Microsoft Copilot at 3% Adoption — LinkedIn AI Breaks Out

Microsoft has poured billions into Copilot (its AI assistant built into Word, Excel, and Outlook) to make it the default AI tool for the world's office workers. As of the end of 2025, only 3% of Office 365 users actually use it regularly. Meanwhile, LinkedIn's AI agent (a feature that automates job searches, outreach, and profile optimization) quietly became a breakout product that senior Microsoft executives are now holding up as a model. Same parent company. Wildly different outcomes.

This contrast is not a fluke — it is a signal about where the entire AI industry is heading as April 2026 brings layoffs, billion-dollar chip bets, and a few genuine surprises all in one week.

Why LinkedIn's AI Agent Runs While Copilot Crawls

The numbers are stark. Office 365 Copilot — available to 365 million+ users and priced at $30 per seat per month — was adopted by just 3% of eligible users by the end of 2025, according to internal Microsoft data reported by The Information. That translates to roughly 11 million active users despite an addressable base more than ten times larger.

LinkedIn's AI agent, by contrast, has become what The Information describes as a "surprise hit" and "surprise bright spot" for Microsoft leadership. The agent handles personalized connection requests, summarizes job listings, and drafts cover letters — tasks professionals were already struggling to do manually at scale.

The difference comes down to integration and intent:

• Copilot asks you to change habits — you have to learn to prompt it inside familiar apps (Word, Teams, Outlook) in entirely new ways, which most workers skip
• LinkedIn's agent fits work you already want done — job searching and networking are tasks users actively seek help with, no habit change required
• ROI is impossible to ignore — LinkedIn's agent shows direct, measurable results (more replies, hours saved per week); Copilot's value is diffuse and hard to prove to IT buyers
• Context wins over compute — LinkedIn has 1 billion+ users with clear professional intent; the agent matches that context precisely instead of trying to be everything

Microsoft's own leadership is now studying the LinkedIn playbook to rethink Copilot's roadmap. If you have tried Copilot at work and found it underwhelming, you are in the 97% majority — and the company that built it knows it.

OpenAI's $20 Billion Cerebras Chip Bet Against Nvidia

One layer below the product headlines, a more consequential infrastructure shift is underway. OpenAI has committed over $20 billion to purchase AI server chips from Cerebras Systems (a chip startup based in Sunnyvale, California) over three years — plus an additional $1 billion toward data center development. As part of the deal, OpenAI receives equity warrants (the right to purchase Cerebras shares at a locked price, giving OpenAI a stake in any IPO upside).

This deal carries two implications the industry is still absorbing:

• It is the first major public confirmation that OpenAI is actively diversifying away from Nvidia's H100 and H200 GPUs (graphics processing units originally designed for gaming, now repurposed as the dominant AI training hardware), which currently control over 80% of the AI chip market
• It gives Cerebras the revenue foundation needed to file an IPO (initial public offering — the first time a private company sells shares to public investors) targeting a $35 billion valuation, a 60% premium over its $22 billion private valuation from just two months prior in February 2026

The backdrop makes the timing striking. TSMC (the world's largest contract chip manufacturer, which fabricates chips for Apple, Nvidia, and AMD) just reported Q1 2026 revenue growth of 40.6% — above every analyst projection — and raised its full-year growth target to 30%+. CEO C.C. Wei called AI-related demand "extremely robust." Every dollar flowing toward Cerebras alternatives is a dollar that once ran exclusively through Nvidia's ecosystem, and TSMC benefits from both sides of that equation regardless.

What OpenAI's Cerebras Chip Shift Means for AI Pricing

If Cerebras scales as OpenAI's backing implies, it could be the first real check on Nvidia's GPU pricing power since 2022. Cerebras builds wafer-scale chips (processors the size of an entire silicon wafer rather than a small die) that claim significantly higher memory bandwidth for certain AI workloads. The OpenAI deal effectively funds a real-world proof of concept at scale that no benchmark could replicate.

Meta Layoffs: 8,000 Cuts as AI Infrastructure Costs Rise

Meta Platforms announced what will be the tech industry's largest single-company layoff in two years: 8,000 employees — exactly 10% of its global workforce — will be cut starting May 20, 2026, with additional reductions expected in the second half of the year. Three simultaneous moves make this layoff unusual even by tech's aggressive standards:

• Meta disclosed it paid $2.3 billion to Broadcom in 2025 alone for custom AI chip design work — a rare and candid admission of what building proprietary AI hardware actually costs at scale
• It raised Quest VR (virtual reality) headset prices by $50 to $100 per unit due to global memory chip cost inflation, pushing the Quest 3S to $349.99 (128GB) and $449.99 (256GB)
• AI research and infrastructure teams are almost entirely shielded from cuts; layoffs are concentrated in business development, recruiting, and redundant management layers

The $2.3 billion Broadcom figure is the kind of transparency the industry rarely sees. Most hyperscalers bury custom chip expenditures inside broad capital expenditure disclosures. Meta's number makes clear why AI infrastructure — not just model software — is consuming so much of the industry's operating budget. Building your own AI chip is a nine-figure annual line item before you ship a single product to users.

DeepSeek, Alibaba, and the Split Running Through Global AI

DeepSeek — the Chinese AI startup owned by quantitative hedge fund High-Flyer Capital, known for releasing models that matched GPT-4 performance at a fraction of the training cost — is seeking outside investment for the first time in its history. The reported valuation target: $10 billion or more. DeepSeek had previously turned down multiple funding approaches; reversing course signals that even the most compute-efficient AI builders are facing rising costs that outside capital can no longer offset internally.

Meanwhile, Alibaba is quietly retreating from its open-source commitments. The company is becoming more selective about releasing Qwen (its family of open-weight AI models — models whose internal parameters are made publicly downloadable for anyone to run or fine-tune) to the public, pivoting toward proprietary control. This mirrors the transition U.S. labs made roughly 18 to 24 months earlier: open-source for community building and talent attraction, proprietary for monetization and competitive moat.

Google, for its part, is negotiating a classified AI deployment contract with the Pentagon to run Gemini models in military settings. This represents a significant policy reversal for a company whose engineers famously forced leadership to exit Project Maven (a 2018 military AI contract that sparked the first major internal employee revolt in Big Tech history). Eight years of competitive pressure from defense-focused AI startups apparently shifted the internal calculus.

Five AI Automation Moves to Make Right Now

The macro figures this week — $20 billion chip bets, 8,000 layoffs, $35 billion IPO targets — are attention-grabbing, but the ground-level takeaway is practical. The AI tools that win are not the ones with the most compute behind them. They are the ones that fit a task someone was already frustrated trying to do. Here is where to direct your attention:

• Benchmark your own AI tool adoption honestly. The 3% Copilot figure is now the industry's most useful reference point. If your team's internal AI adoption is near that level, the problem is almost certainly product-task fit, not user willingness
• Try Anthropic's Claude Design, launched this week — an AI tool for building prototypes, presentations, and product one-pagers directly inside Claude. Relevant for designers and product managers evaluating Figma AI alternatives
• Watch the Cerebras IPO filing for the first public glimpse at what AI chip unit economics look like outside Nvidia's pricing umbrella — it will set a public benchmark for the entire sector
• Track DeepSeek's funding round closely. A $10 billion valuation for a Chinese AI lab would be the first major U.S. venture capital test of cross-border AI investment appetite under current export controls
• If you are in a non-AI role at a major tech company, Meta's May 20 start date and H2 cut forecast is the clearest public signal yet that the industry's talent-sparing phase is over for everyone outside the AI core teams

Related Content — Get Started | Guides | More News