Grok had approximately 117 million MAUs that used Grok’s AI features as of March 31, 2026, up from approximately 89 million on December 31, 2025, according to the amended Form S-1 that SpaceX filed with the Securities and Exchange Commission. On the same date, SpaceX counted approximately 1.9 million SuperGrok, SuperGrok Heavy and SuperGrok Lite paid subscribers, a paying base of about 1.6% of Grok’s monthly users. That gap between reach and revenue runs through the Grok AI statistics that follow.
Revenue moved faster than subscriptions. AI revenues were up 213% sequentially and 247% year-over-year to $2.6 billion in the second quarter of 2026, yet Cloud Services Agreements resulted in $1.6 billion in incremental AI infrastructure revenues for the second quarter. The Grok statistics below cover users, subscribers, pricing, model releases, benchmarks, compute, market share, and funding, and each figure traces to a SpaceX filing, a SpaceXAI announcement, or a named measurement firm.
Key Takeaways
- Only about 1.6% of Grok’s monthly active users paid for a SuperGrok-tier plan as of March 31, 2026.
- About 62% of the AI segment’s Q2 2026 revenue came from compute sold under Cloud Services Agreements.
- Advertising revenues fell about 13.9% year over year in the quarter ended June 30, 2026, the only AI revenue line that shrank.
- AI capex ran about 6.2 times AI revenue in the second quarter of 2026.
- U.S. market share of Grok rose to 17.8% from 1.9% in January 2025, according to data from Apptopia cited by Reuters, while Grok held 2.4% of global generative AI website traffic in Similarweb’s latest reading.
- Grok 4.7 lists at $2 per million input tokens and $6 per million output tokens, against $10 and $50 for Fable 5.1 Max.
- SpaceXAI shipped Grok 4.5, Grok 4.6, and Grok 4.7 within 67 days.
Editor’s Choice
- Grok monthly active users: approximately 117 million MAUs who used Grok’s AI features as of March 31, 2026.
- X and Grok combined MAUs: approximately 550 million in the period ended March 31, 2026.
- Q2 2026 AI revenue: $2.6 billion, up 247% year-over-year.
- Paid subscribers across X and Grok: approximately 6.3 million as of March 31, 2026.
- Nameplate compute: 1.4 GW, up from 1.0 GW in Q1 2026.
- Anthropic compute deal: $1.25 billion per month through May 2029.
- Series E funding: $20 billion, exceeding the $15 billion targeted round size.
Grok AI Statistics: How Many People Use Grok?
These figures are compiled from 47 cited findings: 43 from company, SEC, and government documents (tier 1), 3 from measurement firms (tier 2), and 1 from secondary news coverage (tier 3). Sources run from November 2023 to September 2026. Only primary or named measurement data qualified, and figures are reviewed on a rolling basis when sources publish new editions.
- Grok’s monthly user base grew 31.5% in a single quarter, from year-end 2025 to March 31, 2026.
- Grok users made up about 21% of the combined X and Grok monthly audience.
- The integrated platforms across Grok and X had over 1.3 billion supported accounts active in the last twelve months ended March 31, 2026, up from over 1.1 billion.
- The filing says SpaceX manages the business with the objective of driving sustainable revenue growth and profitability rather than with the primary objective of growing or maintaining MAU levels.
- Apptopia data placed Grok as the third most-used chatbot in the country in January, behind ChatGPT and Google Gemini.
- Approximately 4.4 million of the paid subscribers held X Premium and Premium+ plans rather than a Grok tier.
| Metric (millions) | Dec 31, 2025 | Mar 31, 2026 |
|---|---|---|
| Grok MAUs | 89 | 117 |
| X and Grok combined MAUs | 520 | 550 |
Source: SpaceX Form S-1 and Form S-1/A, May 2026
Does anyone actually use Grok AI?
Yes, at scale. SpaceX reported approximately 117 million MAUs that used Grok’s AI features as of March 31, 2026, which it counts inside a combined X and Grok audience. That audience included approximately 550 million MAUs, so most X users still do not touch Grok in a given month.
