TechLook
The AI infrastructure buildout isn't slowing down — it's accelerating into territory that even the most bullish forecasts didn't predict for this decade. Two stories today capture this tension: Anthropic is reportedly raising at a valuation that puts it above OpenAI's, while OpenAI itself just hit its 2029 compute goal three years early. Both things can't be the whole story at once.
Anthropic's Valuation Doubles to $900B in One Round
Anthropic is weighing a $40–50 billion funding round that would value the company at $850–900 billion — more than doubling its $380 billion valuation from February, and landing it near or above OpenAI's $852 billion post-money mark.
Here's everything you need to know:
- Board decision expected in May 2026
- Would make Anthropic one of the most valuable private companies in history
- Annual revenue run rate jumped from ~$9B (end of 2025) to $30B+ — some sources put it closer to $40B
- Claude Code and Cowork are the primary revenue drivers — coding agents have become the fastest path from model capability to enterprise revenue
- The White House opposed Anthropic's plan to expand Mythos access to ~120 firms (up from ~50), citing cyber-risk and compute-capacity concerns
The political complication is real but secondary. What matters more is what the valuation says about where the money is. Investors are betting that AI agents — specifically agents that do real work inside enterprise environments — represent durable revenue, not a hype cycle. That's a bet on the Software 3.0 world where the product isn't a model, it's a workflow.
For builders: this is another green light. If you're building agents that replace or augment professional workflows — coding, legal, finance, operations — the market validation is unambiguous. The question isn't whether agents will be big. It's whether your specific bet on what they replace is the right one.
Software 3.0: The Paradigm Shift Builders Actually Need to Understand
Andrej Karpathy put a framework to what many builders have been living through: the software stack is being reorganized around a new unit of computation — the agent, not the application.
Here's everything you need to know:
- Software 1.0: Hand-written code — humans wrote instructions for machines
- Software 2.0: Neural networks — humans wrote data, machines learned weights
- Software 3.0: Prompts + context + agents — humans write intent, agents execute in your environment
The concrete example is OpenClaw. Instead of shipping a complex install script, it gives users text instructions to hand to an agent, which reads the environment, debugs, and completes the setup. The "program" is a piece of text. The interpreter is the LLM. The runtime is an agent with tools.
The takeaway isn't that apps disappear — it's that the interface between human intent and machine execution shifts from commands to conversations, and the machine takes on more of the implementation burden. If you're building products today, the question is whether your stack assumes human-in-the-loop at every step, or whether you're designing for a world where the agent handles the execution.
Manifest OS Wants to Replace the Billable Hour With an AI Operating System
Manifest OS raised $60 million in Series A funding at a $750 million valuation to build AI-native law firms under the Manifest Law brand — starting with business immigration, going after the legal services market that's been resistant to copilot-style tools.
Here's everything you need to know:
- Not building another legal copilot — owns the entire legal service layer
- Fixed-fee, outcome-based, AI-powered law firms (vs. traditional hourly billing)
- Legal rates rose 7.4% in 2025 — the billable hour is the target
- Combines AI platform + centralized back-office + licensed lawyer review where the law requires it
The contrast with Harvey is sharp. Harvey makes existing lawyers faster. Manifest asks a different question: what if the customer never needed the lawyer-shaped workflow in the first place? If it works, the future law firm is an AI operating system with lawyers attached where statutes still demand human oversight.
For builders: vertical AI services are a legitimate category, not just a vertical SaaS variant. Manifest is making the bet that owning the end-to-end workflow — not just providing tools to humans who do the workflow — is the winning model in regulated industries.
OpenAI Hit Its 2029 Infrastructure Goal in 2026
OpenAI has already surpassed its 10-gigawatt U.S. AI compute capacity goal — three years ahead of the original timeline. The company added 3 gigawatts in the last quarter alone.
Here's everything you need to know:
- Original Stargate target: 10 GW by 2029
- Achieved: 2026 — years ahead of schedule
- 3 GW added in just the last 3 months
- Reflects the capital intensity of the infrastructure buildout driving AI capability
The infrastructure story is real. But so is the tension it creates. Companies are building compute capacity at a pace that assumes enterprise adoption will match — but the revenue proof points are still concentrated in a narrow range of use cases (coding agents, cloud infrastructure, advertising). If the demand doesn't keep pace, the capex cycle creates stranded assets. If demand does keep pace, the companies that locked in capacity early have a meaningful moat.
For builders: the inference layer is getting cheaper and more available faster than most models predicted a year ago. That's an enabler — but it also means the infrastructure advantage is becoming commoditized faster than anyone expected.
⚡ Quick Hits
-
Apple: Mac Q2 revenue hit $8.4B — up 6% YoY, beating expectations in the low $8B range. Mac mini became top-selling desktop in China during the OpenClaw agent rush. Cook says supply could take months to catch up. Apple's AI story gets framed through iPhone, but Mac is the more interesting developer signal right now.
-
NVIDIA: Launched Nemotron 3 Nano Omni — an open multimodal model optimized for AI agents handling vision, audio, and document-aware tasks. Part of NVIDIA's push to own the inference stack for agentic workflows.
-
Cursor: Running autonomous agents on codebase security review — scheduled scans with results posted directly to Slack. The autonomous code editing agent is becoming a security surface that needs its own security tooling.
-
Meta: Raising up to $25B in bonds for AI infrastructure. Lost ~20M daily users across its family of apps (Iran war disruptions, WhatsApp ban in Russia) — but revenue is growing at its fastest pace since 2021. The user base is shrinking; the revenue per user is growing.
-
Visa: Launched Agentic Ready Program across 50+ banks in Asia Pacific, preparing financial infrastructure for AI agent-initiated payments. The agent-to-agent economy needs settlement infrastructure, and Visa is moving early.
-
Google Translate: Turned 20 with 1 billion monthly users. Free + instant + everywhere beats prettier AI translation products at global scale.
Techlook — AI & tech signal for founders and builders.