🌟 Editor's Note: Recapping the AI landscape from 03/31/26 - 04/06/26.
🎇✅ Welcoming Thoughts
Welcome to the 37th edition of KPAI Weekly.
What’s included: company moves, a weekly winner, AI industry impacts, practical use cases, and more.
Anthropic usage is still very limited right now.
I’ve started using Sonnet instead of Opus in Claude, which makes a huge difference in usage alotted, but hurts in terms of brainpower.
The more money I spend on Claude extra usage, the more I align with the "AI offers intelligence as a commodity” philosophy.
Google released a new open-source model I want to try.
Might be time to get a Mac Mini to run it.
Pretty interesting Impact Industries this week.
Lots of movement on AI for Biology across the space.
Let’s get started—plenty to cover this week.
👑 This Week’s Winner: OpenAI // ChatGPT
OpenAI wins the week headlined by a $122B fundraise. The largest private fundraise in history cements OpenAI atop the week. Comes with a media acquisition, a leadership shakeup, and a policy blueprint for the AI era. Here's the recap:
$122B Funding Round: OpenAI closed the largest private fundraise in history at an $852B post-money valuation, anchored by Amazon ($50B), NVIDIA ($30B), and SoftBank ($30B). 900M+ weekly active users, $2B/month in revenue, 50M+ subscribers. I have to imagine this is the last raise before IPO.
TBPN Acquisition: OpenAI acquired Silicon Valley's daily live tech show TBPN for a reported low hundreds of millions. ~70K daily viewers, $5M in 2025 revenue. Editorial independence preserved under the strategy org. Not sure the spend is justified but pretty interesting. I’m a fan of TBPN so congrats to those guys.
Leadership Reshuffle: Applications CEO Fidji Simo went on medical leave, with Greg Brockman stepping in on product. COO Brad Lightcap moved to special projects. CMO Kate Rouch stepping down for cancer recovery.
OpenAI also published a 13-page policy blueprint called the "Superintelligence New Deal," proposing a public AI wealth fund, robot taxes, and a 32-hour workweek pilot at full pay. ChatGPT also launched hands-free voice mode in Apple CarPlay for all plan tiers.
From Top to Bottom: Open AI, Google Gemini, xAI, Meta AI, Anthropic, NVIDIA.
⬇️ The Rest of the Field
Who’s moving, who’s stalling, and who’s climbing: Ordered by production this week.
🟣 Google // Gemini
Broadcom Partner Extension: Google and Broadcom extended agreements for custom chip development and AI infrastructure through 2031. Anthropic accesses 3.5 gigawatts of next-gen compute through Broadcom starting in 2027. Google increasing revenue from its chips. Strong pair with Gemini.
Open Model Launch: Google released a new family of open, freely usable AI models in four sizes, with long context and 140+ language support. The largest ranks 3rd among all open models globally. This was a very popular launch. I may try it out for myself. More advanced and takes up less space than other open source models on the market.
IOS Dictation App: Google quietly launched a free offline-first iOS dictation app powered by its Gemma models. It strips filler words, polishes your text, and imports personal vocabulary from Gmail — no subscription required. Cool.
⚪️ NVIDIA
Marvell $2B Investment: NVIDIA invested $2B in Marvell, launching a platform letting customers mix NVIDIA and third-party chips when building AI infrastructure. Third straight ~$2B chip ecosystem investment this quarter.
Iran Threat: Iran's IRGC named NVIDIA among 18 U.S. tech companies as "legitimate targets" and threatened the $30B Stargate Abu Dhabi data center. Confirmed strikes already hit AWS Bahrain and Oracle Dubai. Idiots.
Robotics Week: NVIDIA kicked off National Robotics Week showcasing its robot training platforms. Announced a Physical AI blueprint with Microsoft Azure, Uber, and Teradyne Robotics. Cool. NVIDIA will play a big role in the move into Physical AI.
🔴 xAI // Grok
SpaceX IPO Filing: SpaceX filed confidentially with the SEC targeting a $2T+ valuation and a June listing that could raise up to $75B. Combined entity includes SpaceX, Starlink, xAI, and X. This will be a big one. Should provide lots of capital for whichever direction they want to go next.
