Wave Theory Notes: Amazon Goes All-In on Anthropic (Apr 20-26)
This was the week the AI infrastructure arms race got more zeroes added to it. Amazon dropped $25 billion more into Anthropic, OpenAI shipped GPT-5.5, and DeepSeek returned with a model that once again threatens to rewrite the cost curve. A lot happened in five days.
Hey Wave Riders!
Welcome back. This week felt like confirmation more than surprise: the big players are consolidating their bets, the models keep getting better, and the ground keeps shifting under everyone building on top of them. Let's get into it.
Amazon and Anthropic: $25 Billion, Ten-Year Infrastructure Lock-In
The headline number is $25 billion, but the more telling detail is the $100 billion Anthropic committed to spend on AWS over the next decade. This is not just an investment. It is a structural dependency between two organizations, built to last.
Amazon's stake comes in at $5 billion now, with up to $20 billion more tied to commercial milestones. All of this is priced at Anthropic's $380 billion valuation. The trigger, according to Anthropic, was real: surging demand for Claude across enterprise and consumer use cases was creating "inevitable strain" on infrastructure and hitting reliability. The company needed capacity, and Amazon provided it.
What makes this deal different from prior investments is the chip commitment. Anthropic will bring nearly 1 gigawatt of Trainium2 and Trainium3 capacity online by year-end. That is not a vague cloud partnership. That is manufacturing scale being redirected toward a single AI company's compute needs.
The broader context matters: Amazon made a $50 billion deal with OpenAI just two months ago. The cloud giants are not picking winners. They are hedging with hundred-billion dollar bets on multiple horses. The winner of that arrangement, so far, is whoever controls the GPU supply.
GPT-5.5 Is Here, and OpenAI Wants ChatGPT to Be Your Everything App
OpenAI released GPT-5.5 on April 23, less than two months after GPT-5.4. The release cadence alone tells you something: the labs are not waiting for a clean generational leap anymore. They are shipping continuously.
The framing OpenAI used for 5.5 is worth paying attention to. They called it their most intuitive model yet, one that understands what users are trying to do faster and carries more of the work itself. That is the language of agentic AI: less question and answer, more autonomous task completion. It excels at coding, research, data analysis, document creation, and moving across tools until a job is done.
The model is rolling out to paid subscribers now, with GPT-5.5 Thinking and GPT-5.5 Pro available from day one. Free users are not getting access yet. That tiering is deliberate: OpenAI is using its best capabilities to drive paid conversions, which feeds the revenue model that funds the next model.
The "superapp" framing is intentional. OpenAI is not positioning ChatGPT as a chatbot anymore. It is positioning it as the interface you use to get work done, period. Whether that plays out depends on whether users actually trust it to act on their behalf, not just answer questions.
DeepSeek V4: The Open-Source Pressure Keeps Building
DeepSeek dropped V4 on April 24, exactly a year after the original model that sent shockwaves through Silicon Valley. The timing was not accidental.
V4 comes in two versions. The Pro model has 1.6 trillion parameters and a 1 million-token context window, meaning you can feed it an entire large codebase or a year's worth of documents as a single prompt. The Flash version prioritizes speed. Both are open source, both published on Hugging Face, and both priced well below what frontier US models cost to access.
The technical standout is what DeepSeek calls Hybrid Attention Architecture, which improves how the model maintains context across long conversations. For anyone building applications that require multi-step reasoning or persistent memory, that is a real upgrade.
The bigger question is what this does to pricing dynamics across the market. Every time DeepSeek ships a competitive open-source model, it puts pressure on OpenAI, Anthropic, and Google to justify their API pricing. That pressure benefits builders and enterprise buyers. It complicates the unit economics for the closed-model labs.
One year on from the original DeepSeek shock, the pattern holds: Chinese labs are producing frontier-quality models at lower cost, and they are releasing them openly. That is not a one-time disruption. It is a structural feature of this market now.
Quick Hits
Meta announced cuts of around 8,000 roles as part of an efficiency drive while simultaneously increasing capital expenditure on AI infrastructure, a workforce reduction and infrastructure ramp that are not in tension but the same strategy. SpaceX is reportedly repositioning itself as an "AI-first" organization ahead of a potential IPO, with AI framed as central to satellite network operations and broader company strategy. Between Amazon's $50 billion deal with OpenAI in February and this week's $25 billion deal with Anthropic, the two largest US cloud providers have committed over $75 billion to the two leading frontier AI labs in the space of roughly two months. The pace of model releases in 2026 is running ahead of any prior year, with GPT-5.4 in early March, GPT-5.5 in late April, and DeepSeek V4 following within a day, putting the interval between significant releases in weeks rather than quarters. US state-level AI legislation continues to move faster than federal action, with several states advancing disclosure and transparency requirements for AI systems used in hiring, lending, and public services.
This Week's Template: The 5-Part AI Brief
We're cycling back to the foundational template this week, because every other technique in the rotation builds on top of it. If you only adopt one habit from these newsletters, make it this one.
The 5-Part AI Brief Template
Most people type a question and hit send. Professionals write a brief. Here is the exact structure to paste at the top of any complex prompt:
ROLE
Who is the AI for this task?
Example: "You are a senior copywriter who specializes in B2B SaaS landing pages."
AUDIENCE
Who is reading or using the output?
Example: "The audience is mid-level marketing managers at companies with 50-500 employees. They are analytical and skeptical of hype."
TONE
How should it sound?
Example: "Direct and confident. No filler phrases. Short paragraphs. No bullet points unless they genuinely aid clarity."
CONSTRAINTS
What to avoid. What to include.
Example: "Avoid: generic advice, passive voice, sentences over 20 words. Include: one concrete example per section, a clear next action at the end."
SUCCESS CONDITION
How will you know the output is right?
Example: "This is ready when I could hand it to a colleague and they would immediately understand what to do and why."
Paste that structure at the top of any complex prompt and you have done 80% of the work before you type a single word of your actual request. The people getting the most out of AI are not better at prompting. They are better at briefing. That is a professional skill. It is learnable.
Wrapping the Waves
The thread running through this week is consolidation at scale. Amazon locked in Anthropic with a ten-figure infrastructure deal. OpenAI pushed its release cadence even faster. DeepSeek reminded the market that open-source competition is not slowing down. What ties these together is infrastructure: whoever controls compute, distribution, and access is in the strongest position, and those positions are becoming harder to challenge.
Next week, watch for any response from Google and Microsoft. Both have been quieter than expected while Amazon made two massive moves in the span of two months. The silence may not last.
Find me on YouTube and follow @WaveTheoryAI on X for daily AI coverage. See you next Sunday.
Wave Theory Notes publishes every Sunday. If this was forwarded to you, subscribe at thewavetheory.substack.com.