xAI

xAI is a frontier AI company Elon Musk founded in 2023, developer of the Grok model family and operator of the Colossus supercluster in Memphis, Tennessee. Musk founded it after his departure from the OpenAI board (2018) and subsequent public break with the company, positioning xAI as a counter to what he characterizes as ideological drift at OpenAI and other frontier labs. In February 2026, xAI was merged into SpaceX, and on July 6, 2026 it was rebranded SpaceXAI (Source: gizmodo.com); xAI is also the named plaintiff in xAI v. Colorado, the first full First Amendment challenge to a state AI anti-discrimination statute.

Snapshot

Valuation

Revenue and losses (SpaceX AI segment, per S-1)

Compute

Overview

xAI's mission statement and Grok's system prompt emphasize "truth-seeking" and willingness to state "politically incorrect" conclusions when "well substantiated." In legal filings, xAI has formalized this posture into a constitutional theory: that AI training-data curation, system-prompt drafting, fine-tuning, and guardrail design are expressive editorial acts protected by the First Amendment.

In February 2026, SpaceX and xAI merged. SpaceX filed its S-1 offering statement on the evening of May 20, 2026; the prospectus describes the company as an AI competitor rather than a rocket maker, a positioning the merger had implied. Fortune characterized SpaceX as having "essentially re-invented itself from a commercial space pioneer" into "an AI-centric player" against hyperscalers including Microsoft, Google, OpenAI, and CoreWeave (Source: fortune.com). The S-1 disclosed revenue, losses, and a total-addressable-market framing recorded in the Snapshot above. The offering, expected mid-June 2026 at a targeted valuation of $1.5 trillion or more, is the largest public-market test of Musk's "$26.5T AI TAM" framing to date; it is a reference point for AI Bubble Debate and, alongside Musk-controlled SpaceX–xAI cross-investments, for circular AI financing. The S-1 makes the AI-infrastructure side of the Anthropic–SpaceX Colossus 1 compute lease (disclosed by Anthropic in January 2026) visible to public-equity investors for the first time.

On June 4, 2026, S&P Dow Jones Indices decided to keep its profitability rules for index inclusion, blocking a fast S&P 500 entry for SpaceX as an unprofitable AI-heavy firm (Source: arstechnica.com).

On July 6, 2026 the xAI brand was retired and the division was renamed SpaceXAI, formalizing the February 2026 merger into the by-then publicly traded SpaceX. Gizmodo reported that the AI division burned $6.4 billion in the prior year and anchors Musk's pitch of a $28.5 trillion addressable market that includes orbital data centers (Source: gizmodo.com).

Models — Grok family

Grok models are distinctive for X-native deployment giving direct access to a large real-time data stream; comparatively permissive guardrails relative to OpenAI and Anthropic; and an explicit "truth-seeking, non-partisan viewpoint" system-prompt framing.

xAI launched Grok 4.3 on May 4, 2026 with improved reasoning, lower per-token pricing than Grok 4.20, and a Custom Voices and Voice Library product suite (Sources: docs.x.ai; artificialanalysis.ai). On May 2, 2026 xAI began rolling Grok Voice mode into Apple CarPlay, extending its consumer reach into automotive infotainment in parallel with the broader AV deployment ramp (Source: 9to5mac.com).

On July 1, 2026, xAI announced Voice Agent Builder in beta, a no-code platform for configuring production voice agents on Grok Voice. The platform bundles telephony, knowledge retrieval, tools, guardrails, and MCP support, ships with more than 80 built-in voices plus brand-voice cloning from roughly two minutes of audio, and is priced at $0.05 per minute of audio plus $0.01 per minute for telephony (Source: x.ai). See Agentic AI.

Per an internal memo disclosed July 7, 2026, SpaceXAI planned to release Grok 4.5 — its first model trained with Cursor data — as soon as Wednesday, July 8; Musk said beta testing showed performance comparable to Claude Opus and GPT-5.5 (Source: gizmodo.com).

SpaceXAI released Grok 4.5 on July 8, 2026 as planned — its first model since going public — jointly introduced with soon-to-be subsidiary Cursor and built for coding, agentic tasks, and knowledge work. Musk called it "an Opus-class model, but faster, more token-efficient and lower cost," priced at $2 per million input tokens and $6 per million output tokens versus $5/$25 for Claude Opus 4.8; published benchmarks showed it competitive with but just short of best-in-class models. It launched in Grok Build, Cursor, and the SpaceXAI console, with public availability July 9, and was not initially available in the EU (Sources: axios.com; bloomberg.com; techcrunch.com). See Grok (xAI). On July 10, 2026, Musk directed Tesla staff by memo to move to Grok, citing Grok 4.5's lower token costs — an internal steer that followed Tesla's June memo capping employee AI spending at $200 per week (Source: theinformation.com).

