SpaceX Owns Every Layer of AI Now. The Model Is Still the Weak Link.

📊 Full opportunity report: SpaceX Owns Every Layer of AI Now. The Model Is Still the Weak Link. on ThorstenMeyerAI.com — validation score, market gap, and execution plan.

TL;DR

SpaceX has purchased Cursor, gaining control of every layer of the AI stack from hardware to applications. Despite this dominance, the core AI model remains a weak link, raising questions about future capabilities.

SpaceX has finalized its acquisition of Cursor for $60 billion in all-stock, gaining control over every layer of the AI stack, from hardware to applications. This move consolidates the company’s position as a dominant force in AI infrastructure, but the core AI model itself remains a weak link, raising questions about future performance and competitiveness.

On June 16, SpaceX announced the completion of its $60 billion all-stock acquisition of Cursor, a profitable AI coding company founded in 2022. The deal, expected to close in Q3 2026, makes SpaceX the owner of all AI layers, including hardware, data centers, research labs, and application models. Cursor’s revenue, which reached approximately $4 billion annually by June, was built on its AI coding tools, notably its Cursor model, which is now integrated into SpaceX’s broader AI ecosystem.

By acquiring Cursor, SpaceX now controls the entire AI stack: from the Colossus supercomputers in Memphis, with around 555,000 Nvidia GPUs, to the Grok model line developed by xAI, and the application layer through Cursor. The company also owns or has access to significant compute resources, including contracts with rival AI labs like Anthropic and Google, which lease large portions of SpaceX’s supercomputing capacity.

Despite this comprehensive ownership, the core AI model—critical for performance—remains underperforming, with reports indicating it operates at only about 11% of its potential FLOPs utilization, far below the 35–45% typical of production-grade models. This inefficiency has led to internal shifts, including relocating training to new hardware and renting out existing compute capacity, highlighting ongoing challenges in model scaling and optimization.

At a glance
breakingWhen: announced June 16, 2026; deal expected…
The developmentSpaceX announced the acquisition of Cursor for $60 billion, completing its ownership of all AI infrastructure layers but highlighting ongoing issues with model performance.
SpaceX owns every layer of AI — the stack, the rentals, the weak link
AI Dispatch · Infrastructure & Strategy

SpaceX owns every layer
of AI now

The $60B Cursor buy completes the stack: power, compute, research, model, app, distribution. But owning every layer isn’t winning every layer — and the model is the weak one.

$60B
all-stock · Cursor
(Anysphere)
The stack, layer by layer
06
Distribution
X · Tesla · Optimus · Cursor’s developer base
Strong
05
Application — Cursor
~$4B annualized revenue · just acquired
Bought
04
Model — Grok  ← the weak link
Underdelivered vs compute; training moved to Colossus 2
Weak
03
Research — xAI
Folded into SpaceX, Feb 2026
Mid
02
Compute — Colossus 1 & 2
~555K GPUs · orbital data-center plans filed
Dominant
01
Power
On-site gas generation, built faster than utilities interconnect
Dominant
The landlord pivot — renting Colossus 1 to rivals
Colossus 1 · Memphis
220,000+ GPUs · 300 MW
xAI couldn’t parallelize Grok on its mixed H100/H200/GB200 build, so it moved training to Colossus 2 and leased the rest out.
⚠ ran at ~11% utilization — “embarrassingly low”
Anthropicthru May 2029
$1.25Bper month
Googlethru June 2029
$920Mper month
combined ≈ $26B / year in compute revenue
122
days to build the first 100K-GPU cluster
~555K
Nvidia GPUs across the Memphis site
~2 GW
total power capacity
~$18B
in silicon (phase 1 alone ~$4B)
The take

You can buy a coding app and a model team. You can’t buy the research lead that makes your foundation model the one everyone else builds on — which is why Anthropic pays Musk $1.25B/month, not the other way around. Owning every layer bought SpaceX the right to attempt the hard thing. It hasn’t done it yet.

Sources: SpaceX S-1 & SEC filings; WSJ; Reuters; CBS; TechCrunch; Forbes; Business Insider; Introl; Built In (Feb–Jun 2026). Lease figures per SpaceX filings; utilization per a reported internal xAI memo.
thorstenmeyerai.com

Implications of Total AI Stack Ownership by SpaceX

This acquisition positions SpaceX as arguably the most integrated AI conglomerate outside of China, giving it unmatched control over hardware, data, and applications. While this vertical integration is a strategic advantage, the persistent weakness of the core AI model could limit the company’s competitive edge in AI development and deployment. The move also intensifies the concentration of AI infrastructure among a few major players, raising concerns about market dominance and innovation dynamics.

