PublicNASDAQ: SPCXFounded 2023🇺🇸Palo Alto, CaliforniaCEO Elon Musk (division president: Michael Nicolls)
xAI is Elon Musk's artificial-intelligence company, founded in March 2023, acquired by SpaceX on 2 February 2026, and rebranded SpaceXAI on 6 July 2026.
Latest update Oct 19, 2026
01
Grok 4.7 is the current flagship in Cursor, Grok Build and the Grok API at $2 and $6 per million input and output tokens. On 18 September 2026 SpaceXAI released Grok Voice Transcribe 2.0, ranking first among 32 streaming models on Artificial Analysis at unchanged $0.10/$0.20 per hour. Official docs now bill X Search at $5 per 1,000 posts and $10 per 1,000 profiles. On 25 September Musk gave a founder timetable putting Colossus at about 780,000 GPUs now and as many as 1.44 million by late December if three 220,000-GB300 waves land; Bloomberg and Tom's Hardware reported the post, and x.ai still has no matching news item or later SEC restatement. Combined SPCX last $167.60 (about $2.275T on 13.57B shares). Grok MAU remains the March 2026 ~117 million IPO figure. Grok 5 is still unreleased on x.ai/news. Google's ~110,000-GPU Colossus 2 lease had a 30 September full-delivery test; as of 18 October the one-month grace was still running through 31 October, with no public full-count, reduced-count, or termination disclosure and no SpaceX 8-K after 14 August. On 2 October the Eighth Circuit granted an injunction pending appeal in SpaceXAI LLC v. Ellison, No. 26-2806. On 5 October CBS reported SpaceXAI answered the NYC Council subpoena with a letter saying it wants to work with the city but would not attend. On 13 October the New York City Council said SpaceXAI agreed to testify on 16 October after the Council said it would petition the New York State Supreme Court to enforce the subpoena. On 16 October Gabriela Castillo Madrid, senior manager of global government affairs, testified. The company's own letter was not published in those Council releases. On 14 October 2026 xAI released Grok Voice 3.0 in the Voice Agent API at $0.08 per minute. Artificial Analysis's speech-to-speech page lists it at an index score of 86.9. xAI says the model handles millions of Tesla conversations a day. Tesla's support page describes in-vehicle Grok and does not name Grok Voice 3.0.
council.nyc.govUpdated Oct 18, 2026
Analysis · 7 min read · Aug 7, 2026The long view on xAI: strategy, technology and what to watch02
2026-2027
After US District Judge Donovan Frank denied a preliminary injunction on September 4, 2026, xAI noticed an appeal to the Eighth Circuit (docket 26-2806) seeking to block Minnesota from enforcing its anti-nudification law, which carries fines of up to $500,000 per violation, while the First Amendment challenge proceeds. No Eighth Circuit ruling had been published as of September 16, 2026.
H2 2026
xAI opened a public beta of Grok Bot on August 11, 2026, then expanded plan coverage on August 21 and August 26. On September 3, 2026 it made Grok Bot available for enterprises, adding access, network, and audit controls, with two weeks of free usage for Grok and Cursor Enterprise customers (including inviting people without an existing seat). xAI said thousands of organizations have adopted Grok Bot since launch. Android support remains 'coming soon' with no firm date, and as of September 9, 2026 xAI has still given no confirmed date for Grok Bot exiting beta on every consumer surface.
H2 2026
Release the ~6-trillion-parameter Grok 5, still training on Colossus 2 after missing Musk's original Q1 2026 window and a subsequent Q2 window; Musk has claimed a '10% and rising' chance it constitutes AGI. As of September 22, 2026 there is still no confirmed release date and no model card. xAI instead shipped Grok 4.5 (July), Grok 4.6 (August 12), and Grok 4.7 (now available in Cursor, Grok Build and the Grok API). Independent trackers still treat a 2026 Grok 5 ship as possible but unconfirmed.
H2 2026
Following the August 14, 2026 close of the $60B Anysphere acquisition, deepen the joint SpaceXAI-Cursor roadmap: Cursor (50,000+ businesses, including 64% of Fortune 500 companies, and roughly $3B in annualized revenue) now runs on SpaceX's GPU fleet including Colossus. The integration produced Origin, a GitHub-rival code-hosting platform, on August 18; Grok 4.6 shipped in Cursor and GitHub Copilot the same week; and on September 3 Grok Bot became available to Cursor Enterprise customers with organization-wide invites. Remaining work in H2 2026 is the gradual Cursor rebrand and deeper Grok Build / Grok Bot / API sharing.
2026
TargetWeek of 28 Sep 2026 (Musk: next week from 25 Sep)
2026
TargetOne-month Google grace after Sept 30, 2026
Oct 2026
Ramp ~110,000 GPUs of Colossus 2 capacity for Google's $920M/month lease, which runs October 2026 through June 2029 (~$30B). Colossus 2 became operational September 3, 2026, and Google's access is still ramping through September at a reduced fee; per SpaceX's own regulatory filing, if the full committed GPU count is not delivered by September 30, 2026, Google can terminate after a one-month grace period or accept a reduced GPU count at a lower fee, and either party may otherwise exit on 90 days' notice after December 31, 2026 - making the remaining 14 days the near-term risk to the Q2 2026 AI-segment revenue beat. SpaceX has not publicly confirmed that the full GPU count has been delivered.
Nov 2, 2026
Official xAI release notes say that on November 2, 2026 the grok-imagine-image-quality API slug is retired. Requests to that slug will be served by grok-imagine-image-2.0 with quality set to low, at a lower per-image price, with no change to the request or response shape. The grok-imagine-image (1.0) slug is not affected.
