Five days ago, Google committed up to $40 billion to Anthropic. The same 24 hours, Meta and Microsoft announced more than 20,000 job cuts and pointed at AI as the reason. DeepSeek shipped an open-weights model that runs at one-sixteenth the price of GPT-5.5. AbbVie picked Durham for a $1.4 billion campus that will run on AI from day one.
If you've been waiting for the moment AI stops being a story about demos and starts being a story about money, jobs, and concrete, this is it.
Let's get into it.
$10B now at a $350B valuation. Another $30B if Anthropic hits its targets. Plus 5 gigawatts of TPU capacity. The single largest AI investment ever recorded. Suggested image: Google + Anthropic logos with stylized data center photo.
Two announcements landed last Thursday and Friday. They look unrelated. They are not.
First: Google confirmed a deal to put up to $40 billion into Anthropic. Ten billion lands now at a $350 billion valuation. The other thirty is contingent on Anthropic hitting performance milestones. The deal also dedicates 5 gigawatts of TPU capacity to Anthropic, on top of the multi-gigawatt Broadcom partnership announced earlier this month. For context, Amazon's prior $8 billion bet on Anthropic is now worth roughly $13.8 billion on paper. Anthropic's annualized revenue went from $1 billion at the end of 2024, to $9 billion at the end of 2025, to $30 billion as of early April 2026. An IPO is reportedly being targeted as soon as October.
Second: Meta announced 8,000 layoffs, about 10% of its workforce. Microsoft opened its first-ever voluntary buyout program in 51 years, eligible for roughly 8,750 U.S. employees. CNBC put the combined potential cuts at more than 20,000. Meta's chief people officer wrote in a leaked internal memo that the cuts would help "offset the other investments we're making." Those other investments are $115 to $135 billion in 2026 capital expenditure, almost double last year's spend, almost all of it on data centers, GPUs, and the infrastructure for Llama and the new Meta Superintelligence Labs.
Add Alphabet, Amazon, Meta, and Microsoft together and capex hits roughly $700 billion in 2026 alone. Total tech layoffs this year just crossed 92,000.
My take: Three things to pull apart here.
First, capital is being moved from payroll to compute, and the companies have stopped pretending otherwise. For two years, the layoff press releases used soft language about right-sizing and operational efficiency. This week, Meta wrote it down on paper: we are cutting people to pay for the GPUs. Oracle said it earlier this month. Microsoft built its buyout to select against employees least aligned with its AI roadmap. Three of the most valuable companies in the world are converting headcount into hardware in real time, and they're being explicit about it.
Second, the strategic weirdness of Google's deal. Google sells Gemini. Anthropic sells Claude. They compete head-to-head on enterprise AI. And Google just became Anthropic's biggest single backer while running a competitor product in the same building. That's not a contradiction. It's a strategy. Google doesn't need to win the model race outright. It needs to win the infrastructure race, and Anthropic burning $40 billion worth of TPU capacity is a faster path there than any Gemini sales cycle.
Third, watch the cost curve. Both flagship API models from OpenAI and Anthropic cost $5 per million input tokens. That was a tacit agreement that the frontier costs what the frontier costs. DeepSeek V4 just blew that up. V4-Flash is the cheapest small model on the market at $0.14 per million in. V4-Pro is the cheapest larger frontier model at $1.74. They're maybe three to six months behind GPT-5.5 on raw capability, and they're 1/16th the price, with weights you can download. For most production workloads, that math is going to start mattering.
Worth a small caveat on all of this: only about 9% of companies say AI has actually replaced roles entirely. There's clearly some amount of "AI" being used as cover for cost-cutting that would have happened anyway. But the direction is real. Per Menlo Ventures' enterprise survey, Anthropic now captures 40% of enterprise LLM spend, up from 12% in 2023. OpenAI fell from 50% to 27% over the same period. The money is moving fast, both into the labs and out of the workforce.
Two product drops worth actually trying, not just bookmarking.