Banks Forced to Buy Grok: Musk required banks, law firms, and auditors on the SpaceX IPO to purchase Grok enterprise subscriptions. Smart business move IMO.
Grok Imagine 2.0 Delayed: Musk announced Grok Imagine 2.0 needs "a few more weeks of training," pushing the release back. In the meantime, xAI launched Speed and Quality generation modes. Video could be an interesting niche for xAI. Shouldn’t be a priority though.
🔵 Meta // Meta AI
Open-Sourcing New Models: Meta's first models under Alexandr Wang are coming soon with open-source versions planned. Largest models will stay proprietary. I like that they’re moving back to open source. Just not sure how good the models will actually be.
Data Breach Pause: Meta indefinitely paused work with AI data startup Mercor after a cyberattack exposed proprietary training methods. Makes sense, worth noting OpenAI (also working with Mercor) is still investigating.
Hardware Push? Meta's Superintelligence Labs hired Rui Xu to build a dedicated hardware team, signaling a push toward AI-native devices beyond smart glasses and VR. Hardware will likely be Meta’s niche, they can’t compete on AI models alone. Not good enough.
🟠 Anthropic // Claude
Claude Code Source Leak: A production error accidentally published Claude Code's internal source code to the open internet, exposing the full engineering blueprint of Anthropic's flagship coding tool. Not enough to fully run it on your PC but enough to cause a stir.
Coefficient Bio Acquisition: Anthropic acquired Coefficient Bio for ~$400M in all-stock — an 8-month-old stealth biotech startup with fewer than 10 employees building AI tools for drug discovery. This is cool. Top company x Top AI use case.
Usage Drop: Claude has updated its policy to limit how much subscribers can do without purchasing extra usage. It is also restricting third party apps like OpenClaw and moving them to a pay as you go model. Has to be a cost cutting move. I do not like it one bit. Unfortunate.
🤖 Impact Industries 🚑
Robotics // Cyber Tea Farmer
DEEP Robotics deployed its LYNX M20 wheeled-legged robot dogs in China's Zhejiang province for the spring Longjing tea harvest. The robots navigate mountain paths as narrow as 50cm and slopes up to 45 degrees, hauling freshly picked leaves down to collection points before they lose flavor — a logistics bottleneck that has existed for centuries. One of the more grounded real-world robotics deployments this year: no demo environment, no warehouse floor, just a mountain.
Read the Story
Healthcare // Digital Twins
Y Combinator-backed Mantis Biotech raised $7.4M to build AI-generated "digital twins" of human bodies — synthetic patient models built from medical imaging, biometric sensors, and motion-capture data. The platform simulates clinical trial populations to predict outcomes, targeting rare diseases where real patient data is nearly impossible to collect at scale. Already used by NBA teams for sports medicine. The implications for drug development timelines are significant.
Read the Story
💻 Interview Highlight: Sam Altman with Mike Allen (Axios)
Interview Outline: Sam Altman presents a new "blueprint" for the Intelligence Age, urging Washington to prepare for a "new paradigm". The discussion focuses on the immediate risks of AI-enabled cyber and bio-attacks in the next year, while proposing radical shifts in the tax system and energy policy to manage the transition of the labor market.
About the Interviewee: Sam Altman is the co-founder and CEO of OpenAI. He is positioning himself as an "educator" for policymakers, attempting to bridge the gap between rapid research breakthroughs and societal resilience.
Interesting Quote: "Never before in human history have close to a billion people been talking to the same virtual brain.”
Condensed Interview Highlight — Sam Altman (OpenAI)
Mike Allen: What should we expect from the next class of AI models?
Sam Altman: Currently, models help scientists make small discoveries. With the next class, I expect people to say the tools helped them make career-defining discoveries. On the productivity side, you'll hear people say they can do the work of a whole software team by themselves using a few hundred GPUs.
Mike Allen: How real are the biological threats associated with these models?
Sam Altman: The models are getting very good at biology. While frontier models are currently in responsible hands, we aren't far from a world with capable open-source models. The possibility of terrorist groups using them to create novel pathogens is no longer a theoretical thing.
