks platform (formerly Agent Framework) generally available this week, letting developers deploy custom AI agents as fully managed Databricks Apps on serverless compute. It's a direct shot at the fragmented agentic tooling market — and a signal that enterprise AI is moving from prototype to production. Solutions Review

• Agentic Browser Security Is the Newest Attack Surface. Zenity Labs dropped a technical deep dive this week showing how agentic browsers like ChatGPT Atlas and Perplexity Comet — which can read authenticated content autonomously — create a "shadow AI" risk layer inside enterprises. Their new open-source Safe Harbor tool lets agents escape malicious workflows. This is the security story most teams haven't caught up to yet. Solutions Review

• SolveAI Emerges from Stealth with $50M. The London-based startup raised $50M at launch, targeting enterprise-grade AI coding — not consumer demos. The thesis: AI-generated code needs to fit internal architecture standards, security controls, and compliance requirements from day one. A crowded space, but the "last-mile" enterprise problem is real. Tech Startups

Deep Dive:

1. India's AI Summit: The World's Next Compute Battleground

This week, New Delhi became the center of the AI universe. India's four-day AI Impact Summit brought together every major AI lab CEO under one roof — Sam Altman, Dario Amodei, Sundar Pichai, and Demis Hassabis all attended, alongside Prime Minister Modi and French President Macron. The scale of the gathering signals that India is no longer positioning itself as a consumer of AI products but as a sovereign infrastructure player.

The headline number: Adani Group announced plans to invest $100 billion to build AI data centers powered by renewable energy by 2035, with an additional $150B in downstream effects across server manufacturing and sovereign cloud. The Indian government also earmarked $1.1B for a state-backed VC fund targeting AI and advanced manufacturing. Meanwhile, home-grown AI company Sarvam teased its first hardware product — smart glasses called Sarvam Kaze — built on top of their suite of Indian-language speech and vision models. This is no longer just about outsourced engineering talent. India is building the full stack.

TechCrunch

2. Microsoft's Memory Poisoning Warning — and Why It Matters Now

Microsoft researchers disclosed a new attack vector this week: manipulated "Summarize with AI" links that embed hidden instructions capable of altering chatbot memory and biasing future recommendations — invisibly, persistently, across sessions. It's not theoretical. Over 30 organizations across finance, healthcare, legal, and SaaS were observed attempting variations of this tactic in the wild.

Microsoft has classified this as "memory poisoning" and deployed mitigations in Copilot, but the deeper problem is systemic. As AI assistants become ambient recommendation engines embedded across workflows, bad actors now have a new vector: not attacking users directly, but corrupting the AI's long-term context instead. Think of it as subliminal advertising at the infrastructure layer. Detection is hard because there are no visible cues to end users. Security and compliance teams need to add AI memory auditing to their threat models — yesterday.

MarketingProfs / Microsoft Research

Global AI News

• EU publishes first draft Code of Practice on AI-generated content transparency, requiring machine-readable labeling of deepfakes and synthetic text under the AI Act. A second draft arrives in March; final rules apply from August 2026. Fladgate

• EU Council expanded the EuroHPC mandate to build large-scale AI gigafactories across member states, adding a quantum technologies pillar — a direct move to close the compute gap with the US and China. Fladgate

• London Mayor Sadiq Khan launched an AI and future-of-work taskforce, focused on sectors like finance, creative industries, and professional services — framing the city's labor market shift as "seismic." Fladgate

• Samsung is targeting 800 million Gemini-equipped devices by end of 2026, pushing advanced AI features down to mid-tier and budget Android devices in a major expansion of the Google partnership. Crescendo AI News

Market Trends:

• Meta's $60B AMD chip commitment isn't just a purchase agreement — it includes an equity stake option of up to 10% of AMD. This is infrastructure investment as M&A strategy, blurring the line between supply chain and ownership. Tech Startups

• SambaNova raised $350M (led by Vista Equity Partners, with Intel Capital) to scale its SN50 inference chip and SambaCloud platform. SoftBank Japan is its first named customer, deploying SN50 chips in Japanese AI data centers — a strong early commercial signal for the Nvidia alternative thesis. Tech Startups

• Axelera AI pulled in $250M Series B for edge AI semiconductors — another week, another major non-Nvidia chip bet funded. The thesis: inference will move to the edge, and whoever owns that silicon owns the next margin layer. Tech Startups

• Global M&A is hitting record levels, with AI-related deals driving mega-transaction volume. Deals over $5B accounted for more than 73% of total M&A value growth in 2025, per Bain — and Goldman expects that momentum to accelerate further into 2026 as companies consolidate around AI capabilities. CNBC

• World Labs led this week's largest funding round with a $1B raise for its spatial AI models, according to Crunchbase — continuing a pattern of $1B+ rounds becoming normalized in AI infrastructure. Crunchbase News

Want to advertise?