Q1-Q2 2027
Build out the Memphis-area campus toward the stated 1M+ GPU, ~2 GW target: Colossus 3 ('MACROHARDRR' in Southaven, Mississippi, a $20B+ investment purchased December 2025) and a newly confirmed fourth building, 'Minihard' (announced late July 2026, ~220,000 GB300 GPUs), join the original Colossus 1 and 2 structures on the Tulane Road campus. Trackers now put the campus at close to 1M GPUs by Q1-Q2 2027, while SpaceX secures power for a complex whose demand rivals the entire Memphis grid peak.
2027
Elon Musk confirmed on August 29, 2026 that SpaceX is building a foundry in Bastrop, Texas (adjacent to its Starlink factory, on roughly 830 acres quietly acquired between March and June 2026) to cast the nickel-superalloy turbine blades and vanes that are the most bottlenecked part of the gas-turbine supply chain, with major suppliers like GE Vernova sold out of capacity through 2030. Musk said in-house casting can cut up to 18 months off the time to bring new natural-gas turbines online, addressing the power crunch that has constrained buildouts including xAI's Colossus site, though the plan has also drawn early pollution scrutiny given Colossus's own turbine controversy in Memphis.
Jul 2027
Under a July 31, 2026 agreed order with the Mississippi Department of Environmental Quality, phase out the roughly 69 temporary gas turbines powering Colossus as a permitted 1.2 GW natural-gas plant (41 turbines, authorized under a March 2026 Clean Air Act permit) comes online. Removal began as early as August 2026, but construction of the permanent plant is running behind schedule on supply-chain delays for its turbines, and regulators have set a December 31, 2026 deactivation deadline for some units alongside the original July 2027 full-removal date for the rest. The underlying NAACP Clean Air Act lawsuit over the unpermitted units remains unresolved: a preliminary-injunction hearing that could have forced an earlier shutdown was scheduled for August 24, 2026 but was postponed per an August 21 court notice, with no new date set as of late August, while the Department of Justice separately continues pressing to intervene and dismiss the suit on national-security grounds.
Q4 2027
On SpaceX's August 4, 2026 earnings call, Elon Musk said SpaceX and xAI will go exclusively with Nvidia GPUs and Vera CPUs for AI compute going forward ('we're exclusive to Nvidia'), calling the Vera Rubin architecture the best available platform. Beyond terrestrial data centers, SpaceX and Nvidia are jointly developing Starmind AI1, the first of a planned constellation of orbital AI-compute satellites, carrying a full Nvidia Vera Rubin NVL72 rack (72 GPUs, ~250kW peak/175kW average power via 75-meter solar panels and 30-meter radiative cooling panels). Prototype testing is targeted for early 2027, with mass production at a new 'Gigasat' facility and commercial launches at scale targeted for Q4 2027 and beyond, as part of a stated plan to scale installed AI compute to ~2 GW by end-2026 and roughly 10 GW by end-2027.
Sep 21, 2026
Official xAI docs say that from September 21, 2026 at 12:00 PM PT, X Search will be billed at $5 per 1,000 posts fetched and $10 per 1,000 user profiles fetched instead of $5 per 1,000 calls. Parent and quoted posts count toward the post total. The change is posted as a live warning on the pricing page; xAI has not published a developer migration note beyond that notice.
03
On Oct. 14, 2026 xAI released Grok Voice 3.0, a speech-to-speech model, in its Voice Agent API at $0.08 per minute, the same price as Grok Voice 2.0. The company said grok-voice-latest now points to the new model and that grok-voice-2.0 is deprecated. xAI said the model scored 86.9 on Artificial Analysis's Speech-to-Speech Index and that Grok Voice handles millions of conversations a day in Tesla vehicles. Those benchmark and usage figures are the company's.
On Oct. 5, 2026, OpenAI policy lead Morgan Dwyer and Anthropic Frontier Red Team head Logan Graham testified under oath at a New York City Council Committee of the Whole hearing on AI risks, with Google and Meta policy officials also present. CNBC reported Speaker Julie Menin said SpaceXAI was the only summoned lab absent and called that a direct violation of last week's subpoena; CBS said SpaceXAI sent a letter saying it would not attend and that the council is going to court. The council is weighing about 10 AI bills, including third-party validation and whistleblower incentives.
www.cnbc.com www.cbsnews.com www.upi.com council.nyc.gov + OpenAI+ Anthropic
On Oct. 2, 2026, the U.S. Court of Appeals for the Eighth Circuit granted xAI an injunction putting Minnesota's HF 1606 anti-nudification law on hold while the company appeals a Sept. 4 district-court denial. Reuters, via MPR News, and Law360 reported the Friday order is temporary and does not decide the merits. The Aug. 1 law bars covered operators from allowing users to create realistic intimate images of an identifiable person that were not in the original photo. xAI says Grok Imagine has protections against creating nudified images of real people; Minnesota had urged the court to leave the law in force.
On September 29, 2026 President Donald Trump and executives from OpenAI, Anthropic, Google, Nvidia and SpaceXAI (xAI) signed the White House Accord on Super Intelligence Joint Commitment on Frontier Responsibilities, a voluntary four-layer plan covering internal model controls, an internal team, independent external audits and a board committee. Trump the same day signed an executive order directing federal agencies to use Super Intelligence and SI in place of Artificial Intelligence and AI. The accord is described as morally binding and says the steps may later be written into law; Meta CEO Mark Zuckerberg also signed and is described here without a catalog entry.
www.whitehouse.gov www.washingtonexaminer.com www.bbc.com + OpenAI+ Anthropic+ Google DeepMind+ Nvidia
Elon Musk said on X on September 25 that Colossus 1 now holds 150,000 Nvidia H100, 50,000 H200 and 30,000 GB200 chips, while Colossus 2 holds 110,000 GB200 and 440,000 GB300 chips. He said another 220,000 GB300 chips will be fully operational next week, 220,000 in November, and a third 220,000 by late December if the company gets lucky, which would take the combined fleet to about 1.44 million GPUs. Bloomberg called the post the most detailed timetable yet for the Memphis-area cluster; the December wave is conditional and the counts are Musk's, not an independent inventory.
