Price:

AI & TECH

SpaceX buys Cursor to own AI's stack

Thursday, June 25, 2026 · from 3 podcasts
Nerd SnipeBanklessThis Week in AI
  • SpaceX’s $60B Cursor acquisition secures orbital compute for AI training and locks in top coding talent.
  • AI labs are collapsing under GPU costs, pushing research into enterprise engineering.
  • Chinese open-weight models threaten US AI dominance on price and performance.

SpaceX is no longer just launching rockets. It’s launching AI. The $60 billion all-stock acquisition of Cursor, confirmed on June 24, 2026, marks a decisive vertical integration into the AI stack - from orbital infrastructure to developer tools. Jason Calacanis notes the deal values Cursor at 20 times revenue, a premium paid not for code, but for control over the future of agentic development.

The move solves a critical bottleneck. SpaceX had vast GPU clusters sitting idle, lacking the data and talent to train frontier models. By absorbing Cursor, Elon Musk gains access to one of the world’s richest repositories of agentic coding data - the kind that trains AI to write, debug, and deploy autonomously. As Theo from Nerd Snipe explains, this isn’t just an acquisition; it’s the formation of "SpaceX AI," a vertically integrated behemoth bypassing traditional IPO pathways.

"SpaceX had massive GPU warehouses sitting idle because they lacked the data and researchers to train competitive models."

- Theo, Nerd Snipe with Theo and Ben

The shift reflects a broader collapse of the pure research model. OpenAI and Anthropic now burn $1.25 billion a month renting GPUs, according to Nerd Snipe. That financial pressure has forced labs to abandon exploratory research in favor of enterprise engineering and data labeling. As Theo puts it, researchers once chasing "stupid swings" now manage datasets to satisfy quarterly burn rates. The dream of artificial general reasoning is being traded for survival.

Meanwhile, Google is losing ground. The departure of key talent like Adi Esmali, coupled with a retreat from affordable models, signals internal decay. Chinese open-weight models such as GLM 52 are now outperforming Google in front-end coding tasks, undermining the premise of US AI superiority. Arthur Hayes warns that US labs’ reliance on long-term financing for short-lived hardware creates a ticking financial time bomb - one that could dwarf the subprime mortgage crisis.

"The Fed can't print Moore’s Law. Pumping cash into the economy doesn't make a three-year-old chip faster."

- Arthur Hayes, Bankless

The real competition isn’t just technological - it’s structural. SpaceX now controls the full loop: hardware, energy, data, and deployment. While American AI labs strain under debt and commercialization, Musk’s empire is building closed, self-reinforcing systems. The future of intelligence may not emerge from a lab. It may launch from Cape Canaveral.

#Agents#Models#Enterprise#AI Infrastructure#Startups
AnthropicOpenAISpaceXCursorFederal Reserve

Source Intelligence

- Deep dive into what was said in the episodes

Nerd Snipe with Theo and Ben
Nerd Snipe with Theo and Ben

The US Government Banned Claude Fable 5... • Jun 24

Also from this episode: (20)

Other (20)

