Farzad: Why Tesla and SpaceX "Have to Merge" + His CoreWeave Question + SpaceX Says Grok Bot Has 800,000 Paid Weekly Users

Late Tuesday evening, Farzad laid out why he thinks Tesla and SpaceX have to merge: inference compute soaking up idle grid capacity. Earlier, around the SpaceX earnings call, he posted on compute payback and asked what he's missing on CoreWeave. Grok Bot posts and a Wednesday morning apology to Elon rounded out the X reaction.

"Tesla and SpaceX have to merge"

Farzad, Oct 20 at 10:52pm CT: Guys the reason why Tesla and SpaceX have to merge is because distributed compute will be the FASTEST way to bring up AI inference, because there's UNGODLY amounts of waste in the electrical grid in the US.

His reasoning: the grid "has to be built for absolute worst case scenario," so most of the time it's "nowhere near its max capacity." He'd capture that slack with "batteries (Tesla) + inference chips (SpaceX/Terafab) installed freaking EVERYWHERE." His list: Superchargers, businesses, homes, utilities, power stations.

He wrote that those sites "will pay for themselves in a year or two due to the massive premium tokens are going for," and that "the player that can deploy power/chips the fastest will ultimately be the most valuable company in the world." That player, in his words, "appears to be Musk Industry or whatever we're calling this thing."

That's Farzad's projection. The one to two year payback is his estimate, and the post doesn't show the math. Nothing in these posts says Tesla or SpaceX plans to merge or to put inference chips at Superchargers.

He'd floated pieces of it Tuesday afternoon. At 4:23pm CT he posted: Add inference compute to every Tesla supercharger $SPCX $TSLA. At 4:27pm CT he quoted his own Aug 12 post, where his gut told him a majority of $TSLA shareholders "will be wanting to be acquired by $SPCX by the end of the year." This time he added: I think my gut is gonna be right.

SawyerMerritt, 4:38pm CT: In my opinion, SpaceX's Q3 earnings call was the best earnings call from any of Elon's companies to date. The post adds: "SpaceX has one heck of an executive team." Farzad replied at 4:48pm CT: Merge and get these guys to run the whole shebang!

Check the merge post here. The Supercharger post here. Farzad's quote here. SawyerMerritt here. Farzad's reply here.

"6 month to 1 year payback"

Farzad, Oct 20 at 5:04pm CT: So let me get this straight. SpaceX gets a 6 month to 1 year payback on MASSIVE compute clusters.

He went on: "It can then turn around and sell the capacity to customers at MASSIVE premiums as the demand for tokens continues to go exponential." Less than a minute later he replied to himself: Add on Starlink and Starship. Wild shit.

suddersan replied at 5:43pm CT: "I thought payback was 22 months at this point?" The reply also asked what $m per MW Farzad was using. Farzad answered at 6:54pm CT: Company guidance.

So Farzad credits the 6 month to 1 year figure to company guidance, and the 22 months is suddersan's number. Neither post shows a calculation.

Farzad named the condition himself. At 3:14pm CT he wrote that SpaceX is "running at ~20% Net Income in Q3 2026" and that "Now the race begins to outstrip the $19.5 BILLION of quarterly CAPEX." Then: "As long as AI demand holds up, SpaceX is on track to be absurdly cash-flow positive." The ~20% and the $19.5 billion are the figures in his post. On the call, SpaceX put Q3 capital spending at $19.6 billion.

TSLAFanMtl, 6:08pm CT: Fundamental assumption that Elon makes and which, if true, makes the bull case on $SPCX a near certainty, is that the value of AI compute will INCREASE over time as the AI models and tools get better and more productive.

Farzad replied at 9:11pm CT: There's not going to be price equilibrium until we find the maximum amount of intelligence that humans would demand AS LONG AS the AI itself doesn't get to a point where it can create its own compute, at which point we are living in Sci-Fi land.

At 4:20pm CT, Farzad had also RTd an earlier TSLAFanMtl post: So unless AI goes away tomorrow... $SPCX seems to be massively undervalued here. That valuation call is TSLAFanMtl's.

Check Farzad's payback post here and his follow-up here. suddersan here. Farzad's answer here. The 3:14pm post here. TSLAFanMtl here and Farzad's reply here. The undervalued post here. Farzad's RT here.

"What am I missing?"