SpaceXAI said Grok 4.7 is now available in Cursor, Grok Build, and the Grok API at the same $2 and $6 per million input and output tokens as Grok 4.6, with a faster variant at twice the output speed and twice the price. The company said the model uses a larger base than Grok 4.6, ran longer reinforcement learning on multi-hour tasks, and ships a new safeguard stack that scored 62.4% on LatchBio's biosafety benchmark and let 3.3% of risky HackerBench prompts through. The release lands after Grok 4.7 slipped past its mid-September target.
SpaceXAI released Grok Voice Transcribe 2.0, saying internal evaluations show it is twice as accurate as version 1.0 at the same price and that it ranks first for accuracy among 32 streaming models on the public Artificial Analysis leaderboard. Pricing stays at $0.10 per hour of audio for batch transcription and $0.20 per hour for streaming, with speaker diarization, word-level timestamps, and key-term biasing included; Atlassian said the model now transcribes Loom videos after outperforming its prior system. Version 2.0 will soon become the Speech-to-Text API default, with 1.0 deprecated in the coming weeks unless developers pin grok-voice-transcribe-1.0.
X Corp and SpaceXAI moved on September 14 to dismiss their antitrust claims against Apple in a federal case alleging Apple and OpenAI locked up smartphone and chatbot markets. The filing did not say whether a settlement was reached. Claims against OpenAI continue.
xAI's official developer pricing page now warns that from September 21, 2026 at 12:00 PM PT, X Search will be billed at $5 per 1,000 posts fetched and $10 per 1,000 user profiles fetched, replacing the current $5 per 1,000 calls. Every post returned by a search or thread fetch, including parent and quoted posts, counts toward the post total. Web Search and other server-side tools stay on the existing per-call rates.
SpaceXAI apologized after Grok went down for about 3.5 hours from 6:30 a.m. PT on September 3, 2026 across X, iOS and Android. The company said the Memphis models cluster outage also affected unnamed compute partners and that it is taking corrective action. Elon Musk said the team would address the failure; the company did not name other chatbots as sharing the same root cause.
US District Judge Donovan Frank denied xAI's request for a preliminary injunction against Minnesota's HF 1605 anti-nudification law, MPR News reported September 4, 2026, so the state can keep enforcing fines of up to $500,000 per violation while the First Amendment case proceeds. The law took effect August 1. xAI has noticed an appeal to the Eighth Circuit.
Elon Musk confirmed on August 29, 2026 that SpaceX is building a foundry in Bastrop, Texas to cast the nickel-superalloy blades and vanes used in gas turbines, bringing the most bottlenecked part of the turbine supply chain in-house. Musk said in-house casting can accelerate new natural-gas turbines coming online by up to 18 months, addressing a shortage that has left major suppliers like GE Vernova sold out of capacity through 2030 and that has constrained power for AI data centers including xAI's Colossus. SpaceX quietly acquired roughly 830 acres near its existing Starlink factory in Bastrop between March and June 2026.
xAI's Grok 4.6 became available as a fully managed model in Model Garden on Google Cloud's Gemini Enterprise Agent Platform, in public preview with a 500,000-token context window and configurable reasoning-effort settings. The listing extends Grok 4.6's enterprise reach beyond its prior rollouts on GitHub Copilot and Amazon Bedrock earlier in August, giving Google Cloud customers a third major managed route to the model alongside xAI's own API.
xAI's Grok 4.6 reached general availability on Amazon Bedrock on August 19, one week after its initial launch, with AWS adding cross-region inference support across every commercial Bedrock region. The frontier coding and agentic model offers a 500,000-token context window and four configurable reasoning-effort settings, moving Grok from a single endpoint into AWS's standard enterprise routing.
Security researcher Rony Utevsky of Adversa AI disclosed a technique called cryptographic context injection, where a web page embeds an encrypted malicious instruction plus its decryption key so Grok's guardrail scanner cannot read it, but Grok itself decrypts and executes it, exfiltrating a user's name, approximate location, subscription tier and chat history to an attacker's server. xAI was first notified on June 3 and again on August 4 and August 10, but had not patched the flaw as of August 19; researchers say the same technique now works only unreliably against Google's Gemini, suggesting Google has already hardened its filters.
04
xAI's frontier LLM line. Grok 4.7 is SpaceXAI's most capable model for coding and knowledge work, using a larger base than Grok 4.6 with a longer reinforcement-learning run on multi-hour tasks; Grok 4.6 (August 12, 2026) remains the prior flagship; Grok 4.1 Fast offers a 2M-token context window for agentic tool use.
Grok's consumer surface: the grok.com site and iOS/Android apps, deep integration into the X platform for its ~600M monthly users, and in-vehicle Grok voice assistant in Tesla cars - a distribution moat no rival lab has.
xAI's image-and-video generation product, built into the Grok apps and API. Grok Imagine Video 1.5 launched May-June 2026 and topped the Image-to-Video Arena leaderboard on release; Musk declared the product 'done' on July 5, 2026, and a July 31 update added native 1080p output, voice/face-consistent references and up to seven-image multi-reference scene control. On August 7, 2026 xAI shipped Grok Imagine Image 2.0, rebuilding the image side around editing tools - a region-specific magic wand, precise segmentation, background removal, and multi-reference editing accepting up to five input images per generation.
The Memphis-area AI training complex built in 122 days in 2024, now ~555,000 Nvidia GPUs (H100/H200/GB200) across Colossus 1 and the gigawatt-class Colossus 2, targeting 1M+ GPUs and ~2 GW - with a third facility, Colossus 3 ('MACROHARDRR' in Southaven, Mississippi, a $20B+ investment), under construction. xAI trains Grok 5 on Colossus 2 while leasing Colossus 1 to Anthropic and capacity to Google - though its temporary gas turbines are the subject of an NAACP emissions lawsuit and a DOJ national-security intervention.
