Happy Tuesday. Meta made its enterprise ambitions official on Sept. 28, announcing a full enterprise AI platform and pulling MongoDB's CEO to run it. The move puts Meta in direct competition with Microsoft, Salesforce, and ServiceNow for business software dollars, and it arrives under real pressure: the company has committed up to $145 billion in capital expenditure this year with no cloud business to show for it.
The rest of today's issue tracks the infrastructure race beneath that headline. AMD bought spatial AI startup World Labs for $8.2 billion. Anthropic cut per-task costs 30% without touching its prices. Okta rallied a dozen vendors around agent security standards. And a zero-day scare forced a global file-transfer company to order server shutdowns on a Saturday morning.
ENTERPRISE SOFTWARE
Meta Bets $145B in Capex on a Business Software Play
Meta announced its enterprise AI platform on Sept. 28 and hired MongoDB CEO CJ Desai to lead it as Chief Enterprise Platform Officer, reporting to Zuckerberg. The platform bundles Meta's Muse AI assistant, its business agents running on WhatsApp and Instagram, and a developer API under a single offering aimed at business customers. No pricing or launch date has been announced.
For SaaS founders, this is a real market entrant, not a research project. Meta has 100 million-plus business relationships and agent conversations on its messaging apps that grew from 1 million to 10 million weekly in a single quarter. Desai's ServiceNow background signals Meta wants to sell workflow automation, not just AI features. Watch how it prices against incumbents once it opens availability, and whether its lack of a traditional cloud infrastructure puts enterprise IT buyers off.
FUNDING & M&A
AMD Pays $8.2B for Fei-Fei Li's Spatial AI Startup
AMD agreed Sept. 26 to acquire World Labs for $8.2 billion in stock, a 60-plus percent premium over the startup's $5 billion February 2026 valuation. Fei-Fei Li, who co-founded World Labs in 2024, will join AMD as Executive Vice President and Chief Scientist reporting to CEO Lisa Su. World Labs' Atlas model generates 3D environments from photos and video, with applications in robotics and physical simulation.
This deal is about chip differentiation, not just model capability. AMD had already invested in World Labs' February round and demonstrated a 4x speed gain running its Marble product on MI325X chips. Owning the model team closes the loop: AMD can optimize silicon and model architecture together, the same approach that has helped NVIDIA hold its lead. For companies building on AMD compute, Li's presence as chief scientist could accelerate support for 3D and robotics workloads. Close the deal before year-end, and AMD enters 2027 with a credible answer to why its stack is not just cheaper but technically distinct.
ARTIFICIAL INTELLIGENCE
Anthropic's New Sonnet Costs 30% Less Per Task, Same Price Per Token
Anthropic released Claude Sonnet 5.5 on Sept. 28 at the same price as its predecessor: $2 per million input tokens and $10 per million output tokens. The 30% cost reduction comes from the model completing work in fewer steps and fewer tool calls. On Terminal-Bench 4.0, Sonnet 5.5 scored 70.6% against Sonnet 5's 10.3%, and it beat the more expensive Opus 5.5 on three of five benchmarks.
The efficiency story matters more than the price tag. Balyasny Asset Management ran the same 2,441 financial tasks through both models: Sonnet 5.5 averaged 121,000 tokens per task versus 497,000 for Sonnet 5. That is a cost reduction of roughly 75% at identical per-token pricing. For teams running high-volume agentic workflows, the practical bill drops without a contract negotiation. Anthropic also confirmed the $3 per million input pricing increase scheduled for Sept. 1 was cancelled, locking in current rates permanently for Sonnet 5.
SECURITY & PRIVACY
Okta, AWS, Google, and 9 Others Release AI Agent Security Standards
Okta and 11 co-founders including AWS, CrowdStrike, Google Cloud, and Salesforce launched the Blueprint Alliance on Sept. 22, publishing an open reference architecture for securing AI agents across enterprise environments. The group released six shared principles and a framework built around four questions: where agents are, what they can do, what they are doing, and how to shut them down. Okta also made Agent SSO generally available Aug. 24 at no extra charge within standard SSO plans.
Gartner's numbers frame the problem: the average Fortune 500 will run more than 150,000 agents by 2028, up from fewer than 15 in 2025, but only 13% of organizations say their governance is adequate today. If you are deploying agents in your product or your internal systems, the access control question is not optional. The Blueprint Architecture gives procurement and security teams a vendor-neutral checklist to pressure-test your stack against. The fact that Salesforce, ServiceNow, and AWS all signed onto the same document also makes it easier to require compliance in vendor contracts.
Created by the robots at The SaaS Sentinel