  • Theo was an investor in Cursor, which SpaceX AI acquired in a $60 billion all-stock deal. Theo stands to make between $50,000 and $2.5 million from the acquisition.
  • Ben notes SpaceX AI, combining XAI and SpaceX, aims to overcome its lack of data and researchers by acquiring Cursor, despite having vast GPU warehouses. This positions it as a major AI lab.
  • Ben highlights that SpaceX's IPO strategy involves complex regulations, with only about 6% of its equity currently tradeable, and early investors facing vesting schedules stretching over three years.
  • Cursor is launching 'Origin,' a new code review platform built by its acquired company, Graphite, designed to efficiently fork large monorepos. Cursor aims to be a comprehensive enterprise software development suite, including local IDE, cloud agents, and code hosting.
  • Cursor's upcoming model, possibly Composer 3, will be pre-trained from scratch, aiming for general intelligence beyond code, with a claimed size of 1.5 trillion parameters. This marks a significant leap from previous Composer models based on fine-tuned open-weight models.
  • Theo notes rumors placed GPT 4.5 at 2-12 trillion parameters and Fable between 5-10 trillion, emphasizing the ongoing uncertainty about large model sizes. Ben argues strong Reinforcement Learning (RL) can enable smaller models like GPT 5.5 to perform comparably to much larger ones.
  • Ben predicts Cursor's new model will be branded as a Grok variant and exclusively available via XAI APIs, potentially with GPT-4-like pricing. Theo expects the model will likely avoid third-party cloud platforms like Bedrock or GCP.
  • Theo and Ben agree that open-weight models like GLM 52 are becoming genuinely usable for coding tasks, with GLM 52 ranking fourth among top frontier models, comparable to Opus 4.5 in late 2023. However, they still require substantial compute resources.
  • Theo argues Google is significantly falling behind in AI, exemplified by an open-weight Chinese model (GLM 52) outperforming them in coding benchmarks and a continuous exodus of top talent. Ben adds that Google's 'thinking' approach and poor RL have hindered their progress.
  • Theo notes that while OpenAI historically led with major leaps (transformer, reasoning), Anthropic now drives 'big swings' in research and development, exemplified by Karpathy joining. Ben agrees OpenAI has shifted to an engineering-first company, focusing on gradual improvements.
  • Anthropic's expenditure of $1.25 billion per month for GPU rental from xAI highlights the immense burn rates of frontier AI labs, driving them to prioritize commercialization over purely experimental research.
  • Theo suggests a future 'GPU draft' where governments seize graphics cards, warning that NVIDIA's upcoming GB300 GPUs will include GPS trackers. He argues this is less about preventing exports to China and more about manipulating pricing and controlling usage data for purchases over $100 million.
  • The 2022 Chips and Science Act allocated $52.7 billion to boost US semiconductor manufacturing, driven by fears of China. Later in 2022, the US implemented the first ban on chip exports to China, followed by increased Senate attention to AI in 2023 after ChatGPT's launch.
  • In February (implied 2025), Anthropic refused to grant the US government unrestricted use of its models on private compute. Anthropic sought contractual carve-outs against autonomous killing and mass surveillance of US citizens, leading the government to consider them a 'supply chain risk.'
  • While Anthropic clashed with the government over direct model access with restrictions, OpenAI secured deals by offering its models as a service via APIs with built-in safety layers, avoiding direct control over government actions.
  • Theo attributes the Fable ban to a significant knowledge gap between AI developers and non-technical officials. He argues Amazon engineers, unfamiliar with modern AI, misinterpreted Mythos's capabilities when compared to outdated tools, triggering exaggerated alarm.
  • Ben argues Anthropic CEO Dario's 'fear-mongering' about AI's potential dangers through initiatives like Project Glasswing aims to force serious consideration of AI safety. However, Ben views this strategy as counterproductive, as it scares non-technical audiences into irrational actions.
  • Theo describes OpenAI's internal 'Grug speak' optimization, where models use highly compressed, nonsensical language to reduce token count and improve reasoning efficiency, a leaked detail he finds amusing.
  • Theo and Ben discuss how macOS's `syspolicyd` process, designed for security, creates a significant CPU bottleneck for AI agent workloads by extensively monitoring numerous sub-processes. This pushes them to offload tasks to Linux machines.
  • Theo reveals a 'pro tip' for advanced Codex users: a config file setting allows bumping the default parallel agents from three to 20 per thread, potentially burning 'hundreds of dollars of inference an hour.' Ben suggests asking Codex itself to configure this.
#Big Tech#Startups#Coding#Models#Enterprise
Bankless
Bankless

"The Fed Can't Print Moore's Law" - How the AI Crash Sends Bitcoin to $1M | Arthur Hayes • Jun 22

  • Arthur Hayes expects a future AI credit crisis driven by the mispricing of GPU depreciation, where multi-year financing deals clash with a two-year effective hardware lifespan. This capital misallocation will dwarf the subprime mortgage crisis.
  • Hayes argues the subsequent money printing by authorities to bail out the system will fail to revive the AI sector, as the Fed cannot print technological progress or change Moore's Law. That stimulus capital will instead flood into crypto, sending Bitcoin to $1 million.
  • Hayes is currently positioned in cash and T-bills, having sold tokens like Hype and Zcash. He cites risk management, capital preservation, and a search for new asymmetric trades as his reasons for taking profits.
  • He views Ethereum as a high-conviction, large-cap investment with a favorable setup, trading well below its all-time high while other major assets have breached theirs. Hayes would allocate a marginal dollar to Ether over Bitcoin on a chart perspective.
  • Hayes foresees a significant rally in hydrocarbon prices due to global strategic inventory rebuilding and geopolitical risk, viewing current lows as a buying opportunity for energy equities.
  • He identifies Chinese AI models as a critical threat to US AI valuations, citing their potential to commoditize the market by offering comparable performance at a fraction of the cost, eroding the premium for US tech brands.
  • Hayes predicts the 2028 US presidential election could become a referendum on AI, driven by public anger over economic displacement, environmental costs, and wealth inequality, which would create regulatory risk for the sector.
Also from this episode: (3)