A suite of Grok frontier models and tooling (Deep Search, Tool Use) packaged for US national-security, federal, state and local customers - spanning a DoD contract, GSA-wide availability, and embedding in the Pentagon's internal GenAI.mil platform at Impact Level 5.
The AI-powered code editor and coding-agent platform built by Anysphere, acquired by SpaceX for $60B in an all-stock deal that closed August 14, 2026 and folded into the SpaceXAI division. Serves 50,000+ businesses (including 64% of the Fortune 500) at roughly $3B in annualized revenue, now running on SpaceX's GPU fleet including Colossus, and feeding coding data into Grok's training pipeline.
05
(Mar 2026 workforce-data estimate, up from ~1,200 in 2025); the Memphis Colossus site alone has grown to ~3,000
FY2025: $3.2B revenue, -$6.4B operating loss (combined xAI + X, per SpaceX's IPO filing). Q2 2026, the most recent disclosed quarter (reported August 4, 2026): AI-segment revenue $2.561B (+247% YoY) against a $1.26B operating loss, with AI capex of $15.83B out of SpaceX's $18.4B total Q2 capex - up from $818M revenue and a $2.47B operating loss in Q1 2026. The jump reflects the Anthropic ($1.25B/mo), Google ($920M/mo from Oct 2026, ramping at a reduced fee through September) and Reflection AI ($150M/mo from July 2026) Colossus leasing contracts beginning to contribute, worth roughly $27B+ annualized once fully ramped.
(Feb 2026 xAI merger valuation); combined SpaceX (SPCX) closed Tuesday September 15, 2026 at $143.49 (market cap about $1.948T; 13.57B shares outstanding), down from the September 11 $151.21 close and still above the $135 IPO price and the August 3 52-week low of $104.83. The third lockup expiry on September 9 freed about 319 million insider shares; the next unlock is around September 24
(Mar 2026, per SpaceX IPO filing; up from ~89M on Dec 31, 2025 per the same filing) - still the most recent official figure as of September 2026, with no newer SEC disclosure since the June IPO prospectus
(Jan-Feb 2026, ~2 GW capacity target) - still the most recent disclosed count as of September 2026. A third facility (Colossus 3 / MACROHARDRR) and a fourth building ('Minihard', ~220,000 Nvidia GB300 GPUs) are under construction toward the 1M+ GPU goal by Q1-Q2 2027
(Mar 2026, first SEC-audited figure, per SpaceX's S-1; up from 520M in Dec 2025) - below the ~600M Musk had repeatedly claimed; still the most recent official figure as of September 13, 2026, with no newer SEC disclosure since the June IPO prospectus. X and xAI combined have ~6.3M active paid subscribers (~4.4M X Premium/Premium+, ~1.9M SuperGrok tiers)
FY2025: $3.2B revenue, -$6.4B operating loss (per SpaceX's IPO filing). Q2 2026, the most recent disclosed quarter (reported August 4, 2026): AI-segment revenue $2.561B (also reported as $2.56B), up 247% YoY and above the ~$2.18B analyst consensus, on a $1.26B operating loss - driven partly by the new Anthropic, Google and Reflection AI compute-leasing agreements ramping up. AI capex reached $15.83B of SpaceX's $18.4B total Q2 capex (~86%, roughly 39% above the ~$13.2B Street estimate) and roughly 2.4x the AI segment's own quarterly revenue, spooking investors even as total company revenue ($7.8B, +92% YoY) beat estimates.
$76B+ through 2029 across three deals: Anthropic ($1.25B/mo through May 2029, ~$40B+), Google ($920M/mo from Oct 2026 through June 2029, ~$30B) and open-source lab Reflection AI ($150M/mo from July 1, 2026 through 2029, up to $6.3B, signed June 22, 2026 for GB300 capacity on Colossus 2). Colossus 2 itself became operational September 3, 2026. Google's allocation (~110K GPUs) is still ramping at a reduced fee through September 2026; if SpaceX has not delivered the full committed GPU count by September 30, 2026, Google can terminate the agreement (after a one-month grace period) or accept a reduced GPU count at a lower fee, and either party may exit on 90 days' notice after December 31, 2026.
$2 / $6 per million input/output tokens, the same as Grok 4.6; SpaceXAI also serves a faster Grok 4.7 variant at twice the output speed and twice the price. Grok 4.7 launched in Cursor, Grok Build, the Grok API, third-party coding harnesses, and cloud platforms.
(Colossus 1: 150k H100 + 50k H200 + 30k GB200; Colossus 2: 110k GB200 + 440k GB300). Planned additions: 220k GB300 next week, 220k in November, and 220k by late December if lucky, for about 1.44 million total. Founder-disclosed; December wave conditional.
As of 8 Oct 2026, no public full-count delivery, reduced-count acceptance, or Google termination. SpaceX FWP: about 110,000 GPUs, $920M/month from October 2026, one-month grace then terminate or reduced count. Latest SpaceX 8-K on EDGAR is 14 Aug 2026. x.ai news has no restatement of Musk's 1.44 million GPU timetable.
06
Operating status
Grok Voice 3.0 indexed at 86.9; NYC testimony given; Colossus grace open
On October 16, 2026 Gabriela Castillo Madrid testified before the New York City Council after the company agreed on October 13 to appear rather than wait for a state-court petition. Artificial Analysis lists Grok Voice 3.0 at a Speech-to-Speech Index of 86.9. Tesla's support page does not name that model. Google's Colossus delivery grace still runs through October 31, 2026.
Advertising on the X platform (~$3.3B+ annualized by end-2025) plus X Premium subscription tiers (~$1B annualized by Feb 2026), which bundle Grok access - the division's largest revenue base.
SuperGrok (~$30/month) and SuperGrok Heavy (~$300/month) on grok.com and the Grok apps, plus Grok entitlements inside X Premium plans.