Farzad, Oct 20 at 6:53pm CT: I might be extremely dumb but I feel like CoreWeave's strategy is fundamentally broken.

His case: "They are slow bringing up their computer relative to SpaceX, they pay significantly higher interest on the debt, they charge significantly lower price, and they don't serve any of their own models." He closed: "What am I missing?"

That's Farzad's opinion, framed as a question. The post doesn't give figures for interest costs, pricing or how fast either company brings compute online.

Check it here.

"Mind-blowingly good"

At 10:31am CT Tuesday, Farzad RTd SpaceX: Watch live as SpaceX discusses our Q3 2026 earnings at 3:30pm CT today. The call itself is covered in our full breakdown of the SpaceX earnings call.

At 3:37pm CT, Farzad posted: THANK GOD FINALLY VIDEO EARNINGS CALL!!!!!!! A minute later: VIDEO EARNINGS CALL FOR TESLA AS WELL PLEASE!!!!!!!! At 4:20pm CT: This $SPCX call is mind-blowingly good. At 5:23pm CT he posted his own edit of the call, titled BREAKING: Elon Musk Makes Shocking Predictions.

Elon, 4:59pm CT, replying to aaronburnett: "Veterans of the cellular industry look at the chess board of existing pieces and don't see a solution." The post also says SpaceX "is developing entirely new pieces." Farzad quoted it at 5:05pm CT: Goodbye AT&T, Verizon, etc. That reading is Farzad's. Elon's post doesn't name any carrier.

On Wednesday at 7:52am CT, stevenfiorillo posted: I owe $SPCX and @elonmusk a public apology. It goes on: "I thought the valuation SpaceX was getting was insane. Q3 changed my mind." One of the figures listed: "$14.6B revenue vs about $12.9B expected". Farzad RTd it at 8:25am CT and replied: Hi Steve. The figures are stevenfiorillo's.

Check SpaceX here. Farzad's RT here. Farzad's live posts here, here and here. His edit here. Elon here. Farzad's quote here. stevenfiorillo here. Farzad's RT here and reply here.

"No longer impossible"

SawyerMerritt, Oct 20 at 3:57pm CT: Holy shit, Elon just said @SpaceX could exit 2027 with an ARR of somewhere between $500 billion to $1 trillion.

ARR is annual recurring revenue. Farzad quoted the post a minute later and asked $TSLA investors: "what do you think?"

Elon, 5:05pm CT: As crazy as this sounds, $1T in SpaceX revenue in 2028 is no longer impossible. Farzad RTd it at 5:09pm CT.

At 5:22pm CT, Farzad wrote that he's been using AI agents "EXTENSIVELY this year" and is now "completely and totally Grok Bot pilled." From there, he wrote: "it's quite easy to see how SpaceX can reach its $1 trillion revenue target by 2028."

amitisinvesting, 3:50pm CT, posted these as Elon's words: "WE COULD HAVE 20GW OF SI CAPACITY ONLINE BY NEXT YEAR, SMALL CHANCE BUT IT IS POSSIBLE" and "I WOULD BE SURPRISED IF WE BROUGHT ON LESS THAN 13GW OF COMPUTE BY THE END OF NEXT YEAR." Farzad replied at 3:59pm CT: He's gonna do it. He RTd the post too.

The $500 billion to $1 trillion ARR, the $1T for 2028 and the gigawatt lines are Elon's, as posted by Elon, SawyerMerritt and amitisinvesting. "He's gonna do it" and "quite easy to see" are Farzad's bets.

Check SawyerMerritt here. Farzad's quote here. Elon here. Farzad's RT here. Farzad's 5:22pm post here. amitisinvesting here. Farzad's reply here and RT here.

"Most addicting product I've ever used"

munster_gene, Oct 20 at 3:53pm CT: Elon says Grok Bot may over time use most of SpaceX compute. In other words, Grok Bot growth will be over the top.

Munster's take: "Today there are only about 15m daily personalized bots world wide." Then: "My guess is 5m Grok Bot. I expect 50m daily Grok Bot users in 6 months." Farzad RTd it at 3:58pm CT.

SawyerMerritt, 3:59pm CT: SpaceX has announced that Grok @Bot now has 800,000 paid weekly users as of earlier this month, more than doubling from the previous month. Farzad RTd it at 4:00pm CT.

