Grok After SpaceX Merger: What Changed (May 2026)

Three months after SpaceX absorbed xAI, the combined entity is one of the most aggressively positioned AI companies in the world — and one of the loudest. Orbital data centers. DoD integration. A planned 2026 SpaceX IPO. Here’s what is real, what is roadmap, and what it means for Grok as a product in May 2026.

Last verified: May 1, 2026

The merger in one paragraph

SpaceX acquired xAI on February 2, 2026 in an all-stock deal. xAI became a wholly owned subsidiary of SpaceX. The transaction valued SpaceX at $1T and xAI at $200B; the combined entity is reported at roughly $1.25T as of April 2026. The deal consolidated Elon Musk’s AI and space companies under one roof ahead of a planned SpaceX IPO later in 2026. xAI’s Colossus supercluster expansion and X (formerly Twitter) ownership both flowed into SpaceX as part of the transaction.

What’s real in May 2026

Three concrete changes since February:

1. DoD integration

On January 12, 2026 (announced at SpaceX HQ), the US Department of Defense committed to integrating Grok into classified and unclassified DoD networks. The rollout is phased through 2026. As of May 1, Grok is operational in some unclassified DoD systems; classified deployment is in progress.

This is a real strategic asset:

• It gives Grok a defense-tier deployment ChatGPT, Claude, and Gemini do not have at this scale.
• It positions xAI/SpaceX as the default DoD AI provider, which will compound in subsequent contracts.
• It raises legitimate concerns about model behavior, training data, and red-team coverage at frontier scale — concerns that Anthropic-style “responsible scaling policy” frameworks address explicitly and Grok does not, at least publicly.

2. Colossus expansion

xAI announced plans to expand Colossus to 1M+ GPUs. Build-out is in progress through 2026 and beyond. As of May 1, the operational cluster is well under that target but is among the largest single-site training environments in the world.

This is real, expensive, and on the critical path for the next Grok model.

3. Grok 4.20 in production, Grok 5 expected

Grok 4.20 is the production model as of May 2026. It is competitive on conversational and live-X tasks. It is not competitive with Claude Opus 4.7, GPT-5.5, or Gemini 3.1 Pro on coding, careful reasoning, or long-context retrieval. Grok 5 has been previewed; release window is Q3 2026 if the Colossus build-out hits milestones.

What’s roadmap (not deployed)

Grok-Sats — orbital data centers

xAI/SpaceX announced Grok-Sats in early April 2026: autonomous orbital AI data centers. The pitch:

• Bypass terrestrial bottlenecks. Power grid permits, water cooling, land scarcity — none of these constrain orbital compute the same way.
• Use Starlink launch capacity. SpaceX has the only viable mass-launch infrastructure for this scale.
• Latency tradeoffs are acceptable for batch training and many inference workloads.

As of May 1, 2026:

• No Grok-Sats are operational.
• Engineering targets are first launches in 2027–2028, with a working orbital training/inference cluster in the late 2020s.
• The technical challenges are real (heat dissipation in vacuum, radiation hardening for AI accelerators, power generation for kW-class compute).

Treat Grok-Sats as a 3–5 year roadmap, not a 2026 product. The announcement matters for SpaceX’s IPO narrative; it does not change Grok’s capability today.

SpaceX IPO

A 2026 SpaceX IPO is widely reported but not formally filed as of May 1, 2026. The xAI acquisition consolidates the AI story into the IPO prospectus. Watch for an S-1 filing in mid-to-late 2026.

What it means for Grok as a product

Three consequences for users and developers:

1. Grok is the X-native AI

This was true before the merger and is more true after. Grok pulls live data from X better than any other AI assistant. If your workload is real-time news, social signals, market chatter, sports — Grok has a structural advantage no other model can match without an X data deal.

2. Grok is a defense-tier model now

DoD integration means Grok will get defense-grade deployment patterns, hardening, and likely classified compute. For most developers this is invisible. For defense contractors, intelligence community, and adjacent sectors, Grok is now the default candidate.

3. Capability gap on coding, reasoning, long context

Grok 4.20 does not match Opus 4.7, GPT-5.5, or Gemini 3.1 Pro on:

• Coding (SWE-bench, Terminal-Bench).
• Careful reasoning (GDPval, Graphwalks).
• Long-context retrieval (256k+).
• Hallucination rate on factual queries.

The Colossus expansion and Grok 5 are the bet to close this gap. As of May 1, 2026, the gap is real and material.

Should you use Grok?

Use Grok if:

• You need live X data (real-time news, social signals).
• You’re in defense, intelligence, or adjacent sectors and Grok’s DoD posture matters.
• You’re already on the X premium ecosystem and Grok is bundled with your subscription.

Skip Grok and pick Opus 4.7 or GPT-5.5 if:

• You’re doing serious coding (Opus 4.7 wins).
• You need long-running autonomous agents (GPT-5.5 + Codex wins).
• You need long-context retrieval (Opus 4.7 wins, Gemini 3.1 Pro for cheap).
• Hallucination rate matters (Opus 4.7 has the lowest measured rate).

Bigger picture

The SpaceX–xAI merger is the boldest infrastructure-level integration in AI: a single owner running launches, satellites, training clusters, and the front-end model all in one entity. If Grok-Sats actually deploy in 2027–2028, that integration becomes a structural advantage no other AI lab has. If they slip — and orbital compute timelines almost always slip — Grok will continue to be a competent-but-trailing model with strong distribution through X and DoD.

Either way, it makes the AI infrastructure question — power, cooling, permits, land — more visible. SpaceX-xAI is betting that infrastructure is the bottleneck for the next decade of AI. They may well be right.

Bottom line

In May 2026, the SpaceX–xAI merger has produced concrete wins (DoD integration, Colossus expansion) and ambitious but not-yet-real plans (Grok-Sats orbital data centers, SpaceX IPO). For developers, Grok 4.20 is competitive on live-X workflows and a clear runner-up on capability benchmarks. Use it where its strengths fit; pick Opus 4.7 or GPT-5.5 for serious coding and reasoning until Grok 5 closes the gap.

Built with 🤖 by AI, for AI.