Mike Allen: What is the case against the government nationalizing OpenAI and its competitors?
Sam Altman: Historically, projects like Apollo or the Manhattan Project were government-run, but I don't think that would be successful today. We need the US to build superintelligence aligned with democratic values before others do, which requires a deep partnership between companies and the government.
Mike Allen: Does your policy blueprint imply a major shift in how capitalism functions?
Sam Altman: Capitalism has always depended on a balance between labor and capital. If AI proceeds as expected, way too much leverage will move toward capital. We have to think about new ideas to keep capitalism thriving and include more people as that shift happens.
Mike Allen: You've compared intelligence to a utility like electricity. What does that mean for the user?
Sam Altman: You’ll have a personal super-assistant in the cloud that plugs into everything. You won’t think about the "electricity" (the intelligence), just the products it powers. If we build enough infrastructure, the price per unit of intelligence will continue to fall astronomically fast.
👨💻 Practical Use Case (Issue 10 Revisited): API’s
Difficulty: Mid-Level
An API (Application Programming Interface) is basically a way for two apps to talk to each other. Instead of clicking buttons or copying data by hand, APIs let software connect directly in the background. Think of it as a universal translator → your app or website says “send this info,” and the other app knows exactly how to receive it.
API’s are also used within one general application to accomplish smaller tasks. In AI workflows, APIs are the glue that make everything work. Without them (or MCP which we’ll get to), ChatGPT couldn’t plug into your CRM, your AI video tool couldn’t pull from a script, and your custom agent couldn’t push notifications to Slack.
Examples:
Connect your website form to GPT via API so every submission gets auto-summarized and emailed to your team.
Use an API to send customer purchase data to an AI model, which then generates personalized follow-up messages.
Tie a payment processor like Stripe to an AI dashboard that flags unusual activity and drafts customer service responses.
API’s have been around for a while, they’re as old as the internet and just as useful. In 2000, Salesforce released one of the first enterprise API’s and in 2006, Twitter and Facebook opened up their API’s to the public. In my blog discussing building your first GPT Wrapper, API’s play a central role in connecting the chat box to ChatGPT or OpenAI itself.
It’s important to understand how API’s work in order to better understand the tech that is powered by them. Here’s a quick explainer for a better understanding!
Issue 37 Update: Nothing too new to report here just worth a reminder for the non-technical folks in the audience. API’s are a huge part of the AI ecosystem. Whenever you talk to an LLM like Claude and you’re not actually in the Claude interface, you’re likely using an API. MCP (Model Context Protocol) is also becoming popularized with agents, MCP lets AI agents connect to external tools and data sources, whereas APIs are the underlying connection mechanism.
💊 Startup Spotlight
Legion Health
Legion Health — AI-powered prescription refills for mental health care.
The Problem: Accessing mental health care is a major hurdle for hundreds of thousands of people, especially those in "shortage areas." Even stable patients who just need a routine refill for common medications often face long wait times or high costs to see a human psychiatrist just for a renewal.
The Solution: Legion Health has developed an AI app approved to handle psychiatric prescription renewals. The system is strictly limited: it only handles refills for stable patients who have previously been prescribed the drugs by a human. It currently covers a specific set of medications (like Prozac and Zoloft) and excludes anyone hospitalized in the last year. The pilot includes monthly reporting to regulators and pharmacist oversight to ensure safety.
The Backstory: Co-founded by Arthur MacWaters, the San Francisco-based startup is currently using Utah as its primary proving ground. While the company aims for a nationwide rollout by the end of 2026, the move has sparked significant debate among medical experts regarding the risks of automated "over-treatment" and the lack of human nuance in psychiatric care.
My Thoughts: This is pretty interesting. In order to start using AI in more important areas we need controlled tests like what Legion Health is doing. Obviously because AI is (still) so new, mistakes are possible, but in order to get to a point where something like this is commonplace, mistakes need to happen with a watchful eye to prevent harm.
“It’s not likely you’ll lose a job to AI. You’re going to lose the job to somebody who uses AI”
- Jensen Huang | NVIDIA CEO
Till Next Time,
Noah from KPAI