Grok and xAI Revenue Statistics
- SpaceX’s AI segment generated revenue of $3,201 million in 2025, with a loss from operations of $6,355 million.
- Revenue for the year ended December 31, 2025 increased by $581 million, or 22.2%, compared to 2024.
- X and Grok subscription revenue increased by $365 million and data licensing revenue by $88 million in 2025.
- Research and development spending rose to $5,064 million in 2025 from $1,176 million in 2024.
- Revenue for the three months ended March 31, 2026 increased by $91 million, or 12.5%, with Grok and X subscription revenue up $177 million.
- Per SpaceX’s Q2 2026 earnings release, the segment cut its operating loss by 49% compared to Q1 2026 to a loss of $1.3 billion and achieved positive AI-adjusted EBITDA of $1.1 billion.
- Advertising revenue fell about 13.9% year over year in the second quarter of 2026.
Is Grok losing money?
Yes, at the operating line. The AI segment that houses Grok posted a loss from operations of $1,257 million in the quarter ended June 30, 2026, and its 2025 loss from operations was $6,355 million. The same quarter produced positive AI-adjusted EBITDA of $1.1 billion, a non-GAAP measure.
Recent Developments
- July 2026: On July 16, Grok 4.5 posted a 29.0% SWE Marathon resolution rate, against 26.0% for Opus 4.8 (max).
- July 2026: On July 29, Grok Voice Think Fast 2.0 scored 82.9% on the Overall AA Speech-to-Speech Quality Index, against 75.7% for Grok Voice Think Fast 1.0.
- August 2026: On August 4, SpaceX’s second-quarter results showed Cloud Services Agreements totaling $14.1 billion in contracted sales and an agreement to acquire Cursor for $60 billion.
- August 2026: On August 12, Grok 4.6 High scored 61 on the AA Intelligence Index, the same score as GPT-5.6 Sol Max.
- August 2026: On August 26, Grok Bot, launched in beta on August 11, was included with all SuperGrok, Cursor Pro, and Cursor Teams plans.
- September 2026: On September 21, Grok 4.7 arrived at the same price and speed as Grok 4.6 and scored 46.3% on CursorBench 4.0.
Grok Subscribers and Pricing Plans
- Grok’s free plan includes real-time web and X search, voice mode, and connectors.
- SuperGrok includes the Grok 4.6 model, Grok Bot access, and image and video generation.
- SuperGrok Plus offers significantly higher usage, 1080p video creation, and priority access.
- SpaceX’s market-sizing model uses the $30 SuperGrok price for high-income countries and $8 monthly cost in upper-middle and lower-middle-income countries.
- That model puts the consumer subscription market opportunity at approximately $760 billion a year.
- The GSA’s OneGov agreement made Grok AI models accessible to federal agencies for $0.42 per organization for 18 months.
- The offer is valid until March 2027 and includes Grok 4 and Grok 4 Fast.
| Subscriber group | Paid subscribers (millions) |
|---|---|
| X Premium and Premium+ | 4.4 |
| SuperGrok, SuperGrok Heavy and SuperGrok Lite | 1.9 |
| Total X and Grok paid subscribers | 6.3 |
Source: SpaceX Form S-1, March 31, 2026 data
By the numbers: SpaceX’s filing counts approximately 6.3 million active paid subscribers as of March 31, 2026, including approximately 4.4 million X Premium and Premium+ subscribers. X plans made up about 70% of that paid base, so Grok’s own tiers still trail the social network’s subscriptions.
Grok API Pricing Statistics
- Grok 4.7 bills $2 input below 200,000 prompt tokens, rising to $4 input, $1 cached input and $12 output above it.
- Cached input costs $0.50 per million tokens below 200,000 prompt tokens.
- The Grok-4.7 model has a 500,000-token context window.
- Grok 4.7 Fast is the same model at twice the token rates.