Usage-based Grok API access (Grok 4.5 at $2/$6 per million tokens) plus 'Grok for Government' - a $200M-ceiling DoD contract, GSA-wide availability at $0.42 per agency, and Grok embedded in the Pentagon's GenAI.mil at Impact Level 5.
Renting Colossus capacity to rival AI labs: Anthropic pays $1.25B/month through May 2029 for all of Colossus 1 (~220K GPUs, 300 MW), Google pays $920M/month from October 2026 through June 2029 for ~110K GPUs at Colossus 2, and open-source lab Reflection AI pays $150M/month from July 1, 2026 through 2029 (up to $6.3B) for Nvidia GB300 capacity at Colossus 2 - over $76B contracted across the three deals.
07
How past announced dates turned out.
4/5announced dates hit on time
Ship Grok 4.7 after the mid-September slip
Target Late September 2026 · Resolved Sep 22, 2026
SpaceXAI made Grok 4.7 available in Cursor, Grok Build, the Grok API, third-party coding harnesses, and cloud platforms at $2 and $6 per million input and output tokens, with a faster variant at twice the output speed and twice the price. The company said the model uses a larger base than Grok 4.6 and ships a new safeguard stack.
Ship Grok 4.7
Target September 12, 2026 · Resolved Sep 11, 2026
The implied September 12, 2026 release did not happen. On September 11 Musk said Grok 4.7 needed a few more days because reinforcement-learning length penalties were making the model quit solvable hard tasks early and check its work too loosely. Official xAI docs still listed Grok 4.6 as the latest model as of September 13, with no new fixed date.
Await ruling on Minnesota nudification-ban injunction
Target Pending · Resolved Sep 4, 2026
US District Judge Donovan Frank denied xAI's motion for a preliminary injunction on or around September 4, 2026, allowing Minnesota to keep enforcing HF 1605 (civil fines of up to $500,000 per violation) while the First Amendment case proceeds. xAI noticed an appeal to the Eighth Circuit.
Deploy Grok across the Pentagon's GenAI.mil
Target 2026 · Resolved Aug 31, 2026
On August 31, 2026 the Pentagon launched Starshield AI's Grok for Government on GenAI.mil at Impact Level 5, alongside OpenAI's ChatGPT Mil, giving about 3 million DoD personnel access. The portal had already onboarded more than 1.7 million unique users.
Close the $60B Cursor (Anysphere) acquisition
Target Q3 2026 · Resolved Aug 14, 2026
SpaceX's acquisition of Cursor-maker Anysphere became effective August 14, 2026 per a regulatory filing, well ahead of the Q3 2026 target, folding Cursor into the SpaceXAI division as a subsidiary with access to the Colossus GPU fleet for AI training.
08
2026
Testifies at the New York City Council AI hearing
On October 16, 2026 Gabriela Castillo Madrid, senior manager of global government affairs at SpaceXAI, testified at the Council's follow-up AI safety hearing. Speaker Julie Menin said it took a subpoena and the threat of continued legal action to compel the company to appear. The Council release does not quote a company letter.
Agrees to testify after a threatened New York court petition
On October 13, 2026 the New York City Council said SpaceXAI agreed to appear on October 16 at 10 a.m. before the Committee on Technology. The Council said that after SpaceXAI missed the October 5 hearing, it told the company it would file a petition in the New York State Supreme Court to enforce the subpoena. SpaceXAI requested a meeting on October 7 and then agreed to appear. The Council release does not publish the company's letter or any legal objections.
Grok Voice 3.0 ships in the API
On Oct. 14, 2026 xAI released Grok Voice 3.0 in the Voice Agent API at $0.08 per minute, the same price as Grok Voice 2.0, and pointed grok-voice-latest at the new model. The company said it scored 86.9 on Artificial Analysis's Speech-to-Speech Index.
Skips NYC Council AI hearing after subpoena
On Oct. 5, 2026 SpaceXAI did not appear at the New York City Council Committee of the Whole AI hearing. CNBC said the Sept. 28 subpoena required Elon Musk or another SpaceXAI representative to testify and quoted Speaker Julie Menin calling the absence a direct violation and saying the council will pursue the matter in court. CBS New York reported Menin said the company responded with a letter saying it wants to work with the city but would not attend. The City Reporter quoted the same letter paraphrase. The letter itself was not published, so named addressees and any legal objections remain unverified. The Council's Sept. 28 release said noncompliance could go to New York State Supreme Court.
Eighth Circuit grants injunction pending appeal in SpaceXAI LLC v. Ellison
On Oct. 2, 2026 the U.S. Court of Appeals for the Eighth Circuit, in SpaceXAI LLC v. Keith M. Ellison, No. 26-2806, issued a one-sentence order: Appellant's motion for injunction pending appeal is granted. The caption lists the Minnesota district case as 0:26-cv-03425-DWF. The order states no reasoning, duration, or geographic terms. The motion it granted asked the court to bar Minnesota's attorney general from enforcing Minn. Stat. 325E.91 against the appellant while the appeal proceeds; the order does not decide constitutionality or repeal the statute, which took effect Aug. 1 as HF 1606. Reuters via MPR News said the court put the law on hold for now.
Launches Team Bots for shared Grok Bot workflows
On September 28, 2026 SpaceXAI launched Team Bots, shared Grok Bots that teams can endow with files, plugins, credentials and memories, then use individually or in Slack. The company said it uses them for account briefings, engineering coordination and data questions. This is not Grok 5.
Musk signs White House Super Intelligence accord for SpaceXAI
On September 29, 2026 SpaceXAI CEO Elon Musk signed the White House Accord on Super Intelligence Joint Commitment on Frontier Responsibilities. Coverage lists him among the six company signatories with OpenAI, Anthropic, Google, Nvidia and Meta. The accord is voluntary. Tesla is not listed as a separate signatory in the published text.