In court filings around May 4, 2026, Musk acknowledged that xAI used OpenAI models in training Grok, reigniting the model-distillation dispute between the two labs (Source: theverge.com). The Nature subliminal-learning paper (Language models transmit behavioural traits through hidden signals in data — Cloud, Le et al. (Anthropic et al., Nature, April 2026)) reports that distillation between a teacher and a student sharing base models can transmit behavioral traits as well as capability, which makes provenance tracking relevant to safety as well as intellectual property. See Musk v. Altman (and OpenAI / Microsoft / Brockman).

Image generation

xAI made Imagine Image 2.0 generally available on August 7, 2026 as the new Quality Mode on grok.com/imagine and in its iOS and Android apps, with API access described as coming soon. The release adds a magic wand that edits a pointed-at region and leaves the rest untouched, segmentation for selecting precise areas, background removal that exports a subject on a transparent background, multi-reference editing accepting up to five input images in a single generation, and a smart resize that recomposes an image into a chosen aspect ratio. xAI states the model ranks second in the world in both text-to-image generation and image editing, citing the Arena Image Edit and Text-to-Image leaderboards as of August 7, 2026; on both leaderboards as reproduced by xAI, OpenAI's gpt-image-2 ranks first (Source: x.ai). Grok's image generation is also the subject of the child-safety litigation and S-1 risk disclosure recorded under Controversies below.

Grok Build (agentic coding CLI)

xAI launched Grok Build on May 25, 2026, an early-beta coding agent and CLI for SuperGrok and X Premium Plus subscribers. It is xAI's first explicit entry against Anthropic's Claude Code (reported to drive roughly $14B in ARR for Anthropic) and OpenAI's Codex CLI (more than 1M developers in its first month) (Sources: x.ai; devops.com).

At $0.20 per million input tokens, Grok Build is priced roughly 60% below GPT-5.5-class agentic-coding token pricing and roughly 85% below Opus 4.7 Claude Code retail pricing. Its local-first design addresses the Microsoft–Claude Code de-adoption dynamic (per the May 25 Verge report cited in AI Bubble Debate), under which token-cost-driven enterprise cancellations are less likely at xAI's pricing point. Grok was previously absent from the agentic-coding market, where Claude Code's $14B ARR is the largest single-product ARR figure reported in 2026 enterprise AI.

The local-first description in the launch materials (May 25, 2026) was contradicted by security research in July 2026. On July 12, 2026 the researcher Cereblab reported that Grok Build had been uploading users' entire Git repositories — including previously deleted secrets — to a company cloud-storage bucket, without disclosure, through a behavior enabled by a server-side flag. xAI halted the uploading after the report, and by July 14, 2026 Elon Musk had promised deletion of all previously uploaded user data; Cereblab noted that the product's advertised /privacy command was not what fixed the behavior (Source: theregister.com). See AI and Privacy.

On July 15, 2026, xAI open-sourced Grok Build — including its terminal UI and its skills, plugins, MCP-server, and subagent extension mechanisms — on GitHub (Source: x.ai).

xAI, now branded SpaceXAI, expanded Grok into Google Workspace on July 24, 2026 as a free Marketplace add-on covering Docs, Sheets, and Slides, a week after shipping equivalent Microsoft 365 add-ins (Source: stocktwits.com).

Build Mode

xAI launched Build Mode for Grok on July 28, 2026 in Early Beta, exclusively for SuperGrok Heavy subscribers, on web, iOS and Android. The feature generates websites and applications from natural-language prompts (Source: news.futunn.com).

Colossus supercluster

xAI's Colossus facility in Memphis, TN is one of the largest single-site AI training deployments reported. Current scale is approximately 200,000 Nvidia H100-class GPUs (reported 2024–2025), against a Musk public target of 1M GPUs. The facility was deployed rapidly, including reported use of on-site gas turbines prior to full grid interconnect, a recurring data point in AI Environmental Impact. Its compute density allowed xAI to close the frontier gap with OpenAI and Anthropic faster than any previous new entrant. Memphis siting raised environmental-justice concerns about air quality in adjacent majority-Black neighborhoods (see Environmental compliance and siting below). Per July 22, 2026 reporting, SpaceXAI is laying groundwork for at least one new large-scale Texas data center that could match or exceed the roughly 1-gigawatt Memphis hub (Source: theinformation.com; benzinga.com).