AI Systems Performance Engineering: Optimizing Model Training and Inference Workloads with GPUs, CUDA, and PyTorch

AI Systems Performance Engineering: Optimizing Model Training and Inference Workloads with GPUs, CUDA, and PyTorch

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Background of SpaceX’s AI Infrastructure Expansion

Over recent years, SpaceX has invested heavily in building a vertically integrated AI ecosystem. Its Colossus supercomputers in Memphis, built at a cost exceeding $4 billion, set industry records for rapid deployment and scale, with the capacity to train massive models on hundreds of thousands of GPUs. The company has also secured lucrative contracts with rival AI labs like Anthropic and Google, leasing large portions of its compute capacity—an unusual move that reflects both the underutilization of its hardware and strategic positioning.

Previously, SpaceX’s AI ambitions centered around its xAI research division and the Grok model line, developed to bridge hardware and application layers. The acquisition of Cursor, a profitable startup with a strong foothold in AI coding, marks a significant step in consolidating all AI layers under SpaceX’s control, making it a unique entity in the AI landscape. However, despite owning the infrastructure, the core models’ underperformance remains a challenge, as industry benchmarks show.

“The rapid buildout of Colossus was a ‘superhuman’ feat of vertical integration, enabling extremely fast training cycles.”

— Jensen Huang, Nvidia CEO

NVIDIA Tesla L4 24GB PCIe Graphics ACELLERATOR HH/HL 75W GPU 900-2G193-0000-000

NVIDIA Tesla L4 24GB PCIe Graphics ACELLERATOR HH/HL 75W GPU 900-2G193-0000-000

24GB Video Memory

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Unresolved Challenges in AI Model Performance

It is not yet clear whether SpaceX can significantly improve the performance of its AI models, especially given the current low utilization rates and hardware bottlenecks. The core models, which are crucial for competitive AI capabilities, remain underperforming, and internal assessments suggest that further development and optimization are needed. The impact of these issues on future product offerings and market competitiveness is still uncertain.

Hands-On LLM Serving and Optimization: Hosting LLMs at Scale

Hands-On LLM Serving and Optimization: Hosting LLMs at Scale

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Next Steps for SpaceX’s AI Strategy

SpaceX is expected to continue investing in model training and hardware optimization, aiming to boost the efficiency and capabilities of its AI models. The company may also explore expanding its application ecosystem and further leveraging its control over infrastructure to outpace rivals. Monitoring how SpaceX addresses its core model weaknesses will be critical in assessing its future AI leadership.

Kinupute AI Server Mini PC AMD Ryzen AI 9 HX 370(12 Cores | 80 Tops | Radeon 890M), Win-11 Pro, 32G DDR5, 2T PCIE4.0 SSD, Desktop Computers Dual 2.5G LAN, Oculink/USB4.0 /Triple Display 8K/WiFi7, VESA

【High-Speed Performance AI Mini PC】 Mini PC equipped with AMD Ryzen AI 9 HX370 processor, which uses AMD's…

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Key Questions

Why did SpaceX acquire Cursor?

SpaceX acquired Cursor to own all layers of the AI stack, including a profitable application, a trained model team, and a developer distribution channel, enhancing its AI ecosystem.

What are the main challenges facing SpaceX’s AI models?

The primary challenge is the low utilization of its models, with current models operating at only about 11% of their potential FLOPs, limiting performance and scalability.

How does this acquisition affect the AI industry?

This move consolidates AI infrastructure control within a single company, potentially reducing competition and increasing reliance on SpaceX’s hardware and models, raising market concentration concerns.

Will owning all AI layers guarantee success for SpaceX?

Not necessarily. While control over infrastructure and applications provides strategic advantages, the weak performance of its core models remains a significant obstacle to achieving AI leadership.

What is the significance of the low utilization rates?

Low utilization indicates inefficiencies in training and hardware deployment, which could hinder the development of more advanced and competitive AI models.

Source: ThorstenMeyerAI.com

You May Also Like

China: The Visible Hand

China’s government actively directs AI, robotics, and industrial policy through top-down planning, with state-owned capital and institutions leading the effort.

Some Reasons Why Google Had Such A Bad Day

Google faced multiple setbacks today, including technical issues and user dissatisfaction, impacting its services and reputation.

The Simple Way to Tell Automation From Real Intelligence

Learn how to distinguish automation from true intelligence in AI — practical tips, real examples, and key differences explained clearly.

Five Levers, Many Hands

Global responses to AI-driven labor changes are uneven, using five key tools. The response varies by country, reflecting different social and economic structures.