X Search billing listed at per-post and per-profile rates
Official xAI developer docs now price the x_search tool at $5 per 1,000 posts fetched and $10 per 1,000 user profiles fetched, billed per item rather than per call. Parent and quoted posts count. That is the rate card previously posted as taking effect 21 September 2026 at 12:00 PM PT. The pricing page no longer presents it as a future warning. nextSteps lack hashes, so this date is recorded on the timeline rather than resolved as a milestone hit.
2026
Releases Grok Voice Transcribe 2.0
SpaceXAI released Grok Voice Transcribe 2.0, saying it is twice as accurate as version 1.0 at the same price: $0.10 per hour batch and $0.20 per hour streaming, with diarization, timestamps and key terms included. The company said the model ranks first for accuracy among 32 streaming models on the public Artificial Analysis leaderboard and will become the Speech-to-Text API default in the coming weeks, with version 1.0 deprecated unless callers pin grok-voice-transcribe-1.0.
Musk maps Colossus to as many as 1.44 million GPUs
On September 25, 2026 Elon Musk said Colossus 1 holds 150,000 H100, 50,000 H200 and 30,000 GB200 chips and Colossus 2 holds 110,000 GB200 and 440,000 GB300 chips. He said another 220,000 GB300 chips will be fully operational next week, 220,000 in November, and a third 220,000 by late December if the company gets lucky, which would take the combined fleet to about 1.44 million GPUs. Bloomberg reported the post as the most detailed timetable yet; the December wave is conditional.
SpaceXAI introduced Grok 4.7 as its most capable model for coding and knowledge work, available in Cursor, Grok Build, the Grok API, third-party coding harnesses, and cloud platforms at $2 and $6 per million input and output tokens, with a faster variant at twice the output speed and twice the price. The company said the model uses a larger base than Grok 4.6, was trained with a longer reinforcement-learning run on multi-hour tasks, and ships a new safeguard stack that scored 62.4% on LatchBio's biosafety benchmark and allowed 3.3% of risky HackerBench v0.3 prompts through.
Dismisses Apple claims in the OpenAI antitrust suit
On September 14, 2026, X Corp and SpaceXAI moved in federal court to dismiss their antitrust claims against Apple in a suit that alleged Apple and OpenAI locked up smartphone and chatbot markets by integrating ChatGPT into Apple Intelligence. The filing did not explain the reasons or say whether a settlement had been reached. The companies said they will continue to pursue claims against OpenAI.
X Search shifts to per-post and per-profile billing
xAI's official developer pricing page warns that from September 21, 2026 at 12:00 PM PT, the x_search tool will be billed at $5 per 1,000 posts fetched and $10 per 1,000 user profiles fetched, replacing the current $5 per 1,000 calls. Every post returned by a search or thread fetch, including parent and quoted posts, counts; every profile returned by a user search counts. Web Search stays at $5 per 1,000 calls.
Grok 4.7 slips past the mid-September target
Musk said on September 2, 2026 that the 2.1-trillion-parameter Grok 4.7 was about 10 days out, implying around September 12. On September 11 he said the model needed a few more days because reinforcement-learning length penalties were making it quit solvable hard tasks early and check its work too loosely. As of September 13 xAI had still not published a model card, API identifier or pricing, and official developer docs continued to list Grok 4.6 as the current flagship.
On September 3, 2026 xAI made Grok Bot available for enterprises, adding access, network, and audit controls so organizations can govern always-on agents at scale. Grok and Cursor Enterprise customers received two weeks of free usage and can invite their whole organization, including people without an existing seat. xAI said thousands of organizations have adopted Grok Bot since the August 11 public beta, with the heaviest use outside engineering.
Pentagon puts Grok for Government live on GenAI.mil
On August 31, 2026 the US Department of Defense launched Starshield AI's Grok for Government on GenAI.mil at Impact Level 5, alongside OpenAI's ChatGPT Mil, joining Google Gemini on the department's secure generative-AI portal. The Pentagon said the tools are tailored to warfighter needs for about 3 million military and civilian personnel, and that GenAI.mil had already onboarded more than 1.7 million unique users. Grok for Government is described as offering deep-thinking inference, Auto/Fast/Expert reasoning modes, customizable workspaces, persistent projects, and reusable playbooks for tasks such as acquisition market research and supply-chain management.
Memphis models cluster outage takes Grok offline
On September 3, 2026 Grok went down for about 3.5 hours from 6:30 a.m. PT across X, iOS and Android after an outage at SpaceXAI's Memphis models cluster. The company apologized to users and unnamed compute partners and said it is taking corrective action; it did not name other chatbots as sharing the same root cause.
Minnesota judge denies anti-nudification injunction
On or around September 4, 2026, US District Judge Donovan Frank denied xAI's motion for a preliminary injunction against Minnesota HF 1605, the state's anti-nudification law that took effect August 1, allowing Minnesota to keep enforcing civil fines of up to $500,000 per violation while the First Amendment case proceeds. xAI noticed an appeal to the Eighth Circuit.
Sues photographer over Grok-generated CSAM
On August 24, 2026 xAI sues Bentonville, Arkansas photographer Russell Bloodworth, who was arrested and charged with 100 counts of possessing child pornography and 100 counts of unlawfully creating deepfake visual material after authorities found roughly 1,700 sexualized images and videos of minors on his phone, many generated with Grok from photos of his juvenile clients. xAI's suit seeks to hold Bloodworth liable for legal costs from related lawsuits, arguing he used adversarial prompts to defeat Grok's safety guardrails - a shift to offense amid mounting CSAM-deepfake litigation against the company.
