Welcome, Startup enthusiasts!

Good Morning. It’s July 28, 2026.

🚀 The Startup Khan Newsletter

July 28, 2026 — AI’s Real Currency Is Compute

🧠 Top Story: Fish Audio Raises $52M to Build More Controllable AI Voices

Palo Alto-based Fish Audio raised a $52 million seed round led by Coreline Ventures and Capital Today, with participation from 359 Capital, HF0, Play Time, Parable, and other investors. (TechCrunch)

The startup develops AI voice-generation models for creators and enterprises. Its technology includes more than 15,000 natural-language controls, allowing users to adjust how generated speech sounds and behaves. Fish Audio says its open-source and hosted products have attracted more than 8 million users and reached $21 million in annual recurring revenue. (TechCrunch)

Founder Shijia Liao, a former Nvidia researcher, initially trained the company’s voice model using a single GPU and released it as an open-source project. The Fish Speech repository has since accumulated more than 31,000 GitHub stars, while companies including HeyGen and Sanas use Fish Audio’s enterprise platform. (TechCrunch)

Why founders should care

The voice-AI market is moving beyond novelty.

Creative teams want emotion and expression. Enterprises want consistency, control, and reliability. The companies that win will not simply produce realistic voices—they will give customers precise control over the final output.

👉 Founder takeaway: Generative products become enterprise products when users can reliably direct the result.

💰 Other Startup Moves

1️⃣ Recursive Superintelligence signs a $410M AWS compute deal

Recursive Superintelligence signed a multiyear $410 million computing agreement with Amazon Web Services.

The startup emerged from stealth in May with $650 million in funding and is developing open-ended, self-improving AI systems. Founder and CEO Richard Socher said the agreement may be one of the company’s smallest compute deals over the next several years. (TechCrunch)

The message for founders is blunt:

👉 In frontier AI, fundraising is increasingly a mechanism for purchasing compute.

2️⃣ Granola brings AI meeting notes to the Apple Watch

AI note-taking startup Granola launched an Apple Watch app that lets users record and transcribe in-person conversations without pulling out a phone.

The app can be placed on a watch face for quick access, surfaces meeting reminders, and connects with Granola’s iOS application. The company said internal testing showed that a significant portion of mobile usage shifted from the iPhone to the watch. (TechCrunch)

This is a smart distribution lesson: instead of creating another dedicated recording device, Granola moved onto hardware customers already wear.

3️⃣ London’s robotaxi market gets more crowded

Baidu began testing autonomous vehicles in London through partnerships with Lyft and Freenow. It joins Waymo, Uber, and Wayve as companies prepare for commercial robotaxi deployments in the city. (TechCrunch)

For mobility startups, technical performance is only one challenge. Regulation, public trust, local partnerships, insurance, and fleet operations will determine who reaches commercial scale.

🧰 Tool of the Day: E2B

E2B provides secure, isolated cloud environments where AI agents can execute code, use a terminal, access files, and complete tool-driven tasks.

Its JavaScript, TypeScript, and Python SDKs let developers create and manage temporary sandboxes without giving an autonomous agent access to the company’s production infrastructure. (E2B)

Useful founder applications include:

Running AI-generated code safely
Testing coding agents
Processing unfamiliar files
Automating data analysis
Isolating risky or unpredictable workflows

👉 Best use: Give agents a disposable workspace before giving them access to valuable systems.

📊 Trend Check: The Power Grid Is Becoming an AI Constraint

PJM Interconnection, operator of the largest U.S. electrical grid, plans to temporarily curtail electricity to data centers and other major users during shortages beginning in June 2027.

The policy will apply to facilities using at least 50 megawatts. Data-center electricity use is projected to be four times higher in 2035 than it is today. (TechCrunch)

This creates opportunities in:

On-site energy generation
Cooling and efficiency
Workload scheduling
Battery storage
Regional compute markets

The next AI bottleneck may not be chips.

It may be electricity.

🌟 Founder Spotlight: Shijia Liao

Shijia Liao built Fish Audio after becoming frustrated with the limited expressiveness of existing synthetic voices.

He started narrowly: one model, one GPU, and an open-source release. That project created a developer community, product feedback, distribution, and eventually enterprise demand. (TechCrunch)

The playbook:

👉 Use open source to prove technical credibility before scaling the commercial layer.

⚡ Quick Win: Calculate Your Compute-to-Revenue Ratio

Measure:

Monthly AI infrastructure spending ÷ monthly recurring revenue

Then track it every month.

If infrastructure costs grow faster than revenue, test smaller models, caching, batch processing, routing, or usage limits before the problem compounds.

🧩 Final Thought

Today’s startup signals reveal the infrastructure beneath the AI boom:

Fish Audio is making generated voices controllable. Recursive is converting capital into compute. Granola is expanding AI into wearable workflows. E2B is giving agents safer places to operate.

Intelligence gets the headlines.

Compute, control, distribution, and energy determine whether it becomes a business.

— The Startup Khan

Your 5-minute startup edge — funding news, tools, and founder playbooks.

Your 5-minute startup edge — funding news, robotics tools and founder playbooks.