Claude Design — Anthropic Labs released this on April 17, powered by Opus 4.7. You describe a pitch deck, landing page, or prototype in plain language and get a working draft back. Refinement happens in conversation: comment inline, edit text directly, or pull on the small sliders Claude generates on the fly to tweak spacing and color. It can read your codebase or design files and apply your team's design system automatically to every project. Available in research preview for Pro, Max, Team, and Enterprise. Figma stock dropped 7% the day it shipped. Try it →
ChatGPT Images 2.0 (gpt-image-2) — Released April 21. The headline feature is finally fixing the thing AI image models have always been bad at: text. Menus look printer-ready. Non-Latin scripts (Japanese, Korean, Hindi, Bengali in particular) got a serious upgrade. The model can search the web, plan layouts, generate up to 8 consistent panels from one prompt, and verify outputs before finalizing. Up to 2K resolution. It's OpenAI's first image model with built-in reasoning, and it took the #1 spot on the Image Arena leaderboard within 12 hours. Read →
Three years ago, the AI conversation was about capability. Two years ago, it was about funding. Last year, it was about deployment. This week, it became about substitution.
The Meta memo is the part I keep rereading. The line about offsetting investments isn't a euphemism. It's a sentence that says, in plain English, we are trading 8,000 jobs for compute capacity. Microsoft built a buyout that filters specifically against the workforce least likely to thrive under their AI plans. Oracle is cutting people to fund data centers. The companies with the most to lose from being honest about this are the ones being most honest about it.
If you work in tech and you've been telling yourself the AI conversation doesn't apply to your role yet, this is the week to revisit that. Not because the sky is falling. The 9% number is real and the macro pressures are real. But the direction is locked in. The companies writing the biggest checks for AI infrastructure are the same ones writing the biggest pink-slip lists, and they have stopped trying to make those two facts sound unrelated.
The good news, such as it is: AI is also the cheapest force-multiplier anyone has ever had access to. The same tools the hyperscalers are spending hundreds of billions to build are sitting in your browser for $20 a month, or in DeepSeek's API for fourteen cents per million tokens. The question worth asking this week isn't whether AI will affect your work. It's whether you're using it the way the people deploying it at scale are using it. As infrastructure, not a curiosity.
Last Wednesday, Governor Stein announced AbbVie is putting $1.4 billion into a 185-acre pharmaceutical manufacturing campus in Durham, near Research Triangle Park. It's the company's largest single capital investment ever and its first major project in North Carolina. 734 jobs at an average salary of $118,000. Engineers, scientists, lab techs, manufacturing operators. Construction starts this year, completion by end of 2028.
The detail that matters for this newsletter: AbbVie was specific that the campus will integrate "advanced manufacturing and laboratory technologies with artificial intelligence" from the start. This isn't a legacy plant being retrofitted. It's being designed AI-native, for immunology, neuroscience, and oncology production. AbbVie's portfolio includes Humira, Botox, and Rinvoq. We're talking about AI in the loop for drugs that show up in millions of medicine cabinets.
Why this matters locally: The Triangle's AI story has been mostly software (Pryon, GTM Buddy, Swarm, the All Things AI scene). The biotech story has been mostly traditional pharma. AbbVie just picked Durham specifically to merge the two and cited the Triangle's research base and workforce as the reason. That's the Triangle's actual edge. Not competing with the Bay Area on pure AI talent, but being the place where AI meets biotech, manufacturing, and applied research. AbbVie's $1.4B vote of confidence makes that thesis a lot easier to sell.
deepseek-chat and deepseek-reasoner endpointsLast Wednesday of April. See you next week.
Daniel
BullCity AI · Durham, NC
P.S. If you've used AI to materially change your own job in the last 90 days (automated something, replaced a tool, picked up a new responsibility), hit reply and tell me what you did. I want to write about what's actually working at the individual-contributor level, not just what hyperscalers are doing with their billions.
P.P.S. If you got hit by the layoff wave at Meta, Microsoft, or anywhere else this week, I'm genuinely sorry. The Triangle is hiring. AbbVie's bringing 734 jobs to Durham. Pryon, Red Hat, IBM, and a dozen smaller shops are looking. Reply if you want intros.