Musk tells newly acquired Cursor staff Grok is 'losing' to Anthropic
In his first address to Cursor's 1,000+ employees following the August 14, 2026 acquisition close, Elon Musk told staff he is 'not used to losing' and named Anthropic as the AI industry's current frontrunner, framing the Cursor deal as a way to close that gap. The admission followed earlier internal reporting that xAI staff themselves often used Anthropic's Claude rather than Grok, and came as Nicolls's SpaceXAI leadership continued to benchmark Grok's progress directly against Anthropic's models.
Cursor launches Origin, a GitHub-rival code-hosting platform
On August 18, 2026 Cursor, now part of SpaceXAI following the Anysphere acquisition, launches Origin - a code-hosting platform built to handle repositories, pull requests and collaborative editing 'at agent scale,' positioned as a direct competitor to Microsoft's GitHub. The launch rolled out to paid users the same day GitHub suffered a major outage, and Origin already integrates with Vercel, Buildkite and Depot.
Confirms 'Minihard,' a fourth Memphis-area data center
In late July 2026 Musk confirms a new SpaceXAI facility, nicknamed 'Minihard,' rising alongside the existing Colossus 2 and Colossus 3 ('MACROHARDRR') buildings on the Tulane Road campus in Memphis - a third structure there, housing roughly 220,000 Nvidia GB300 GPUs in a denser configuration Musk calls the second of several standardized modular data-center units. With Minihard, the Memphis-area campus is projected to approach the 1-million-GPU target by Q1-Q2 2027; the announcement lands alongside the start of turbine removals tied to the Mississippi Department of Environmental Quality's July 31 agreed order.
Closes $60B Cursor (Anysphere) acquisition
SpaceX's all-stock acquisition of Cursor-maker Anysphere, first agreed June 16, 2026, became effective August 14, 2026 per a regulatory filing - well ahead of the Q3 2026 target - with Cursor publicly confirming it had joined 'the SpaceXAI team' the next day. Cursor becomes a wholly owned subsidiary with access to SpaceX's GPU fleet, including Colossus, for AI model training; the companies say they have already jointly trained Grok models on trillions of tokens of Cursor's coding data, extending the AI-compute push that began with the February 2026 xAI-SpaceX merger.
Named in Resemble AI report as top source of deepfake attacks
Cybersecurity firm Resemble AI's H1 2026 Deepfake Threat Report, published August 12, 2026, found Grok was behind 87% of the synthetic files tied to 821 documented deepfake attacks in the first half of the year, affecting at least 15,736 confirmed victims and totaling roughly 3.46 million synthetic images, videos and audio files. Resemble AI said about one in six attacks involved non-consensual intimate imagery, including of minors, and released a new detection model, DETECT-World, aimed at catching content from generators it had not previously encountered.
On August 12, 2026 xAI releases Grok 4.6, reusing Grok 4.5's 1.5T-parameter foundation but with improved supervised fine-tuning and reinforcement learning aimed at long-running agents, coding and knowledge work. The model scores 61 on the Artificial Analysis Intelligence Index, matching OpenAI's GPT-5.6 Sol and just behind Anthropic's Claude Fable 5, and ships same-day in the Cursor code editor, Grok Build, and via API at $2/$6 per million input/output tokens with a 500,000-token context window.
Ships Grok Imagine Image 2.0 with precision editing tools
On August 7, 2026 xAI shipped Grok Imagine Image 2.0 as Quality Mode on grok.com/imagine and in its iOS and Android apps, centering editing as a core feature with a magic-wand tool for region-specific changes, precise segmentation, background removal, and multi-reference editing accepting up to five input images per generation. xAI said the model ranks second in the world on the Arena leaderboards for both text-to-image generation and image editing, behind OpenAI's gpt-image-2.
Agrees to remove unpermitted Colossus turbines by July 2027
On July 31, 2026 SpaceX formalizes an agreed order with the Mississippi Department of Environmental Quality committing to phase out the roughly 69 temporary gas turbines powering Colossus, with removal beginning as early as August 2026 and complete by July 2027, as a permitted 1.2 GW gas plant (41 turbines, authorized under a March 2026 Clean Air Act permit) comes online. The agreement follows the NAACP's Clean Air Act lawsuit over the unpermitted units; the Department of Justice separately moved in June 2026 to intervene and dismiss that suit, citing Grok's use in Department of War targeting, intelligence and recruitment missions as a national-security interest.
Grok Imagine declared 'done'; major video update ships
On July 5, 2026 Musk posts 'Done with Grok Imagine,' and on July 31 xAI ships a major update adding native 1080p output, voice/face-consistent references and up to seven-image multi-reference scene control. Grok Imagine Video 1.5 had already topped the Image-to-Video Arena leaderboard on its June 2026 launch, priced roughly 86% below OpenAI's Sora.
Minnesota 'nudification' ban takes effect after judge denies xAI's block request
On July 31, 2026 U.S. District Judge Donovan Frank denies xAI's request for a temporary restraining order, ruling the company waited nearly three months after the law's signing and filed only days before its effective date. Minnesota's first-in-the-nation ban on AI 'nudification' tools takes effect August 1 while xAI's First Amendment challenge continues on the merits.
Sues Minnesota over AI 'nudification' ban
On July 27, 2026 xAI sues Minnesota Attorney General Keith Ellison in federal court, challenging the state's first-in-the-nation law banning AI tools that create nonconsensual nude images before it takes effect August 1. The suit does not contest banning distribution of nonconsensual intimate images but argues the law's broader scope violates the First Amendment, and follows global backlash over Grok generating sexualized images, including of children.
Rebrands to SpaceXAI; Grok 4.5 launches with Cursor
On July 6, 2026 xAI completes its rebrand to SpaceXAI with a new logo and @SpaceXAI handle (Grok keeps its name). On July 8–9 it releases Grok 4.5, its first model co-developed with Cursor - a mixture-of-experts model Musk calls 'Opus-class, but faster, more token-efficient and lower cost,' trained on tens of thousands of Nvidia GB300s and priced at $2/$6 per million tokens.