- Voice API pricing is $0.08 per minute for its agent, $15.00 per million characters for text-to-speech, and $0.10 per hour for batch speech-to-text.
- Imagine API image generation starts at $0.02 per image and video at $0.05 per second.
- Grok 4.5 is priced at $2 per million input tokens and $6 per million output tokens, the same list price that Grok 4.7 carries.
Flat list pricing across three releases means the per-token cost of Grok’s frontier tier has not moved since the summer, while each release raised the benchmark scores that buyers compare.
Grok Model Release Timeline
- After announcing xAI, the company trained a prototype LLM, Grok-0, with 33 billion parameters.
- Grok 3 was trained on the Colossus supercluster with 10x the compute of previous state-of-the-art models.
- Grok 4 used Colossus, a 200,000 GPU cluster, to run reinforcement learning training, and new infrastructure and algorithmic work increased training compute efficiency by 6x.
- Grok 4.5 was trained alongside Cursor and incorporates a 1.5 trillion-parameter V9 foundation model.
- Grok 4.6 builds on Grok 4.5 with a particular focus on long-running agents.
- Grok 4.7 uses a new, larger base model compared to Grok 4.6.
- Grok 5 is currently being trained at COLOSSUS II, according to the S-1.
What model powers Grok in 2026?
Grok 4.7 is the current flagship. SpaceXAI’s model documentation lists Grok 4.7 as its flagship model for code and everything else, with a 500,000-token context window. The SuperGrok plan page still lists the Grok 4.6 model, so consumer plans can trail the newest API release by a version. The AI model tracker lists these Grok versions next to rival labs’ launches.
When was Grok AI launched?
xAI launched Grok-1 in November 2023 and has since released four major versions and notable variations thereof. That run culminated in Grok-4.3 in April 2026 before the summer releases of Grok 4.5, Grok 4.6, and Grok 4.7 on the SpaceXAI news page.
Grok 4.7 Benchmark Statistics
- On DeepSWE v1.1, Grok 4.7 scored 71.0% at xHigh effort, against 72.7% for GPT-5.6 Sol Max.
- On AA Briefcase v1.1, which covers multi-hour office work, Grok 4.7 scored 1,657 against 1,678 for Fable 5.1 Max.
- On HealthBench Professional, Grok 4.7 scored 56.7% against 60.5% for GPT-5.6 Sol Max.
- Grok 4.7 topped LatchBio’s biosafety benchmark at 62.4%.
- On HackerBench v0.3, a SpaceXAI benchmark, Grok 4.7 allowed only 3.3% of risky dual-use prompts through, the kind of misuse measured in AI jailbreaking statistics.
- Every score in this section comes from SpaceXAI’s own announcement, so treat the comparison as vendor-reported rather than independently tested.
Is Grok the smartest AI right now?
No single benchmark settles it. On SpaceXAI’s own Grok 4.7 table, Fable 5.1 Max scored 51.8% on CursorBench 4.0 and 57.9% on Terminal-Bench 4.0, while Grok 4.7 scored 64.0% on EEBench and 19.6% on the Harvey Legal Agent Benchmark. The answer depends on the task, and the rankings shift with each release.
Grok Benchmark History: From Grok-1 to Grok 4.6
- Grok 3 (Think) attained 79.4% on LiveCodeBench, and Grok 3 mini reached 95.8% on AIME 2024.
- On the agentic Vending-Bench, Grok 4 reached $4,694.15 net worth and 4,569 units sold, against $2,077.41 for Claude Opus 4.
- Compared to the previous production model, Grok 4.1 was preferred 64.78% of the time in blind pairwise evaluations.
- Grok 4.1 in its non-reasoning mode ranked #2 at 1465 Elo, and the Grok 4.1 benchmark breakdown walks through the rest of that release.
- Grok 4.6 High scored 69.9% on CursorBench v3.2, against 66.7% for Grok 4.5 High.
- On APEX-Agents, Grok 4.6 High reached 57.5% against 47.1% for Grok 4.5 High.
