Our juggernaut of an IPO report — 60-plus pages and a new kind of IPO coverage for Epistrophy Capital Research — is available to customers by request $5,000.

Below is an excerpt from that SpaceX IPO report.

The announcement Friday was stunning. SpaceX (SPCX: Nasdaq), just a week from the planned IPO announced a deal, with may caveats, that would let it rent out xAI unused computing power for up to $920 million per month to Alphabet (GOOG: Nasdaq) — which owns 6.1% of SpaceX. Three weeks before that, it had signed a caveats-filled $1.25 billion per month deal with Anthropic (14% owned by Alphabet).

Nevermind that the Alphabet deal is a “nine-month contract [with] more easy outs than a kid’s t-ball game,” as investor James Chanos tweeted. It’s a dramatic pivot of xAI from articial intelligence provider to server farm. It’s as much as $2.17 billion per month — $26 billion annualized — renting out data centers that Grok, xAI’s own AI model, couldn’t fill.

These are not signs of a thriving AI business. They are a pivot away from its failure — and an attempt to obscure that most of what SpaceX calls “AI revenue” traces to Twitter advertising, and most of what it calls “AI debt” financed a social network. The question the S-1 cannot answer: will xAI’s revenue growth ever resemble that of OpenAI or even DeepSeek?

xAI launched in 2023 as a credible challenger. Grok positioned itself explicitly as the anti-woke alternative — “maximally truth-seeking,” willing to answer questions rivals refused, even sharing a screenshot of a user asking how to make cocaine as a product demonstration. Twitter users, Tesla (TSLA:NASDAQ) drivers and right-wing social media circles flocked to the free service. Real-time Twitter data gave Grok a training advantage no competitor could replicate. Early growth rates were compelling.

Despite entering a market with 8-to-12-month GPU waitlists, xAI impressively assembled 555,000 Nvidia (NVDA: NASDAQ) GPUs — each H200 trading at up to $46,000, rental prices still rising in 2025, the opposite of normal hardware depreciation. Colossus 1 launched in 122 days. Colossus 2 followed. Total hardware cost: $18 billion. Total 2025 AI capex: $12.7 billion. The infrastructure was world-class.

The over-build was massive because the underlying product, Grok, was failing.

What is SpaceX’s favorite measurement of Grok? GPQA. GPQA Diamond is a 198-question multiple-choice benchmark in physics, chemistry and biology published in November 2023 by researchers at NYU, Cohere and Anthropic — filtered specifically to exclude anything solvable by web search. When it launched, PhD-level experts scored 65–70%. OpenAI’s GPT-4, the most capable model in the world at the time, scored 39%. The test was designed to find the ceiling. It found one.

The SpaceX S-1 cites Grok’s GPQA Diamond performance six times as evidence of frontier scientific reasoning. But there’s a reason the never mention the GPQA grade. It’s weak.As of April 2026, Anthropic’s Claude sits at 94.6%, Alphabet’s Gemini at 94.3% and OpenAI’s models at 93.6%. Grok is at 88%. It entered the leaderboard in early 2025, over a year after the other three labs, climbing impressively from 67% to 88%. But Claude moved from 83% to 94.6% over the same period. DeepSeek, which entered even later, has already pulled ahead at 90.1% — spending a fraction of xAI’s $12.7 billion in 2025 AI capital expenditure. The company that burned the most money is fourth on the test it chose to feature in its own prospectus. The S-1 does not mention DeepSeek once.

The gap showed up in user behavior.

Source: SimilarWeb, April 2026

Grok claims 117 million monthly active users, but the S-1 acknowledges the figure is inflated by passive X integration. Only 1.9 million subscribers pay for SuperGrok, a conversion rate of 1.6 percent. The S-1 flags “periods of decreased Grok app downloads and user activity” as a known business risk. In January 2026, Anthropic discovered that xAI’s own engineers had been quietly accessing Claude through the Cursor IDE and cut off their access. The people building the competitor to Claude were using Claude.

xAI tried everything to drive usage. In July 2025, system prompts encouraging “politically incorrect” claims produced antisemitic posts praising Hitler within days. Musk’s response: “Never a dull moment on this platform.” In December 2025, image generation launched; within days users were generating sexualized images of women and posting them publicly. Over nine days: 1.8 million sexualized images of women, 23,000 of children. Grok posted its own apology on X. xAI’s response to CNBC: “Legacy Media Lies.” Regulatory investigations opened in France, Malaysia, India, the United Kingdom and the European Union. An Amsterdam court banned the feature outright, threatening daily fines of €100,000.