SpaceX's record IPO; $60B Cursor acquisition
The combined SpaceX–xAI lists on Nasdaq as SPCX on June 12, 2026, raising ~$85.7B (with overallotment) in the largest IPO in history - shares priced at $135 closed their first day up 19% at $160.95, a ~$2.1T market cap. Four days later SpaceX agrees to acquire Cursor-maker Anysphere for $60B in stock, the largest acquisition of a venture-backed startup ever, feeding Cursor's coding data into Grok's training pipeline.
Colossus becomes a compute-leasing business: Anthropic, Google and Reflection AI deals
In May 2026 Anthropic agrees to pay $1.25B per month through May 2029 (over $40B total) to rent all of Colossus 1's ~220,000 GPUs and 300 MW for Claude inference; on June 5 Google signs a $920M-per-month deal (~$30B through June 2029) for ~110,000 GPUs at Colossus 2 starting October 2026; and on June 22 open-source AI lab Reflection AI signs a third deal, paying $150M per month from July 1, 2026 (up to $6.3B through 2029) for Nvidia GB300 capacity at Colossus 2 - xAI having shifted its own training to Colossus 2.
Grok goes into classified Pentagon systems; NAACP sues over Colossus 2
In February 2026 the Pentagon agrees to embed Grok in GenAI.mil, its internal AI platform serving 3 million military and civilian employees at Impact Level 5, including use in sensitive workflows. In April the NAACP sues over Colossus 2's gas-turbine emissions in the Memphis area - alleging 1,200–2,000 tons of smog-forming NOx annually - with the Trump administration moving in June to intervene and dismiss the case.
SpaceX acquires xAI in the largest private merger ever
On February 2, 2026 SpaceX acquires xAI in an all-stock transaction valuing xAI at $250B and the combined company at $1.25T (0.1433 SpaceX shares per xAI share), folding Grok and X into SpaceX - partly to fund xAI's cash burn and pursue Musk's plan for space-based data centers. By late March all eleven co-founders other than Musk had departed; ex-Starlink VP Michael Nicolls became president of the division on April 10.
$20B Series E at ~$230B valuation
On January 6, 2026 xAI closes a $20B Series E - above its $15B target - with Nvidia, Cisco, Fidelity, Valor Equity Partners, StepStone, QIA, MGX and Baron Capital participating. Tesla contributes $2B despite a November 2025 Tesla shareholder vote that failed to endorse the investment once abstentions were counted.
2025
xAI launches Grokipedia, an AI-generated encyclopedia positioned as a Wikipedia alternative, on October 27, 2025, and releases Grok 4.1 on November 17, 2025 - briefly #1 on LMArena's text leaderboard - alongside Grok 4.1 Fast with a 2-million-token context window.
Grok 4 launches; $200M Pentagon contract amid controversy
xAI unveils Grok 4 and the $300/month SuperGrok Heavy tier on July 9, 2025 - days after Grok generated widely condemned antisemitic posts on X. On July 14 the DoD's Chief Digital and AI Office nonetheless awards xAI a $200M-ceiling contract alongside its 'Grok for Government' launch, and in September the GSA offers Grok to federal agencies for $0.42 per agency for 18 months.
Acquires X Corp to form XAI Holdings
On March 28, 2025 Musk announces that xAI has acquired X Corp in an all-stock deal valuing X at $33B ($45B including debt) and xAI at $80B, merging the AI lab with the social platform's data, distribution and 600M monthly users under X.AI Holdings Corp.
On February 17, 2025 xAI releases Grok 3, its first model trained on the full Colossus cluster, adding step-by-step 'Think' reasoning and the DeepSearch web-research agent.
2024
Colossus supercomputer built in 122 days
xAI stands up Colossus in a former Electrolux factory in Memphis, Tennessee - 100,000 Nvidia H100 GPUs online in 122 days from conception, called the world's largest AI training cluster at launch. It doubles to 200,000 GPUs by December 2024, though its unpermitted methane gas turbines spark a lasting environmental fight with South Memphis residents.
Grok-1 open-sourced; Grok-2 adds image generation
xAI releases Grok-1's weights open-source on March 17, 2024, ships Grok-1.5 with a 128K context window, and on August 14, 2024 launches Grok-2 - its first model with image generation.
2023
On November 4, 2023 xAI unveils Grok, a conversational AI with a 'rebellious streak' and real-time access to X posts, initially offered to X Premium+ subscribers - establishing the X platform as Grok's primary distribution channel.
xAI is incorporated on March 9, 2023 and announced publicly on July 12, 2023, with Musk and eleven founding AI researchers (including Igor Babuschkin, Christian Szegedy, Greg Yang and Yuhuai Wu) setting out to 'understand the true nature of the universe' - and to challenge OpenAI, which Musk co-founded and left.
09
$372.8Bdisclosed across 7 rounds with a stated amount
Cumulative disclosed raise · dated rounds
Series A
Nov 2023
Early backers aligned with Musk
Series B
May 2024
Andreessen Horowitz, Lightspeed, Sequoia, Tribe Capital, Valor Equity Partners
Series C
Dec 2024
Fidelity, BlackRock, Sequoia, MGX, Nvidia, AMD
Debt + equity raise
Jul 2025
$5B debt + $5B equity (>$120B valuation)
Morgan Stanley-arranged debt; SpaceX ($2B equity)
Series E
Jan 2026
Nvidia, Cisco Investments, Fidelity, Valor Equity Partners, StepStone, Qatar Investment Authority, MGX, Baron Capital, Tesla ($2B)
SpaceX all-stock acquisition
Feb 2026
xAI valued at $250B ($1.25T combined)
SpaceX (share exchange: 0.1433 SpaceX shares per xAI share)
IPO (as part of SpaceX, Nasdaq: SPCX)
Jun 2026
~$85.7B raised at $135/share (largest IPO ever)
Public markets; first-day close $160.95 (~$2.1T market cap)
