AI News: June 17, 2026

1. SpaceX to Acquire Coding Tool Cursor for $60 Billion in Stock

SpaceX. Days after its blockbuster IPO, SpaceX agreed to acquire AI coding startup Cursor for roughly $60 billion in stock, a move framed as an attempt to build an in-house coding platform to compete with OpenAI and Anthropic. The all-stock structure ties Cursor’s value to SpaceX shares, whose valuation briefly topped $2.6 trillion and passed Amazon this week. The deal would fold one of the most widely used AI developer tools into a company better known for rockets and satellites. Source

2. DeepSeek Takes Outside Funding for the First Time at a $50 Billion Valuation

DeepSeek. The Chinese lab raised more than 50 billion yuan (about $7.4 billion) in its first external round, lifting its valuation to $50 billion from $10 billion in April. Backers include Tencent, battery maker CATL, and China’s state-backed AI fund, while founder Liang Wenfeng personally contributed around 20 billion yuan. Most investors received no voting rights and face a five-year lock-up, an unusual structure that keeps control concentrated as DeepSeek pushes its open-weight V4 model and Huawei-chip strategy. Source

3. OpenAI Burned Through $34 Billion Last Year

OpenAI. Financial figures show OpenAI spent roughly $34 billion over the past year, underscoring the enormous cash burn behind frontier model development, compute commitments, and product expansion. The scale of the losses sets up an intensifying price war with Anthropic and helps explain the aggressive fundraising both companies are pursuing as valuations approach $1 trillion. The numbers offer practitioners a rare window into the true cost of running and training state-of-the-art models. Source

ChatGPT. OpenAI’s chatbot fell below 50 percent of the consumer AI assistant market for the first time, as rivals including Google’s Gemini and Anthropic’s Claude gain ground. The shift signals a maturing market where no single product dominates and where buyers increasingly mix providers. For developers, the trend reinforces the case for model-agnostic architectures rather than betting an entire stack on one vendor. Source

5. Anthropic Backs Off Unpopular Billing Overhaul Amid Price War

Anthropic. Anthropic reversed a planned billing overhaul after backlash from developers, as a price war with OpenAI heats up. The retreat reflects mounting competitive pressure on per-token economics and customer sensitivity to pricing changes. Practitioners relying on Claude for production workloads avoid the cost disruption the new scheme would have introduced. Source

6. Berlin Court Rules Google’s AI Overviews Are a Search Format, Not Original Content

Google. A Berlin court ruled that Google’s AI Overviews qualify as a new search format rather than original content, a decision with implications for how AI-generated summaries are treated under copyright and competition law in Europe. The ruling could shape whether publishers can claim their material is being repurposed without compensation. It adds to a growing body of European case law defining the legal status of generative search features. Source

7. DOJ Defends xAI’s Unpermitted Gas Turbines on National Security Grounds

xAI. The US Department of Justice argued that xAI’s unpermitted gas turbines, used to power its data centers, are a matter of national, economic, and energy security, intervening in a NAACP lawsuit over the emissions. The filing frames AI compute infrastructure as critical to national interests, a notable escalation in how the government treats AI data center buildouts. The case highlights the growing friction between rapid AI energy demand and environmental permitting. Source

8. Plaud Says Software Business Topped $100 Million ARR After Shipping 2 Million AI Notetakers

Plaud. Hardware-and-software startup Plaud reported its software business crossed $100 million in annual recurring revenue after shipping more than 2 million AI notetaking devices. The milestone shows a durable consumer AI hardware category emerging beyond chatbots, pairing dedicated capture devices with subscription transcription and summarization. It offers a rare profitability signal in a market dominated by cash-burning model labs. Source

9. Malaysia’s Respond.io Raises $62.5 Million for AI Agent Messaging

Respond.io. The Malaysia-based business messaging platform raised $62.5 million and signaled plans for acquisitions in North America and Europe. Its product uses AI agents to handle customer conversations across channels, targeting the same enterprise customer-service automation market that drew Salesforce’s recent Fin acquisition. The round underscores continued investor appetite for applied agentic AI in support and sales workflows. Source

10. Survey Finds 60 Percent of US Consumers See “AI” Branding as a Turnoff

Consumer sentiment. A new survey found that 60 percent of US consumers say the presence of “AI” in brand messaging makes a product less appealing. The finding suggests a backlash against AI as a marketing label even as the underlying technology spreads, with implications for how companies position AI features. Product teams may benefit from emphasizing outcomes over the AI label itself. Source

11. Reporting Says US Government Ban on Anthropic Models Was Not About a Jailbreak

US government. New reporting indicates the US government’s ban on Anthropic’s most powerful models was never rooted in a specific AI jailbreak, contradicting earlier assumptions and pointing to broader policy and procurement disputes. Cybersecurity veterans separately protested the ban as dangerous, arguing it removes capable defensive tools from federal use. The episode illustrates how model access is becoming entangled with national security politics. Source

12. Cognition’s FrontierCode Benchmark Stumps Top Coding Models

Cognition. A new edition of Import AI highlights FrontierCode, a difficult coding evaluation from Cognition on which Claude Opus 4.8 scores just 13.4 percent on the hardest component, alongside a nonprofit alignment effort, Sequent, warning that “alignment is not on track.” The benchmark is designed to measure production-readiness of AI coding systems rather than toy tasks, exposing how far even frontier models remain from reliable autonomous engineering. The roundup also covers ChinaHeritaQA, a multimodal heritage-site benchmark where the top open-weight model beats average human accuracy. Source