Meta fires 8,000 to fund its $135B AI bet, Google Cloud Next goes agentic tomorrow, and OpenAI's new image model is running in the wild
Meta is cutting a tenth of its workforce to fund one of the largest infrastructure bets in tech history — and reorganizing what's left into AI-focused pods under a new chief AI officer. Google's biggest cloud conference of the year opens tomorrow in Las Vegas with agents as the undeniable headline. And OpenAI's next-generation image model is already live in A/B tests, producing images so realistic people can't tell them apart from photographs.
Meta is cutting 8,000 jobs and reorganizing around AI — the bet is $135 billion
Meta confirmed it will lay off approximately 8,000 employees — roughly 10% of its global workforce — with the first wave scheduled for May 20. The cuts span Reality Labs, the Facebook social division, recruiting, sales, and global operations. A second round is expected in the second half of 2026. This isn't cost-cutting for its own sake: the company has set its capital expenditure guidance at $115–135 billion for the year, nearly double its 2025 spend, directed almost entirely at data centers, GPUs, and AI model infrastructure.
The organizational change running alongside the layoffs is arguably more significant. Meta is restructuring teams into "AI-focused pods" under Alexandr Wang — co-founder of Scale AI, who joined Meta after the company acquired a 49% non-voting stake in his firm for $14.3 billion and appointed him its first-ever Chief AI Officer. The pods report through Meta Superintelligence Labs (MSL), which already shipped Muse Spark on April 8 — a natively multimodal reasoning model positioned as Meta's answer to GPT-5 and Gemini 2.0 across WhatsApp, Instagram, Facebook, and Meta's AI glasses.
thenextweb.com ↗Google Cloud Next opens tomorrow — and the entire conference is built around agents
Google Cloud Next 2026 runs April 22–24 in Las Vegas, and agentic AI is not just a track — it's the organizing principle of the entire event. Google is expected to unveil redesigned infrastructure built for persistent, always-on agents that operate continuously rather than in discrete request-response cycles. Sessions confirmed include enterprise agent deployments at companies like Mars Incorporated, physical AI orchestrators for digital twins, and a particularly notable session on rebuilding legacy systems with Gemini CLI and MCP-powered agents.
On the infrastructure side, Google previewed its TurboQuant algorithm — a quantization approach for large language model inference that significantly reduces memory usage, opening a path to running capable models on consumer devices like smartphones and laptops without cloud roundtrips. That's a direct answer to on-device inference pressure from Apple, Qualcomm, and Microsoft's Copilot+ push. The Marvell chip partnership for inference-optimized TPUs sits in the same strategic frame: Google is building the full stack for the inference era, not just the top layer.
biztechmagazine.com ↗OpenAI's next image model is already in A/B tests — and the results are genuinely unsettling
OpenAI's next-generation image generation model, internally codenamed "maskingtape-alpha" and referred to publicly as GPT-image-2, is live in A/B testing inside ChatGPT. Users in the test group are reporting capabilities that represent a step change from current models: near-perfect text rendering within images, highly realistic UI and screenshot generation, and overall photorealism that makes outputs difficult to distinguish from actual photographs. The model runs on an entirely new architecture — a single-pass generation approach, replacing the previous two-stage inference pipeline — which also makes it faster.
An official launch is expected between late April and mid-May, possibly bundled with a GPT-5.4 update. The model is not yet available via API as of this writing. The strategic timing is notable: OpenAI is actively testing GPT-image-2 as Google prepares for Cloud Next (where Gemini's multimodal image capabilities will likely be featured prominently) and as Adobe's Firefly enterprise push gains ground. OpenAI's image generation revenue has been a meaningful contributor to the company's reported $2 billion in monthly revenue.
dataconomy.com ↗Today's stories share a single underlying dynamic: the AI industry is entering its consolidation phase. Meta is firing people to concentrate resources on AI infrastructure. Google is hosting its biggest enterprise conference with agents as the singular theme. OpenAI is quietly locking in distribution while it can. The land-grab is happening in real time, and everyone is choosing their bets.
The Meta story is the most human one. 8,000 people losing jobs so the company can spend $135B on compute is a stark articulation of what "AI transition" means on the ground. Alexandr Wang running the new AI org is a fascinating choice — Scale AI's entire business was built on human data labeling, which is now being automated away. The irony isn't lost on anyone in the industry. Whether Wang can translate that into coherent frontier research is genuinely uncertain.
The Google Cloud Next signal tomorrow is worth watching. If Google announces robust, production-grade MCP security tooling or addresses the architecture flaw we covered yesterday, that's a meaningful inflection. If they treat MCP as purely an integration surface without acknowledging the security debt — that's a signal too. Las Vegas, 9 AM Pacific.