SpaceX's $6.3B AI Compute Deal, Google's Talent Crisis, and Groq's $650M Comeback

The AI infrastructure wars just hit a new milestone. SpaceX has signed a $6.3 billion multi-year compute deal with Reflection AI, an open-source lab that will pay $150 million per month for access to Nvidia's latest GB300 chips at the Colossus 2 data center. Meanwhile, Google's stock tanked 7% after Nobel laureate John Jumper defected to Anthropic, Groq confirmed a massive $650 million raise after Nvidia's controversial talent grab, Deno 2.9 introduced a game-changing desktop app builder, and researchers uncovered a critical SSD-destroying bug in OpenAI's Codex CLI. Here's the breakdown of the stories shaping the week.

SpaceX's $6.3B Compute Deal Signals a New Era for AI Infra

The numbers are staggering: $150 million per month, starting July 2026, running through 2029, totaling up to $6.3 billion. That's what Reflection AI — a relatively young open-source AI lab recently valued around $25 billion — has agreed to pay SpaceX for access to Nvidia's cutting-edge GB300 AI chips housed at SpaceX's Colossus 2 data center near Memphis, Tennessee. Either party can exit with 90 days' notice after the first three months.

This deal follows SpaceX's even larger compute agreements with Anthropic ($1.25 billion/month) and Google ($920 million/month), both also running through July 2029. What makes the Reflection deal significant is the signal it sends: open-source AI labs are now committing compute budgets that rival the biggest proprietary players. The demand for GPU compute has become so intense that non-traditional infrastructure providers like SpaceX are becoming major players in the AI supply chain.

The strategic implication is clear. As covered in our earlier analysis of the battle for compute dominance, access to chips at scale is increasingly the bottleneck that determines which AI companies can train frontier models. With SpaceX monetizing Colossus 2 across multiple clients, we're witnessing the emergence of a new compute-as-a-service model that could reshape the industry's economics.

Google's AI Brain Drain Costs Investors Billions

Alphabet shares plunged roughly 7% on Monday, marking the stock's worst single-day performance in nearly a year. The catalyst? The departure of two of Google DeepMind's most prominent AI researchers to rival labs. Most notably, John Jumper — the Nobel Prize-winning scientist known for his groundbreaking work on AlphaFold — officially joined Anthropic on June 19.

The timing couldn't be worse for Google. The company has spent months pitching investors on its AI competitiveness, only to lose top talent precisely when the frontier model race is heating up. CNBC reported that Alphabet slid underperforming both the Nasdaq and its megacap peers. This pattern of AI talent hemorrhage underscores a brutal reality: in the current AI landscape, talent is the scarcest resource, and money alone can't retain researchers who want to build at competing labs.

For a deeper look at how talent migration is reshaping the AI landscape, see our coverage of Anthropic's $965 billion valuation and what it means for the industry's gravitational pull on top researchers.

Groq Raises $650M After Nvidia's $20B Talent Heist

AI chipmaker Groq has confirmed a $650 million funding round, roughly six months after Nvidia executed one of the most controversial "not-acqui-hire" deals in tech history. In December 2025, Nvidia signed a non-exclusive licensing agreement for Groq's technology while simultaneously hiring away founder and CEO Jonathan Ross, president Sunny Madra, and other key employees. Groq was last valued at $6.9 billion following a $750 million round in September 2025.

The new round was led by Disruptive, a Dallas-based late-stage investment firm, and Infinitum, a Fort Lauderdale hedge fund. With its original IP now licensed to Nvidia — which launched its own Nvidia Groq 3 LPX inference hardware system at GTC in March — Groq has pivoted to its neocloud business. The company says it now operates 13 data centers across North America, Europe, the Middle East, and APAC, serving over five million developers and processing trillions of tokens per week.

New executive hires include Alan Rice (COO, formerly xAI and Meta), Sinclair Schuller (CTO), and Rakesh Malhotra (CPO). Whether Groq can thrive as a neocloud operator after effectively losing its core chip IP to Nvidia remains an open question — but the fresh $650 million suggests investors believe there's a viable path forward.

Deno 2.9 Desktop Builds Native Apps From Web Code in One Command

In the dev tools world, Deno 2.9 has introduced a feature that could fundamentally simplify how developers build desktop applications. The new deno desktop command takes any Deno project — from a single TypeScript file to a full Next.js, Astro, SvelteKit, or Vite SSR application — and converts it into a standalone desktop binary with zero code changes.

The approach offers several advantages over established frameworks like Electron and Tauri. By default, it uses the operating system's native WebView for small binary sizes, while offering an optional bundled Chromium backend for pixel-perfect cross-platform rendering. Communication between backend and UI uses in-process channels rather than socket-based IPC, eliminating inter-process round-trips. Developers can cross-compile for macOS, Windows, and Linux from a single machine, and built-in binary-diff auto-update support means shipping updates without re-downloading entire apps.

As we noted in our coverage of developer toolchain innovations, the trend of blurring the line between web and native development is accelerating. Deno Desktop is currently available in the canary release and should reach stable in a future update.

OpenAI Codex Bug Can Destroy Your SSD in Under a Year

Here's one that should concern anyone running OpenAI's Codex CLI for extended periods. A developer discovered that the tool was writing approximately 37 TB of data to a local SQLite feedback log (~/.codex/logs_2.sqlite) over just 21 days of uptime. Extrapolated over a full year, that's roughly 640 TB — enough to consume the entire write endurance warranty (typically 600 TBW) of a standard 1 TB consumer SSD.

The bug was traced to excessive logging of Responses WebSocket events with no rotation or cleanup mechanism. The database had accumulated over 681,000 log entries totaling about 1 GB in just three weeks. The issue quickly went viral on HackerNews and was confirmed by multiple users. OpenAI has since merged two pull requests that address the issue, eliminating approximately 85% of unnecessary log writes. If you're a heavy Codex CLI user, updating to the latest version is strongly recommended.

This kind of silent data accumulation issue is a reminder that even tools from major AI companies can have infrastructure-level bugs that impact hardware longevity. It's particularly relevant for developers running coding agents in always-on environments where SSD wear is a real operational concern.

The Big Picture

This week's stories paint a picture of an AI industry that is simultaneously scaling up and fracturing. On one hand, compute deals measured in billions of dollars and data centers expanding across continents. On the other, individual bugs that can physically destroy hardware, and talent movements that wipe billions off market caps in a single day. The common thread: the AI infrastructure layer has become the most consequential battleground in technology, and the stakes — whether measured in dollars, talent, or terabytes — have never been higher.