Compute leasing and financing

SpaceX has disclosed two major compute leases at xAI data centers, and July 2026 reporting on Meta's planned cloud business described SpaceX as also leasing Colossus capacity to Reflection AI (Source: techcrunch.com). The first, disclosed by Anthropic in January 2026, is an Anthropic–SpaceX $45B / $1.25B-month lease at the Colossus 1 Memphis data center. The second, disclosed in a SpaceX SEC filing on June 5, 2026, has Google paying approximately $920 million per month from October 2026 through June 2029 for roughly 110,000 Nvidia GPUs of compute capacity at xAI data centers; Google cited "surging" demand for its Gemini Enterprise agent platform (Source: cnbc.com). The disclosure landed days ahead of SpaceX's planned IPO at a valuation above $1.75 trillion. The arrangement, in which a hyperscaler rents frontier compute from a direct model-layer competitor, is described as an instance of circular AI financing and adds Google to the set of customers underwriting xAI/SpaceX infrastructure capex.

On July 9, 2026 — the day after the Grok 4.5 release — Musk wrote on X that "I was clearly wrong about Anthropic," calling it "obviously currently the leader in AI" and pledging not to cut off the Colossus 1 hosting arrangement (300 MW at $1.25 billion per month through May 2029) (Source: techcrunch.com).

Hardware

The Wall Street Journal reported on July 1, 2026 that SpaceX showed investors a "handset-like" AI device prototype ahead of its planned IPO, described as slimmer than an iPhone, running a proprietary operating system, and integrating xAI technology; Musk called the report "utterly false" (Sources: techcrunch.com; reuters.com). See AI Hardware Devices.

Elon Musk told investors on SpaceX's first earnings call that the company will build its AI services exclusively on Nvidia systems, in an account published August 5, 2026: "We think the Vera Rubin architecture is the best architecture. We think it's the best AI computer, and we greatly value our close cooperation and partnership on many levels with Nvidia. So we're exclusive to Nvidia." Musk said SpaceX expects to end 2026 with more than 2 gigawatts of compute and to approach 10 gigawatts by the end of 2027, and that it plans to deploy Nvidia's Vera Rubin NVL72 rackscale system both on the ground and in space through its Starmind satellite programme, with Starmind launches beginning in 2027. Nvidia stock rose roughly 2% in pre-market trading on August 5; SpaceX stock fell more than 10% (Source: finance.yahoo.com). The single-supplier commitment runs against the multi-silicon strategies of Anthropic, OpenAI and Google DeepMind, each of which has added custom or second-source accelerators. See Semiconductor Supply Chain, Data Center Siting / AI Power Politics.

SemiAnalysis published an analysis on August 7, 2026 accepting the trajectory as achievable, after Musk said on the earnings call that SpaceX "conservatively" aims to deliver an incremental 6 to 8 gigawatts in 2027 alone. At roughly $50 billion per gigawatt, SemiAnalysis calculates that implies $300 billion to $500 billion of 2027 capital expenditure, and projects a path to a $300 billion annual revenue run rate for SpaceX by the end of 2027 assuming only half the incremental compute is monetised. It puts SpaceX capacity at about 2 gigawatts by the end of 2026, reports the Southaven plant growing from 27 turbines (about 495 MW) in February 2026 to 69 turbines (1.7 GW) in July 2026, and names Microsoft the likely largest offtaker on more than 10 gigawatts of binding contracts signed year-to-date. The remainder of the analysis is paywalled and was not read (Source: newsletter.semianalysis.com). The revenue projection is SemiAnalysis's own model rather than company guidance, and the $300 billion run-rate figure would exceed the AI segment's reported $2.6 billion quarterly revenue by roughly two orders of magnitude.

Environmental compliance and siting

Beyond the Memphis siting concerns at Colossus, TechCrunch reported on May 13, 2026 that xAI is running nearly 50 gas turbines unchecked at its Mississippi data center, raising environmental-compliance pressure alongside a broader data-center political backlash in May 2026 (Cleveland City Council permit rejection on May 14; a one-year moratorium in Hill County, TX; a Sen. Warren investigation; and an AOC–Sanders moratorium bill that went viral) (Source: techcrunch.com). See AI Data Centers, AI Environmental Impact.