The 15m, 5m and 50m are Munster's estimates, and the 50m is a forecast. The 800,000 is SpaceX's figure, as SawyerMerritt reported it.

Farzad, 3:51pm CT: Grok Bot stickiness is easy to see if you use the product. It's bananas good.

At 11:18pm CT he went further: Grok Bot is probably the most addicting product I've ever used. He wrote that you get "real, actual economical results" if you build "a good, repeatable, high-quality process," and that Grok Bot "makes entrepreneurship into a video game". His open question: "will the money people make from Grok Bot offset the money they spend on tokens?" If yes, "trending towards abundance." If no, "trending towards dystopia." His last line: "It feels like some sort of endgame."

CharlesSteadman replied to Farzad at 11:30pm CT that it's "certainly addictive" but sometimes "just plain hopeless". Farzad, 11:35pm CT: Yeah my incredible to hopeless ratio is probably 10:1 right now. That's down from 100:1 lol.

Earlier, at 1:33pm CT, nerdalert shared a summary Grok Bot "pulled together" after its "first ~24 hours." The first item: "Got about $1,000 back from an airline". That claim is nerdalert's. Farzad RTd it at 2:00pm CT and replied: Grok Bot is so good.

Check munster_gene here. Farzad's RT here. SawyerMerritt here. Farzad's RT here. Farzad's posts here and here. CharlesSteadman here. Farzad's reply here. nerdalert here. Farzad's RT here and reply here.

Earlier Tuesday

At 9:30am CT, SawyerMerritt posted a quote from Elon's new interview with Walter Isaacson, including: "My so-called wealth is simply a percentage ownership in SpaceX and Tesla." Farzad quoted it at 10:20am CT: Elon's best quality is that he can (eventually) honestly self-analyze and improve. Bodes well for his leadership position in AI. That's Farzad's read.

At 11:28am CT, Farzad wrote that he's "starting to think that the end state for AI will be specialized models at the individual human level". He extended that to organizations, companies and teams, with models that "RL/RSI constantly," shorthand for reinforcement learning and recursive self-improvement.

ALScyborg, 12:07pm CT, on The Boring Company: "A Tesla-powered tunnel truck that carries a standard 20-ft shipping container." The post also says "The Vegas Loop is heading to 28 stations by year end" and "Two Prufrock machines are digging Nashville's Music City Loop at once." Those are ALScyborg's claims. Farzad RTd it at 1:19pm CT.

At 2:46pm CT, Farzad posted: Once the context/memory issue is solved with agents, it's going to get really, really interesting. His projection: "An agent that never forgets, always understands context, and always knows what to do next is probably not that far away."

Check SawyerMerritt here. Farzad's quote here. The 11:28am post here. ALScyborg here. Farzad's RT here. The 2:46pm post here.

New Farzad videos

Farzad posted a long run of videos Tuesday evening. At 5:37pm CT, check Why Amazon Blocks AI Shopping Agents: They Never Scroll here. At 5:38pm CT, check Why Tesla Says It Ditched Hand-Written Code in FSD v12 here.

Check Elon: SpaceX Could Hit $500B to $1T a Year by 2027, at 5:51pm CT, here. SpaceX Went From a Loss to $2.8B Operating Profit in 3 Months, at 6:21pm CT, here. SpaceX's AI Business Grew 830% in One Year, at 6:51pm CT, here. Elon: Phone Carriers Don't See SpaceX's New Chess Pieces, at 7:21pm CT, here. Elon: Old Nvidia H100s Now Rent for More Than When New, at 7:51pm CT, here.

Check Elon Just Described the Megapod on a SpaceX Earnings Call, at 8:21pm CT, here. Elon's Target: 100x Today's Starlink Bandwidth in 3 Years, at 8:51pm CT, here. Elon's New Floor: A 13 Gigawatt AI Compute Target for 2027, at 9:21pm CT, here. Elon: SpaceX Will Try to Catch the Whole Starship Next Month, at 9:52pm CT, here.

Check SpaceX Is Casting Turbine Blades to Skip a 12-18 Month Wait (Elon's Claim), at 10:23pm CT, here. SpaceX's Chip Fab Timeline: Test Chip 2027, Volume 2028, at 10:59pm CT, here. SpaceX Says One Starship Launch Did the Work of Ten Falcon 9s, at 11:24pm CT, here.