On March 12, 2026, Musk posted: “xAI was not built right first time around, so is being rebuilt from the foundations up.” That same week: “Grok is currently behind in coding” — the most commercially valuable AI capability of the moment. The billions were already spent. Ten of xAI’s eleven co-founders had by then departed, including Jimmy Ba, whose research was foundational to Grok’s architecture, who left amid reported tension over Musk’s demands for faster performance gains. The people who designed the machine being rebuilt are no longer there to rebuild it.

The S-1 calls this an AI company. The revenue says otherwise. In 2025, the AI segment generated $3.2 billion in total revenue — of which $1.8 billion came from Twitter advertising. Strip out Twitter subscriptions and data licensing and Grok’s standalone revenue shrinks to a fraction of the headline number.

Source: SPCX S-1

The debt picture is worse. Musk bought Twitter in 2022 for $44 billion, loading it with $13 billion in leveraged debt at rates as high as 13%. Twitter advertising peaked at $4.5 billion in 2021 and had collapsed 62% to $1.7 billion by 2024. He transferred the platform and all its debt onto xAI in March 2025. xAI then added $5 billion of its own debt at 12.5%. SpaceX retired all $17 billion with a $20 billion bridge Goldman Sachs (GS:NYSE) loan at 4.58%. The IPO proceeds are contractually required to repay that bridge within six months of closing. This is, in material part, a Twitter bailout dressed as an AI IPO.

With Grok unable to fill the data centers, SpaceX signed Anthropic at $1.25 billion per month three weeks before the IPO filing, and Google at $920 million per month two days before Amendment No. 2. Both contracts carry 90-day mutual termination clauses. As Chanos noted, Nvidia could simply rent its GPUs directly and undercut the entire model. The infrastructure meant to power a frontier AI model is now a landlord play.

The potential solution is 25-year-old Michael Truell. He started coding at 11, interned at Google and Two Sigma as a teenager, won a national programming prize at 17 and dropped out of MIT to build Cursor (private) — reaching $100 million in ARR in January 2025, $2 billion by February 2026 and $3 billion by May, all with 300 employees, faster than any software company in history according to Fortune and Bloomberg. SpaceX has an option to acquire Cursor for $60 billion in SPCX shares within 30 days of the IPO, or pay a $10 billion breakup fee. A CEO of one of the largest public AI companies told us: “Truell is the real deal — he could make that entire xAI business work. And if they lose him they might not have another shot.”

SpaceX entered 2024 with $791 million in net income. It ended 2025 with a $4.9 billion net loss — the difference being xAI. The AI segment generated $3.2 billion in revenue against $6.4 billion in operating losses. The product it was built to power ranks fourth on the benchmark the S-1 cites six times as proof of frontier performance. The engineers who built it were using a competitor’s product. The founders who designed it have left. The infrastructure meant to be a moat is now being rented to the companies whose models xAI’s own staff preferred. Elon Musk set out to build the machine that understood everything. What he built was a very expensive data center looking for customers.

Oracle (ORCL: NYSE) reports fiscal fourth-quarter 2026 results June 10 after market close – during a San Francisco Giants day game at Oracle Park, no less.. Wall Street, via 34 analysts, expects $1.96 in non-GAAP earnings per share on $19.10 billion in revenue. The range on EPS runs from $1.76 to $2.06.

Who cares?

This quarter is so very important in the history of the AI buildout and, indeed, for Oracle’s product stack and execution that Wall Street’s favorite metrics couldn’t be more useless. 

The uncertainty is not about demand. It is about whether Oracle can convert a $553 billion backlog into revenue as fast as management has promised.

The company entered Q4 on the strongest run in its four-decade history. Fiscal third-quarter revenue of $17.2 billion grew 22 percent year-over-year, the first quarter in more than 15 years in which both organic revenue and non-GAAP EPS grew at 20 percent or better. 

Oracle Cloud Infrastructure revenue reached $4.9 billion, up 84 percent. AI infrastructure revenue grew 243 percent. Multi-cloud database revenue grew 531 percent. 

The big number? remaining performance obligations — contracted future revenue — which stood at $553 billion at quarter’s end, up from $138 billion a year earlier. Demand clearly exceeds supply – but how quickly can Oracle book what it  has supplied?

