2026.04.25 — Big Tech realignments and the agent era in full swing

Google is investing up to $40 billion in Anthropic, and Meta is cutting 8,000 jobs in the name of AI efficiency. China's DeepSeek has closed the frontier-model gap once again with V4. Today marks an inflection point where the race for AI infrastructure and the commercialization of agents accelerate at the same time.

---

1. Google to invest up to $40 billion in Anthropic — the compute race intensifies

• Source: TechCrunch / CNBC | 2026.04.24 | TechCrunch link
• What happened: Google and Anthropic announced that Google has committed up to $40 billion in investment. $10 billion goes in immediately, with the remaining $30 billion released in stages as performance milestones are hit. The deal includes 5 gigawatts of Google Cloud TPU compute over the next five years. Anthropic is currently valued at around $350–380 billion, and some investors are reportedly floating valuations above $800 billion.
• Why it matters: Even though Google competes with Anthropic in the model market, it deepened the relationship as an infrastructure supplier. It signals that the frontier AI race has shifted from 'model performance' to 'securing compute'.

2. DeepSeek V4 launches — a 1.6-trillion-parameter open-weight model takes on the frontier

• Source: TechCrunch / Bloomberg | 2026.04.24 | TechCrunch link
• What happened: Chinese AI lab DeepSeek released two preview versions, V4 Flash and V4 Pro. The Pro model, at 1.6 trillion total parameters, is said to be the largest open-weight model to date, using an MoE (mixture-of-experts) design with 49 billion active parameters. Both models support a 1-million-token context window, and DeepSeek says they posted top-tier results on coding benchmarks. A new technique it calls a 'hybrid attention architecture' reportedly improves context retention in long conversations.
• Why it matters: With the US–China AI gap narrowing again, open-weight models have pulled close to commercial-model quality. Companies now have far more options for self-hosting and fine-tuning.

3. Meta announces 8,000 layoffs — aimed at offsetting AI investment

• Source: Variety / Axios | 2026.04.23 | Variety link
• What happened: Meta announced it will lay off 8,000 people, about 10% of its workforce, effective May 20, and freeze a further 6,000 open positions. Meta has guided 2026 capital expenditure to $115–135 billion, a sharp increase from $72 billion in 2025. The stated rationale for the cuts is to offset AI infrastructure spending, including Meta Superintelligence Labs. Microsoft is also reported to have offered voluntary retirement to roughly 8,750 employees the same week.
• Why it matters: The pattern of expanded AI investment turning into workforce restructuring at Big Tech is now unmistakable. Layoffs justified by AI efficiency are becoming the industry standard.

4. Korea's chip 'twin giants' SK hynix and Samsung post record Q1 results

• Source: Hankook Ilbo / ZDNet Korea | 2026.04.24 | Related ZDNet Korea coverage
• What happened: SK hynix and Samsung Electronics each reportedly posted their best-ever quarterly results in Q1 2026. Analysts put SK hynix's operating margin in the 70% range, above Nvidia and TSMC. The spread of AI accelerators has sent DRAM and NAND prices soaring, and the growth of agent-based autonomous AI workloads is driving simultaneous demand for HBM, server DRAM, and eSSDs. Intel also suggested that as agentic AI spreads, CPUs could overtake GPUs as a share of the data center.
• Why it matters: Entering the AI agent era, memory chip demand has moved into a new supercycle. The Korean semiconductor industry's leadership is becoming even more entrenched.

5. Adobe rebrands Experience Cloud as 'CX Enterprise' — a shift to an agent-centric platform

• Source: MarketingProfs (Adobe Summit roundup) | 2026.04.24 | Related coverage
• What happened: Adobe announced it is rebranding Experience Cloud as 'CX Enterprise' and turning it into an AI-first platform built around agent-driven workflows. Persistent AI agents dubbed 'Coworkers' coordinate work across systems and operate continuously toward business goals. The GenStudio lineup is expanding, data collaboration features are being strengthened, and integrations with major AI ecosystems are in the works.
• Why it matters: In enterprise DAM and content automation, including AEM (Adobe Experience Manager), an agent-first architecture is becoming the standard. Governance, data quality, and ROI measurement remain the main barriers to adoption.

6. OpenAI launches 'workspace agents' for ChatGPT Business

• Source: MarketingProfs (OpenAI roundup) | 2026.04.24 | Related coverage
• What happened: OpenAI officially launched workspace agents for ChatGPT for Business, Enterprise, and education users. Teams can build and share their own AI agents that carry out tasks across tools such as Slack and Gmail. With context gathering, workflow tracking, approval requests, and the ability to learn and improve over time, they are seen as the evolution of custom GPTs.
• Why it matters: The shift from assistive AI to agents that 'actually get the work done' is now in full swing, setting up direct competition with Anthropic's Claude agents and Google's Workspace agents.

---

Today's insight

• Compute has become the new currency. The $40 billion Google–Anthropic deal, Meta's $115 billion capex, Amazon's additional $5 billion investment — at bottom, all of this money goes to securing GPUs, TPUs, and power. This dependence on infrastructure is making the web of cooperation and dependence between Big Tech and AI labs ever more tangled.
• Agent commercialization is the keyword of 2026. It is no coincidence that OpenAI, Adobe, and Google announced agent platforms in the same week. The shift from 'AI that talks' to 'AI that works' has entered a genuine competitive phase in the enterprise market.
• The flip side of AI efficiency is workforce restructuring. Meta, Microsoft, Amazon, and Salesforce are all cutting staff on the grounds of expanding AI investment. This is the moment the first concrete cases of AI adoption directly hitting jobs are appearing.