- Grok 4.5 resolves SWE Bench Pro tasks with 15,954 output tokens on average, about 4.2 times fewer than Opus 4.8 (max).
| Model | Benchmark | Score |
|---|---|---|
| Grok-1 | HumanEval | 63.2% |
| Grok-1 | MMLU | 73% |
| Grok 3 (Think) | AIME 2025 | 93.3% |
| Grok 3 (Think) | GPQA | 84.6% |
| Grok 3 | Chatbot Arena Elo | 1402 |
| Grok 4 | ARC-AGI V2 | 15.9% |
| Grok 4 Heavy | Humanity’s Last Exam (text-only subset) | 50.7% |
| Grok 4.1 Thinking | LMArena Text Arena Elo | 1483 |
| Grok 4.6 High | AA Intelligence Index | 61 |
Source: SpaceXAI model announcements, November 2023 to August 2026
Key finding: In LMArena’s Text Arena, Grok 4.1 Thinking held the #1 overall position with 1483 Elo, a margin of 31 points over the highest non-xAI model. Our AI benchmark coverage shows arena leadership changes hands within a few release cycles, so that lead says more about pace than permanence.
Colossus Supercomputer Statistics
- COLOSSUS and COLOSSUS II collectively provide approximately 1.0 gigawatt of compute power, according to the S-1.
- The next phase of expansion at COLOSSUS II will bring online at least 220,000 additional GB300 processors and over 400 additional megawatts once fully operational.
- An industry benchmark to bring online a 100-megawatt greenfield data center is approximately two years.
- Colossus started running workloads just 19 days after the first servers were delivered.
- xAI ended 2025 with over one million H100 GPU equivalents across Colossus I and II.
- Nameplate compute expanded with the continued build-out of Colossus II and significant incremental capacity under construction.
| Cluster | Processors | Compute power (MW) | Days to bring online |
|---|---|---|---|
| Colossus, first cluster | approximately 100,000 H100 | approximately 130 | 122 |
| Colossus II, first cluster | approximately 110,000 GB200 | approximately 210 | 91 |
| Colossus II, second cluster | 110,000 GB300 | 220 | 64 |
Source: SpaceX Form S-1, May 2026
How Much Is xAI Spending? AI Capex Statistics
- In 2025, capital expenditures for the AI segment were $12,727 million.
- Six-month AI capex reached $23,551 million by June 30, 2026, against $3,316 million a year earlier.
- AI depreciation and amortization reached $1,885 million in the quarter ended June 30, 2026, against $811 million a year earlier.
- Anthropic agreed to pay $1.25 billion per month through May 2029 for compute capacity across COLOSSUS and COLOSSUS II, a customer whose own scale appears in OpenAI vs Anthropic statistics.
- At the full monthly rate, that contract is worth about $15 billion a year.
- The agreements may be terminated by either party upon 90 days’ notice.
Why it matters: AI capex reached $15,828 million in the quarter ended June 30, 2026, while AI revenues were $2.6 billion. Renting spare capacity to outside model developers recovers part of that outlay, which is why compute contracts drove the second-quarter revenue jump.
Grok Market Share Statistics
- ChatGPT’s share slumped to 52.9% from 80.9% in January last year, while Gemini’s grew to 29.4% from 17.3%, per the same Apptopia data.
- Similarweb’s January 2 reading, its first Global AI Tracker of 2026, put Grok at 3.4% of generative AI website traffic.
- In Similarweb’s September 7, 2026 update, ChatGPT held 55.5% and Gemini 25.6% one month earlier.
- StatCounter’s August 2026 AI chatbot table lists ChatGPT at 79.4%, Google Gemini at 10.9%, and Perplexity at 4.31%, and Grok does not appear in it at all.
- Grok runs inside X as well as on its own site and apps, so web-referral trackers see only part of its usage. Keep that in mind when setting Grok beside ChatGPT, Claude, Gemini and Perplexity usage data drawn from web panels.