That supply constraint is the central technical and financial problem Oracle is spending its way through. We were lucky enough to attend an off-the-record OCI Analyst Summit at Oracle’s Nashville Headquarters in May. Mahesh Thiagarajan, executive vice president of Oracle Cloud Infrastructure, said the company had signed multi-gigawatts of data center capacity before competitors recognized the scale of AI infrastructure demand.

That commitment, made in October 2023, explains a backlog that has quadrupled in three quarters without a commensurate acceleration in reported revenue — the gap between what is contracted and what has been built and handed over to customers.

Building infrastructure at this scale requires solving problems that have nothing to do with software. Oracle has secured more than 10 gigawatts of power and data center capacity coming online over the next three years, with more than 90 percent funded through partners rather than Oracle’s own balance sheet, OCI 

Thiagarajan explained the architecture of that funding: Oracle designs the facilities, dictates the technical specifications, controls land and power arrangements and takes title only after the asset is complete and revenue-generating. Partners absorb construction risk. In other words, Oracle is literally building their own data centers using someone else's capital.

The result is data center capacity that stays off the balance sheet until it earns revenue.

Financing that pipeline has not required Oracle to seek capital on unfavorable terms. A $14 billion raise — $10 billion from PIMCO with the remaining $4 billion oversubscribed three times and closed overnight — illustrates the inbound demand. The giant slug of cash from PIMCO – arguably the smartest money on Wall Street – shows that big investors are hunting Oracle. Oracle isn’t hunting them. 

The company has announced intent to raise up to $50 billion in additional debt and equity through calendar 2026.

The operational metrics inside the pipeline are the leading indicators the market should watch on June 10. Oracle delivered more than 400 megawatts of data center capacity in Q3, with 90 percent delivered on or ahead of schedule. Between Q2 and Q3, the company tripled manufacturing sites, quadrupled rack output and cut the time from rack delivery to revenue recognition by 60 percent. Gross margin on AI capacity delivered in Q3 held at 32 percent, above the company’s own 30 percent guidance. 

They’re scaling a giant, capital-intensive business and scaling-profitability. 

Q4 could show more of the same. A Q4 print showing sustained 32 percent or higher AI-capacity margins, alongside megawatt delivery north of 500, would validate the unit economics of the build. Slippage on either would not.

The supply chain beneath that construction program carries risks the earnings slide does not capture. TSMC remains, in Thiagarajan’s characterization, the one industry-wide bottleneck for hardware supply — a constraint that limits every hyperscaler regardless of capital position. Oracle manages more than 60 rack configurations across varying power densities to preserve flexibility as GPU generations turn. The company is paying to install battery energy storage measured in thousands of megawatts against 15-year grid contracts — a capital commitment designed to smooth AI workload power draw and buy goodwill with utility regulators, at a cost Thiagarajan put in the hundreds of millions. 

The U.S. electrician workforce, roughly 900,000 people with an average age above 50, is a structural constraint on how fast anyone can build. Oracle says it employs 50,000 people indirectly through construction activity today.

The RPO composition question will define the post-earnings conversation. A figure above $600 billion would confirm the demand trajectory; a sequential deceleration would need explaining even if year-over-year growth remains extraordinary. And everyone can agree they have enough OpenAI RPOs. 

Any suggest that non-OpenAI RPO is larger than what the market has modeled would be good news showing diversity of customers and risk in the rapidly evolving AI world. .

For Q4 specifically, Oracle guided cloud revenue growth of 40 to 44 percent in U.S. dollars. OCI has exceeded the headline cloud guidance range in each of the past three quarters.

The FY2027 CapEx disclosure deferred from Q3 will arrive with Q4 results. On the March call, CFO Doug Kehring declined to provide FY2027 CapEx guidance, saying the company would return to it after fiscal year-end. That number — and the framing of how much of it will be funded through partner structures versus Oracle’s own capital — is the fulcrum on which the FY2027 revenue target of $90 billion turns. At OCI’s stated Q3 AI gross margin of 32 percent, that scale of revenue would produce an infrastructure earnings stream with no direct precedent in Oracle’s history.

And what of software?