On May 18, 2026 the Department of Justice signaled it is preparing a "possible intervention" to defend xAI's operation of unpermitted gas turbines powering data centers near Memphis, asserting it can participate "as a matter of right" in civil-rights groups' Clean Air Act suit seeking to halt the turbines, citing national security (Source: insideaipolicy.com). It is the first publicly reported instance of the Trump DOJ invoking national-security grounds to defend an AI data center's environmental-compliance posture against a civil-rights Clean Air Act suit, framing AI compute as a national-security interest at the litigation level. It pairs with the parallel Pentagon DPA framework in the defense domain. See AI Environmental Impact, AI Data Centers.

On June 22, 2026 the DOJ moved to dismiss the NAACP's Clean Air Act citizen suit against xAI over the unpermitted turbines, asserting a constitutional right to intervene in and override such citizen-enforcement actions (Source: insideaipolicy.com). The motion escalated the DOJ's posture from the May "possible intervention" to an attempt to terminate the suit on a theory that would curtail private citizen enforcement of environmental statutes more broadly.

A Reuters analysis published July 14, 2026 found that xAI has installed 59 natural-gas turbines for its Colossus 2 data center in Tennessee and Mississippi without securing permits, with the resulting pollution falling hardest on predominantly Black communities (Source: reuters.com). See AI Environmental Impact.

Congressional scrutiny followed. In a letter that became public on July 30, 2026, House Energy and Commerce ranking member Frank Pallone (D-NJ) asked Elon Musk for a visit to the Colossus data center sites and posed questions on potential pollution impacts on communities in Tennessee and Mississippi. Pallone wrote that in its meeting with committee staff the company "was -- by far -- the worst prepared, the most unwilling to engage in" — the quotation is truncated at the publisher's subscriber wall (Source: insideaipolicy.com).

SpaceXAI released Grok Bot in beta on August 11, 2026, a cloud-hosted agent that signs into a customer's existing application accounts to perform work such as drafting emails and organizing spreadsheets. It is available to SuperGrok Heavy, Cursor Ultra and Cursor Premium Teams subscribers on macOS, Windows, Linux and iOS, with Android listed as coming soon, priced at $120 per seat per month through Cursor Premium Teams and $200 per month for individuals (Source: venturebeat.com). See Agentic AI.

Litigation — xAI v. Colorado

xAI is the named plaintiff in xAI LLC v. Weiser (D. Colo. 1:26-cv-01515, filed April 9, 2026), the first full constitutional challenge to a state AI anti-discrimination statute. The complaint's theory is that every stage of AI development (training-data curation, system-prompt writing, RLHF, fine-tuning, guardrails) is expressive editorial conduct analogous to newspaper editing or platform curation; on that theory, the Colorado AI Act's reasonable-care duty is a compelled-speech regime subject to strict scrutiny.

The complaint asserts six counts: (1) compelled editorial modification, (2) viewpoint discrimination, (3) compelled disclosure, (4) Dormant Commerce Clause, (5) Due Process vagueness, and (6) Equal Protection, the last attacking CPAI's diversity / historical-discrimination carve-out as a race-conscious classification.

See the AI and the First Amendment concept page for the doctrinal stakes and Colorado AI Act (SB 24-205) and SB 25B-004 (Date Amendment) for the underlying statute.

Separately, Musk has brought suits against OpenAI alleging breach of its founding nonprofit mission (see Musk v. Altman (and OpenAI / Microsoft / Brockman)).

Government and defense contracting

xAI is one of seven companies signed to the U.S. Department of War's classified-network agreements announced May 1, 2026, alongside OpenAI, Google, Nvidia, Reflection, Microsoft, and AWS, framed under the SpaceX umbrella; Anthropic was excluded (Sources: theinformation.com; war.gov). See DOD — Department of Defense (AI Deployer), Anthropic.

SpaceX's talks to provide billions of dollars' worth of data-center capacity for the Pentagon's AI buildout became public on July 17, 2026 (Source: wsj.com), extending the compute-leasing business that already serves Anthropic, Google, and Reflection AI to a government customer and competing for territory held by CoreWeave and Amazon (Source: reuters.com).

