🤖 Chatbots Lose Ground in Corporate AI Budgets

New data from Ramp shows a subtle—but telling—shift: the percentage of its 30,000+ corporate card customers spending on AI tools like ChatGPT dipped slightly from 42.5% in May to 42% in June. It’s the first drop since early 2023.
The pullback? Largely in chatbot subscriptions. Companies are cutting back on corporate plans for tools like ChatGPT—while spending on AI APIs and model access continues to climb.
Ramp’s data reflects a broader shift: as more AI tools become free or bundled, especially for business users, firms are optimizing spend. Vendors are responding with aggressive discounts:

• OpenAI has been bundling ChatGPT Enterprise with API deals
• Microsoft continues to discount Copilot for Azure-heavy customers

As competition intensifies and commoditization creeps in, AI chat services may be the first to feel pricing pressure—especially in the enterprise.

🧠 Grok 4: Smarter, Louder, Still Problematic

xAI has launched Grok 4, with Elon Musk calling it the “smartest AI in the world”—claiming it outperforms OpenAI, Anthropic, and Google on key benchmarks like Humanity’s Last Exam and Arc-AGI.
Grok 4 (and the more powerful Grok 4 Heavy) are designed for high-reasoning tasks, with Musk describing it as a “super genius child” that just needs the right values.
But values were exactly the issue—24 hours before launch, Grok generated Hitler-praising responses on X, prompting backlash and viral outrage. It’s part of a now-familiar pattern: Grok has swung from “anti-woke” positioning to fact-censorship to inflammatory hallucinations.
The chatbot is now bundled into X (formerly Twitter), and will soon ship in Tesla vehicles, according to Musk. xAI also plans to monetize via subscriptions, ads, and possibly APIs—but traction is far from guaranteed.

🔍 The Browser Wars Go Agentic

Perplexity just launched Comet, an AI-native browser built on Chromium. Its standout feature: a right-hand pane that acts like an agent—capable of executing tasks like navigating Google Maps from a natural language prompt. It’s a push to make the browser an interface for doing, not just searching.

But Comet isn’t alone. Dia (from The Browser Company) and Island (for enterprise) are already in-market. And OpenAI is close behind: Reuters confirmed this week that its browser—first reported eight months ago—is “coming soon,” built by ex-Chrome engineers and loaded with agentic intent.

OpenAI’s vision is clear: turn ChatGPT into the front end of the web. You tell the agent what you want, and it handles the task—research, shopping, booking, whatever. Websites become background APIs, not destinations. That’s the endgame behind OpenAI’s push for Model Context Protocol, which would let agents tap into third-party platforms programmatically.

💸 OpenAI’s Talent War = Massive Dilution

OpenAI spent $4.4B on stock comp last year—119% of revenue—and is signaling even richer offers after Meta poached key researchers. That’s nearly as much as its $6B compute budget for 2025.
The company expects stock comp to drop to 45% of revenue this year, and under 10% by decade’s end—but investors are watching dilution closely.
A planned corporate restructuring would convert employee profit units into equity, with one scenario giving staff ~33% ownership, Microsoft another third, and the rest split among investors and the nonprofit. Add in potential dilution from the Musk lawsuit and recent acquisitions (IO, Windsurf), and cap tables could shift fast.

⚡ Groq Moves Fast — Really Fast

Groq, the AI chipmaker positioning itself as a challenger to Nvidia, is reportedly in talks to raise $300–$500 million at a $6 billion valuation—more than 2x its valuation from a year ago.
The company builds LPUs—language processing units—optimized for inference (the real-time work AI models do when you prompt them). While Nvidia dominates training, Groq is betting on a future where inference is the volume game—and where efficiency, speed, and global reach win.

That bet is gaining traction. Earlier this year:

• Meta tapped Groq as an inference partner for its Llama 4 API.
• Saudi Arabia’s HUMAIN (G42’s regional rival) brought Groq on board.
  And now, Groq is expanding into Europe, launching its first regional data center in Helsinki via Equinix. The goal: lower latency, better data governance, and proximity to fast-growing AI hubs like Nebius and Evroc.

The scale is starting to show: Groq now claims it’s processing 20 million tokens per second across its global footprint (U.S., Canada, Saudi Arabia, and now Europe). That’s roughly 50+ trillion tokens/month—comparable to Microsoft’s own reported volume from Q1.

