ELON, EXPLAINED
Investor Brief · 2026 H1

MACROHARD, in brief.

SpaceXAI's bet to run a software company entirely with AI agents

$3.20B
SpaceX AI segment revenue (2025)
SEC: AI segment revenue of $3,201M in 2025, the reporting segment that houses Grok and MACROHARD
$12.73B
AI segment capex (2025)
SEC: AI segment capital expenditures of $12,727M, versus $3,832M Space and $4,178M Connectivity, the compute-buildout signal
$(6.36)B
AI segment operating loss (2025)
SEC: loss from operations of $(6,355)M; Segment Adjusted EBITDA of $(1,237)M, an early-stage, pre-product economic profile
~1.3B accounts / ~550M MAU
Grok + X reach
SEC: ~1.3B supported accounts in the LTM and ~550M monthly active users as of March 31, 2026, the distribution layer
~555,000 GPUs / ~2 GW
Colossus 2 compute base
NVIDIA GB200/GB300 cluster in Memphis with MACROHARD on the roof; ~555k GPUs and ~2 GW by Feb 2026 per Grokipedia and Elon Musk on X
~$250B (reported)
xAI acquisition value
Reported all-stock value of xAI at SpaceX's Feb 2026 acquisition; private-company figure, label as reported, not audited fact

Overview

MACROHARD, also called Digital Optimus, is SpaceXAI's pre-product effort to emulate the function of an entire software company with AI agents that operate real software through screen, keyboard, and mouse. Elon Musk frames it as the natural attack on firms like Microsoft, whose output is purely digital, and has called it over time probably the most important project at the company. For an investor, MACROHARD sits inside SpaceX's AI reporting segment, which the S-1 shows generating $3.2B of 2025 revenue against $12.7B of capital expenditure, so the brief is a bet on a very large software market funded today by very large compute spending.

The bull case

  • The target market is the largest on Earth: enterprise software and the knowledge work it serves. Microsoft alone reported $281.7B of FY2025 revenue (Productivity and Business Processes $120.8B, Intelligent Cloud $106.3B, More Personal Computing $54.6B), and MACROHARD aims at the labor the software supports rather than the software seats.
  • Distribution is already built. SpaceX's S-1 reports the combined Grok and X platform at roughly 1.3 billion supported accounts and about 550 million monthly active users, with X Money rolling out as the in-app payments layer, so a finished agent ships into an audience competitors must still acquire.
  • The compute backbone is vertically integrated and self-funded. Colossus 2 in Memphis runs roughly 555,000 NVIDIA GPUs at about 2 GW, and SpaceX is pairing it with TERAFAB, a Tesla and SpaceX fab targeting over a terawatt of compute per year, which would reduce dependence on third-party foundry supply.
  • The inference cost structure is differentiated. Elon Musk has said MACROHARD will run on the roughly $650 Tesla AI4 chip already in millions of vehicles, paired with frugal use of expensive NVIDIA hardware, an asset-light inference path no rival agent company can replicate at fleet scale.
  • The dual-mind architecture maps to a clear division of labor: Grok as the System 2 deliberate reasoner deciding what to do, and a Tesla-built agent as the System 1 executor that drives the screen and keyboard, with recursive self-improvement where current agents help train the next generation.
  • It rides a documented cost-decline curve. SpaceX produced an exponential fall in launch cost per kilogram, and the same compounding shape in AI inference cost is the lever that turns full company emulation from a demo into a margin-positive product over time.

The bear case

  • It is unproven and pre-product. As of mid-2026 MACROHARD, Digital Optimus, and the Grok Computer agent are in preparation, not generally available, so there is no shipping revenue line attributable to MACROHARD and no external proof that full digital emulation generalizes beyond demos.
  • Capital intensity is extreme and front-loaded. The AI segment posted $12.7B of 2025 capex against $3.2B of revenue, a $(6.36)B operating loss, and negative Segment Adjusted EBITDA, so the economics depend on a payoff that has not yet arrived.
  • Compute supply is a hard constraint. The thesis leans on Colossus 2 GPU deliveries from NVIDIA and on TERAFAB, an unbuilt fab Grokipedia estimates at $20B to $25B, and any slip in either caps the training and inference scale MACROHARD needs.
  • The competition is well capitalized and shipping. Microsoft Copilot, OpenAI Operator and agents, Anthropic Claude computer use, and Google and Salesforce agents are already in market, and full company emulation is downstream of artificial general intelligence that may not arrive on the stated schedule.
  • Microsoft's distribution and switching moat is real. Decades of compliance certifications, sticky enterprise contracts, and entrenched IT estates slow customer migration even if MACROHARD ships and works.
  • Key-person and structure risk is concentrated. The project was reorganized through 2026, leadership changed, and a meaningful part of the valuation rests on an Elon Musk execution premium that has been right on SpaceX and uneven on certain Tesla timelines.

Catalysts to watch

The economics

MACROHARD has no standalone unit economics yet; it is funded inside SpaceX's AI segment, which in 2025 generated $3.2B of revenue, a $(6.36)B operating loss, and $12.7B of capex, a profile of a compute-heavy build well ahead of monetization. The stated cost edge is structural: training concentrates on Colossus 2, while inference is meant to run on the roughly $650 Tesla AI4 chip across the existing fleet, shifting the marginal cost of an agent toward already-deployed, already-powered hardware. On framing, the SpaceX S-1 cites a near-term total addressable market of about $5.7 trillion, including roughly $740B for AI infrastructure and $2.4T for consumer subscriptions, against a reported xAI acquisition value near $250 billion and a June 2026 IPO that implied roughly $1.78 trillion for SpaceX, with the AI-infrastructure and software opportunity sized larger than the launch business that built the company.

The bottom line

MACROHARD is the highest-ceiling, lowest-proof line in the SpaceX story: a credible attack on the largest software and knowledge-work market on Earth, carried by a built distribution layer in Grok and X, a self-funded compute backbone, and a genuinely differentiated low-cost inference path. The counterweight is that it is still pre-product while the AI segment burns billions, so the investor question is timing and execution rather than ambition. Watch the first public agent release and the post-IPO AI-segment disclosures as the signals that turn the thesis from narrative into numbers.

Sources

This is an educational brief, not investment advice and not a recommendation to buy or sell any security. Figures trace to primary filings, official statements, and Grokipedia; privately held valuations are labeled as reported or estimated.

Keep exploring: Explore all · State of Elon · Net Worth Tracker · Glossary