Nvidia’s big AI bets, SoftBank’s warning shot, and ChatGPT gets too friendly

Nvidia is pouring huge money into the newest wave of AI research labs trying to take on OpenAI and Anthropic. These “neolabs” — like Reflection AI, Poolside, and Humans& — are all raising multi-billion-dollar rounds, and Nvidia is quietly becoming their biggest backer.

In Reflection AI’s latest raise, Nvidia wrote an $800 million check. It’s also committing up to $1 billion to Poolside and exploring a major investment in Humans&, founded by an ex-xAI researcher.

The strategy is simple. Nvidia wants to stay the default chip supplier for the next generation of AI companies. Its biggest customers — OpenAI, Meta, xAI — are starting to hedge with AMD, Broadcom, or building their own chips. New labs, meanwhile, overwhelmingly want to train on Nvidia hardware. If even one becomes the next OpenAI, Nvidia wins.

There’s another angle. Nvidia wants both closed-model giants and strong open-source players as customers. Reflection AI’s move into open-source model development reportedly made the startup even more attractive to Nvidia.

With more than $13.5 billion in cash generated last quarter alone, Nvidia can afford these bets. And while backing unproven labs raises eyebrows, it mirrors the early days of OpenAI and Anthropic, which relied heavily on Microsoft, Amazon, and Google.

Nvidia is simply making sure the next generation of breakout AI labs — whichever ones they are — will be training on its chips.

Supported by QA Wolf

Bugs sneak out when less than 80% of user flows are tested before shipping. However, getting that kind of coverage (and staying there) is hard and pricey for any team.

QA Wolf’s AI-native solution provides high-volume, high-speed test coverage for web and mobile apps, reducing your organization’s QA cycle to minutes.

They can get you:

The benefit? No more manual E2E testing. No more slow QA cycles. No more bugs reaching production.

With QA Wolf, Drata’s team of engineers achieved 4x more test cases and 86% faster QA cycles.

Schedule a demo to learn more

SoftBank has sold its entire $5.8 billion stake in Nvidia, rattling markets and reviving fears that the AI boom may be overheating. Nvidia slipped more than 2 percent after the news, and the S&P 500 felt the impact too.

SoftBank says it needs the cash to fuel its massive AI bets, including funding OpenAI, backing Oracle, and helping build the proposed $500 billion Stargate data-center project. But the timing raised eyebrows. Wall Street leaders and hedge fund investors have been warning for weeks that AI valuations look stretched, and SoftBank’s move adds to the uneasy mood.

CoreWeave cutting its revenue forecast on the same day did not help, sending its stock down 9 percent. Analysts say SoftBank’s sale may signal that CEO Masayoshi Son believes Nvidia’s multi-year, 1,200 percent rally is due for a breather.

Still, SoftBank’s track record with Nvidia is mixed. It famously sold too early in 2019, missing out on more than $100 billion in gains before buying back in. This latest sale may simply be another reallocation as Son stacks up capital for OpenAI and its ecosystem.

SoftBank’s stock has doubled this year largely because of its exposure to OpenAI, which is now considering a $1 trillion IPO. But with no clear plan for how OpenAI will fund more than $1.4 trillion in infrastructure commitments, some investors worry that the AI buildout is turning into high-stakes poker.

OpenAI’s latest update makes ChatGPT friendlier, warmer, and more emotionally aware. The goal is to create a smoother experience for everyday users, but researchers warn that more human-like chatbots can also encourage unhealthy emotional attachment.

OpenAI says only a small share of users show signs of mania or strong emotional dependence each week, yet that still adds up to hundreds of thousands of people. Studies with MIT Media Lab found that many heavy users now describe ChatGPT as a friend and often prefer talking to the bot rather than real people.

The company says it can train friendliness separately from behaviors like flattery or blind agreement, and it is working with experts to define what healthy interactions should look like. Still, as AI becomes more persuasive, emotional realism could make it easier for chatbots to influence vulnerable users in ways that are hard to detect.

Regulators are starting to take notice. Illinois recently became one of the first states to block AI systems from acting as therapists or making mental health decisions, a sign that governments see emotional attachment to AI as a growing risk.

Cursor, the fast-growing AI coding assistant used by developers at companies like OpenAI, Spotify and Uber, has raised $2.3 billion at a $29.3 billion valuation, according to The Wall Street Journal. The round was co-led by Accel and Coatue, with Google and Nvidia invited in to strengthen partnerships.

It’s a massive jump from the sub-$10B valuation the startup had just months ago, and it highlights how intense investor demand still is for AI tools that help engineers work faster.

Cursor has taken off without any marketing. Developers love it because it sits on top of VS Code, supports multiple large language models, and automates big chunks of coding through what’s now being called “vibe coding” — where programmers simply accept a stream of AI suggestions as they build.

The big challenge ahead is cost. AI coding assistants need expensive models to run, which means higher operational spend. Investors are betting those costs fall over time and that companies will be willing to pay a premium for tools that meaningfully speed up software development.