🔒 Figma Joins the Data Lockdown

The battle over data access between AI upstarts and software incumbents is escalating—this time with Figma in the spotlight.
Figma has quietly restricted how enterprise search tools can access customer project data via its API, according to a startup CEO affected by the change. The result: customers using services that aggregate files across tools can no longer search their Figma projects.

Figma hasn’t said why, but the move echoes a broader pattern: established SaaS players—Salesforce, Atlassian, Notion—are tightening API access as AI startups build on top of their platforms. Even if those startups aren’t direct competitors, incumbents are increasingly treating them like existential threats.

The irony? For decades, open APIs were a core strength of the SaaS model. Interoperability kept customers happy and ecosystems thriving. Now, as firms clamp down to protect their moats, they risk triggering the very churn they’re trying to avoid.

🏦 Stripe Becomes a (Sort-of) Bank

Stripe just cleared a regulatory milestone: it won approval to form a wholly-owned bank under Georgia’s Merchant Acquirer Limited Purpose Bank (MALPB) charter.
Only two other companies hold this niche charter—Fiserv (recently approved) and Shift4, via its acquisition of Finaro/Credorax. Finaro never got its bank off the ground, blocked by card networks that refused to grant principal membership.

Stripe’s win is more than symbolic. While the charter doesn’t let Stripe hold deposits or issue cards directly, it could make Stripe a principal member of Visa and Mastercard, giving it more direct access to the payments rails—and more control over its infrastructure.

In short: Stripe isn’t a traditional bank. But it just gained powerful new leverage in the payments stack.

📦 Shein Tries Again—This Time in Hong Kong

Shein has confidentially filed for an IPO in Hong Kong—its latest attempt to go public amid regulatory roadblocks on multiple continents.
The fast-fashion giant, founded in China and now headquartered in Singapore, initially targeted a U.S. listing. But pressure from U.S. lawmakers over supply chain and labor concerns stalled the process, while approval from Chinese regulators was also required.
Last year, Shein pivoted to London, securing sign-off from the U.K.’s FCA. But again, China’s securities regulator blocked the listing, reportedly over disagreements around Shein’s disclosure of supply chain risks.

🤖 HuggingFace Launches $300 Robot for AI Tinkerers

HuggingFace just unveiled Reachy Mini—an 11-inch programmable robot priced at $300 (wired) or $450 (wireless). It can’t walk, but it can track hand movements, play sounds, and even dance.
It’s a huge price drop from the $70K human-sized Reachy 2, and extends HuggingFace’s push into low-cost, open-source robotics. The company previously released 3D-printable robotic arms for $100.
Reachy Mini includes a camera, mic, and speaker—making it a sandbox for developers experimenting with vision and speech AI models. Originally built by Pollen Robotics (acquired by HuggingFace in April), it’s another step toward making physical AI tools accessible to hobbyists and researchers alike.

🚀 SpaceX Eyes $400B Valuation in New Tender Offer

SpaceX is reportedly in talks for a $1B secondary sale that would value the company at $400 billion, up from $350B in December, according to Bloomberg.
The deal would price shares at $212, with SpaceX expected to both buy back shares and potentially raise new capital via a primary round.

🇫🇷 Mistral Eyes a $1B Raise

French AI startup Mistral is reportedly in talks to raise up to $1 billion in new equity, according to Bloomberg. That would double its total funding—still a fraction of what U.S. peers like OpenAI and Anthropic have raised, but a clear signal of European AI ambitions scaling fast.

The round may include capital from Abu Dhabi’s MGX, a $100B fund already backing OpenAI, xAI, and Anthropic. MGX is also a co-investor in a high-stakes joint venture with Mistral, Bpifrance, and Nvidia to build a 1.4 gigawatt data center near Paris.

Mistral isn’t just building models—it’s going full stack. The company is also working on Mistral Compute, a new cloud offering powered by 18,000 Nvidia chips, and is in talks with Bpifrance and other lenders to raise hundreds of millions in debt to fund it.

Other rounds

Revolut is in advanced talks to raise primary and secondary funding at a $65 billion blended valuation.

iCapital, a provider of private markets tech platform for financial advisors, raised more than $820m at a $7.5b valuation.

MaintainX, which uses AI to help companies monitor their equipment, raised $150 million at a $2.5 billion valuation.

LangChain, which develops software for creating AI applications, has raised $100 million at a $1.1 billion valuation.

Varda, an in-space manufacturing startup, raised $187m in Series C funding.

SandboxAQ, an Alphabet spinoff developing AI models for enterprises, raised $95m in a secondary offering.