EU regulatory risk: The European Commission has launched a new formal investigation against X under the Digital Services Act to assess whether it properly assessed and mitigated risks associated with the deployment of Grok’s functionalities into X in the EU, including manipulated sexually explicit images.
Where does Grok rank on AI?
Grok’s rank depends on the channel, and on the web, DeepSeek’s traffic still sits above it. Apptopia data placed Grok as the third most-used chatbot in the country in January, behind ChatGPT and second-placed Google Gemini. Similarweb’s latest global reading put Grok at 2.4% of generative AI website traffic, behind DeepSeek at 3.4%.
Grok Imagine, Voice and Grok Bot Statistics
- Imagine produced approximately 10 billion images and over 2 billion videos per month, on average, for the quarter ending March 31, 2026.
- Grok Voice, a real-time speech engine, was developed in under six months.
- Grok Voice Think Fast 2.0 reached a time to first audio of 0.70 seconds, against 1.25 seconds for Grok Voice Think Fast 1.0 and 2.98 seconds for Gemini 3.1 Flash (High).
- Grok Bot is included with SuperGrok, SuperGrok Plus, SuperGrok Heavy, Cursor Pro, Cursor Pro+ and Cursor Ultra plans, one of the agent products covered in AI agent adoption statistics.
- Grok Bot for Enterprise gave Grok and Cursor Enterprise customers free usage for the next two weeks, announced Sep 3, 2026.
| Benchmark | Grok Voice Think Fast 2.0 | Grok Voice Think Fast 1.0 | GPT-Realtime-2.1 (High) | Gemini 3.1 Flash (High) |
|---|---|---|---|---|
| Overall AA Speech-to-Speech Quality Index | 82.9% | 75.7% | 79.1% | 69.5% |
| Big Bench Audio | 97.2% | 97.1% | 96.0% | 96.6% |
| Full Duplex Bench | 95.1% | 77.8% | 95.7% | 74.3% |
| τ-voice Bench | 56.5% | 52.1% | 45.7% | 37.7% |
Source: SpaceXAI Grok Voice Think Fast 2.0 announcement, July 2026
xAI Funding and Valuation Statistics
- Series E investors included Valor Equity Partners, Fidelity Management & Research Company, Qatar Investment Authority and MGX, with NVIDIA and Cisco Investments as strategic investors.
- Strategic investors NVIDIA and AMD participated in the Series C.
- The three rounds xAI announced on its own site add up to $32 billion, a total that can be set against other labs in the AI funding tracker.
- The SpaceX deal valued SpaceX at $1 trillion and xAI at $250 billion, Reuters has reported.
- SpaceX’s initial public offering brought net proceeds of approximately $85.7 billion, with shares trading from June 12, 2026, under the ticker SPCX.
How much money has xAI raised?
xAI’s Series B, Series C, and Series E announcements total $32 billion. Series E funding alone reached $20 billion, exceeding the $15 billion targeted round size, and it closed shortly before SpaceX bought the company.
Who owns Grok AI in 2026?
SpaceX owns Grok, alongside the Space and Connectivity businesses covered in SpaceX statistics. SpaceX says it acquired xAI in February 2026 as the foundational platform for its AI segment, and its Class A shares began trading on Nasdaq on June 12, 2026 under the ticker SPCX. Grok’s developer now publishes under the SpaceXAI name.
Conclusion
Grok reached approximately 117 million MAUs as of March 31, 2026, and AI revenues climbed to $2.6 billion in the second quarter. The paying base is still small, though, and compute rentals rather than Grok subscriptions carried the latest quarter, so developers benefit most right now from flat token pricing while investors carry a capex bill that runs several times revenue.
Grok 5 is currently being trained at COLOSSUS II, and SpaceX said it expects to close the Cursor deal in the third quarter. Those two events will show whether the next model can lift paid conversion, or whether renting compute stays the larger business this year.




















