Policy positions

xAI's litigation theory, that model outputs and pre-output design choices are protected expressive conduct, is the most aggressive such claim tested in court to date. The company is aligned with the industry effort, alongside America's AI Action Plan and Executive Order 14365 — Ensuring a National Policy Framework for AI, to preempt state AI regulation. Musk signed the 2023 FLI Pause Letter while operating one of the industry's most aggressive compute buildouts, a tension noted in AI 2027. Musk has been publicly supportive of chip export controls, though xAI itself has not been a prominent voice in the export-control policy process.

People

xAI's co-founders and leadership include:

• Elon Musk — founder, principal public voice, named plaintiff in the Colorado suit.
• Igor Babuschkin — co-founder, engineering leadership (ex-DeepMind, ex-OpenAI).
• Jimmy Ba — co-founder, research (U. Toronto, deep-learning theorist).
• Greg Yang — co-founder, theoretical foundations.

Per TechCrunch (May 14, 2026), the merged SpaceXAI entity has lost more than 50 researchers and engineers since the February 2026 SpaceX–xAI merger, at least 11 to Meta and 7 to Mira Murati's Thinking Machines Lab; the pre-training team was reported down to a handful of people after lead Juntang Zhuang's exit, against a backdrop where Anthropic and OpenAI continued to net-add researchers (Source: techcrunch.com). Forbes reported on May 15, 2026 that co-founder Igor Babuschkin is in talks to raise up to $1 billion for a new AI startup, the most senior co-founder departure since the merger (Source: tbpn.substack.com). An Igor Babuschkin page would be warranted if the new venture launches publicly.

Controversies

Beyond the environmental-siting and litigation matters covered above:

• Grok outputs: periodic high-profile incidents of Grok producing violent, antisemitic, or conspiracy-coded content (for example, the "MechaHitler" episode referenced in Sycophancy and Hallucination).
• Nonconsensual deepfakes: a WIRED investigation published June 11, 2026 found Grok still hosting sexualized, nonconsensual deepfakes of celebrities and U.S. Rep. Alexandria Ocasio-Cortez months after the company pledged safeguards. The same day, Canada's Privacy Commissioner issued preliminary findings that xAI violated federal privacy law by lacking "appropriate safeguards from the outset" — landing days before SpaceX's June 12 IPO (Source: wired.com). See AI and Privacy.
• Grok-generated child sexual abuse material: xAI faces a series of lawsuits brought by young girls against X and xAI over Grok-generated child sexual abuse material. In one suit reported on or about July 8, 2026, a man allegedly used Grok to generate roughly 7,000 sexually explicit images of his minor stepdaughter while xAI reported only a single prompt to authorities (Source: arstechnica.com). On July 15, 2026 xAI itself sued a South Carolina man for allegedly using Grok to generate such material — reported as likely the first suit by an AI lab against a user over generated content — with the complaint stating xAI suspended more than 52,000 accounts and made over 73,000 NCMEC reports in 2026 (Source: reuters.com). See Grok CSAM suits (Doe plaintiffs v. X / xAI).
• Music-labels dispute resolved: On July 17, 2026, X and the major record labels ended their dispute over music on the platform (Source: reuters.com).
• Training-data scraping: use of X user content for training, raised under copyright and privacy lenses.

Relationships

• instance-of: AI Race Dynamics, AI and the First Amendment
• depends-on: AI Environmental Impact (Colossus Memphis supercluster context), Semiconductor Supply Chain
• contradicts: Colorado AI Act (SB 24-205) and SB 25B-004 (Date Amendment) (via xAI LLC v. Weiser — Complaint (D. Colo. 1:26-cv-01515)); the premise of state-level AI anti-discrimination regimes generally
• related: OpenAI (origin and ongoing litigation counterparty), Elon Musk, Texas Responsible AI Governance Act (TRAIGA / HB 149) — Source Summary (parallel federalism axis), AI Environmental Impact
• supports: Executive Order 14365 — Ensuring a National Policy Framework for AI, America's AI Action Plan — pro-deployment, anti-state-regulation posture

See also

Primary: xAI LLC v. Weiser — Complaint (D. Colo. 1:26-cv-01515), Elon Musk, AI and the First Amendment, Colorado AI Act (SB 24-205) and SB 25B-004 (Date Amendment). Also referenced on AI Environmental Impact, FLI — Pause Giant AI Experiments: An Open Letter, AI 2027, Executive Order 14365 — Ensuring a National Policy Framework for AI, Sycophancy and Hallucination, EU General-Purpose AI Code of Practice (Final Version, 2025), Frontier AI Safety Commitments (Seoul, 2024), Situational Awareness: One-Year-Later Retrospectives.