The database franchise, which built the company and is financing the infrastructure bet, has not stood still. Oracle’s May 12 announcement of a production-grade MCP server — delivered at no incremental charge through OCI Database Tools, using OAuth 2.0 identity propagation across Oracle Database, Oracle Database@Azure, Database@AWS and Database@Google Cloud — extends Oracle’s access layer into enterprise AI agent deployments. The economics are deliberate: charge nothing for the connector, monetize the database traffic it generates. Oracle has embedded more than 1,000 AI agents across Fusion and industry applications. The database is the authenticated path to enterprise data. Every additional AI query that touches an Oracle database is revenue Oracle did not need to sell separately.

The market that gave Oracle a 27 percent single-day gain in April after a Bloom Energy press release noted an ahead-of-schedule fuel cell delivery has now priced in a version of the infrastructure story. What it has not priced in is the probability that the OCI revenue growth rate, which has accelerated from 54 percent in Q1 to 66 percent in Q2 to 84 percent in Q3, has further to run. If June 10 delivers a fifth consecutive quarter of accelerating OCI growth, the FY2027 $90 billion target stops looking like a forecast and starts looking like a floor.

My kids do not listen to me. But Charles Payne of Fox Business apparently does! He remembered an appearance last year where I talked about the then-quietly-booming optical photonics sector. It’s quiet no more, which gave me a chance to come back on his show to shed some more light on the sector (see what I did there?) I always take pains to say I’m not picking stocks, but when it works it’s nice to be seen.

SEE THE HIT HERE →

Earlier in the week Connell McShane of NewsNation had me on to talk about AI, Trump’s white house and, why not, the New York Knicks! 🏀 (Go New York Go!)

We’ll be in our San Francisco office at the Ferry Building all week. Last week a new reporter sent me an email asking me to quickly jump on camera for an interview, but I was in a meeting and missed it. Reporters! Feel free to call, any time: 415-672-4838. Or text or something!

Written reports are available to clients, with video summaries on YouTube, and of course our popular summaries of the summaries on Instagram, TikTok, and YouTube Shorts.

^{We certify that (1) the views expressed in this report accurately reflect our views about all of the subject companies and securities and (2) no part of my compensation was, is or will be directly related to the specific recommendations or views expressed in this report.}

^{Important disclosures}

^{Important disclosures are available by calling (347) 619-2489 or writing to Epistrophy Capital Research, One Ferry Building, Suite 201, San Francisco, CA 94105.}

^{Epistrophy Capital Research is an independent research provider and does not operate as a financial institution. Epistrophy explicitly does not provide investment advice, stock recommendations, or solicitations to buy or sell any securities.}

^{The research reports provided by Epistrophy Capital Research contain opinions derived from publicly available information, issuer communications, recognized statistical services, and other reputable sources considered reliable. However, Epistrophy does not independently verify the accuracy or completeness of such information and explicitly disclaims responsibility for any errors or omissions.}

^{Opinions and analysis contained within Epistrophy's research reports are current only at the time of publication and are subject to change without notice. Readers must independently verify facts and conduct their own due diligence before making investment decisions.}

^{Epistrophy Capital Research does not consider or evaluate individual investor circumstances, including investment objectives, financial situations, or risk tolerance. Investing in securities, particularly small-cap and micro-cap stocks, involves substantial risks, including significant volatility and potential loss of principal. Readers are strongly advised to consult their financial advisor or another qualified professional before acting on any information provided. Readers should assume Epistrophy Capital Research, its principals or its contributors may have positions, long or short, any of the companies discussed and the Epistrophy Capital Research principals or contributors may have had or currently have business interests in the companies discussed.}

^{Past performance referenced in Epistrophy reports is not indicative of future results. Security prices can fluctuate widely, and investors should be aware that investments can result in significant financial losses. Epistrophy Capital Research or its Unless explicitly stated otherwise, prices quoted in reports reflect market closing prices from the previous trading day.}

^{Epistrophy Capital Research publications are intended solely for direct recipients and should not be redistributed or shared with third parties without explicit permission from Epistrophy Capital Research LLC.}

^{Epistrophy Capital Research reports are provided strictly for informational purposes and do not constitute a comprehensive analysis of any company, security, or industry. No content within these reports should be considered accounting, tax, legal, or professional advice.}

^{Links or references to third-party websites or external resources are provided solely for informational convenience. Epistrophy Capital Research expressly disclaims endorsement of, and responsibility for, the content, accuracy, or reliability of such external information. Accessing third-party information is done entirely at the user's own risk.}

^{For additional details, clarification, or specific inquiries regarding Epistrophy Capital Research reports, please contact Epistrophy Capital Research LLC directly.}