Startups (2)

  • He credits BitMEX with inventing the modern perpetual swap, emphasizing the funding rate mechanism and socialized loss model as the key innovations that enable high-leverage, 24/7 trading for retail.
  • Hayes believes perpetual swaps will dominate because they cater to retail demand for leverage and continuous markets, and he expects Hyperliquid to eventually flip Binance as the offshore leader due to its superior, cost-efficient product.

Markets (1)

  • Despite creating the instrument, Hayes personally trades only spot, advising against leverage for most investors due to crypto's inherent volatility. He occasionally participates as a liquidity provider in perp markets when rates are attractive.
#Models#BTC Markets#Macro#Big Tech#AI Infrastructure
This Week in AI
This Week in AI

Why AI Models Aren’t the Product Any More | TWiAI Ep 18 • Jun 18

  • Jason Calacanis says SpaceX plans to acquire Cursor for $60 billion in stock, giving Cursor access to SpaceX’s massive compute resources for model development.
  • Ali Ansari argues the real product in AI is not the model but the agent layer, evaluations, harness, and UI built atop it. The frontier of intelligence will be defined by application companies owning their proprietary workflows.
  • Jason Calacanis warns that platform companies like OpenAI or Anthropic will study and eventually compete with their most successful application-layer customers, citing historical examples from Microsoft, Facebook, and Apple.
  • Ryan Daniels describes Crosby Legal as an AI-first law firm that uses AI internally to provide scalable legal services at flat rates, aligning incentives by eliminating billable hours.
  • Ali Ansari says Micro One pivoted from an AI recruiter tool to a marketplace providing pre-vetted human experts who train AI models, achieving about $300 million in ARR by April 2026.
  • Jason Calacanis notes SpaceX’s market cap hit $2.88 trillion after its IPO, making it the fourth most valuable US company. He calculates the Cursor acquisition at a 20x revenue multiple.
  • Ryan Daniels explains that Cursor once accounted for 40-50% of Anthropic’s total revenue, but Anthropic’s launch of Claude Code forced Cursor to build its own model, leading to the SpaceX deal.
  • Ali Ansari defines model distillation as using an open-source baseline model and conducting massive post-training that changes most weights, creating a distinct model without reinventing core reasoning.
  • Ryan Daniels argues only a handful of companies can build frontier models due to capital and compute constraints, creating a binary divide between model-makers and everyone else.
  • Jason Calacanis cites Claude Code’s rumored $2.5 billion revenue run rate and Cursor’s $4 billion run rate, predicting the AI coding market will reach $100 billion soon.
  • Ryan Daniels says Crosby’s vertically integrated law firm creates a proprietary feedback loop where lawyers train AI on subjective legal judgment, making human expertise more valuable as models improve.
  • Ali Ansari predicts nearly 100% of future AI data spend will focus on the agent and application layer, with orders of magnitude more agents built than base models.
  • Jason Calacanis describes his vision for an internal venture AI trained on Slack and Notion data to analyze investment history and missed opportunities, calling Slack’s corpus the ultimate dark data pool.
  • Ryan Daniels forecasts AI lawyers could match the average practicing attorney by late 2027, forcing courts to consider ethical access to AI counsel for self-representation.
  • Ali Ansari and Ryan Daniels collaborated on a multi-turn contract redlining benchmark using real lawyer negotiations to evaluate AI models, finding current models perform at only 10-20% of human capability.
  • Ali Ansari proposes an industry-led AI safety consortium where competing model companies create adversarial benchmarks for self-regulation, similar to the MPAA for movies.
#Big Tech#Coding#Agents#Startups

Related Stories