Why OpenAI Is The Next Amazon - GoodAlexander
Summary
GoodAlexander expects SpaceX’s IPO to trade well while draining liquidity from almost every adjacent speculative asset. The next year may bring roughly $200 billion of issuance across three coveted companies, versus $130–140 billion raised by crypto VC over five years. SpaceX could represent a roughly $25 billion slug, with about 30% allocated to retail versus the usual 5–10%. His base case: “SpaceX itself trades well, every other speculative asset trades down.”
The SpaceX valuation rests less on current growth than on Elon Musk’s ability to make investors “dream the dream” of AI data centers in orbit. Musk has filed to launch one million satellites and, in GoodAlexander’s account, proposes putting compute equivalent to 20% of US power supply in space, where cooling and local resistance are less constraining. The thesis may take years to falsify, which is precisely why “the sun is really big” works as an equity narrative.
His preferred SpaceX trade is short legacy telecom, not short SpaceX or Tesla. Satellite-heavy telecom baskets had risen roughly 120%, while some slow growers were up around 180%; if Musk’s architecture works, incumbents lose leverage across connectivity and compute. He refuses to short Tesla absent Chinese EV access to the US: “I try not to short things that could invent the cure for cancer.”
OpenAI is GoodAlexander’s favored AI IPO because he sees the same monetization blind spot that investors once had with Amazon. Amazon’s retail operation was once marked at zero, yet 300 million active shoppers supported a $69 billion-a-year advertising run rate by Q1 2026; OpenAI, he says, has 900 million ChatGPT DAUs. A six-week ad pilot reportedly generated $100 million against a $2.4 billion revenue target, making ads the overlooked option embedded in its consumer distribution.
Anthropic’s Fable model may be technically exceptional while its reported economics remain less attractive than they look. GoodAlexander says Anthropic highlights a non-GAAP operating-profit measure despite remaining deeply free-cash-flow negative, and treats model-training capex as though it were temporary even though stopping investment would destroy future competitiveness. Thread Guy says Fable finds many bugs in GPT-5.5 code that Opus 4.8 missed; GoodAlexander agrees its diagnoses are generally right.
The deeper enterprise-AI risk is that model costs rise while generated software cannibalizes the market it is meant to serve. Two Fable agents may cost around $100 an hour—effectively another full-time engineer—while customers such as Uber spend heavily without producing an obviously better app. Consumer commerce looks greener because models understand intent, while frontier cryptography, biology and mathematics remain the strongest cases where AI can enlarge rather than replace an end market.
The political-market overlay creates a liquidity warning and a possible threat to Palantir. GoodAlexander speculates that official intervention may have supported S&P futures, while stablecoin supply remains stalled near $180 billion. He treats the stablecoin stagnation as a warning for crypto and non-AI risk assets, and sees Palantir as exposed Trump beta trading poorly. Anthropic’s “forward-deployed” work with Goldman Sachs also tests Palantir’s core claim that LLMs require its ontology: “What if Karp’s wrong?”
Deep dive
1. SpaceX is a liquidity event before it is a fundamental bet
GoodAlexander’s starting point remains market plumbing: crypto VC raised roughly $130–140 billion over five years, while the coming year could contain about $200 billion of equity issuance from SpaceX, OpenAI and Anthropic. “The market cannot go up” cleanly through that much supply from three of the world’s hottest companies.
SpaceX is unusually consequential because roughly 30% of the offering may go to retail, versus a normal 5–10%. GoodAlexander describes roughly a $25 billion slug coming out, capable of “suck[ing] the liquidity out of every speculative asset,” irrespective of whether the company ultimately executes.
The setup was already overheated: GoodAlexander points to an S&P telecom ETF containing satellite exposure that rose approximately 120% in a year. As a former telecom analyst, he regards these as historically dull companies whose operating performance did not justify a moonshot.
Thread Guy’s pushback—worth keeping—is that recent Broadcom and Oracle results showed “beating” is insufficient when expectations are extreme. GoodAlexander agrees that a SpaceX decline would be “horrendously bearish,” but thinks every interested bank, issuer and AI company is incentivized to ensure the flagship offering succeeds.
2. Orbital compute is the narrative underwriting the multiple
SpaceX may be growing around 30%—perhaps nearer 15% before recent revenue deals—while Anthropic and OpenAI are described as growing 500–800% and roughly 500%, respectively, at perhaps half SpaceX’s valuation. That mismatch forces the question GoodAlexander asked after watching Musk with Jamie Dimon: “What do you got, Elon? Like, what’s the pump?”
The answer is the sun. Musk has filed for one million satellites and, in GoodAlexander’s retelling, wants to launch compute equivalent to 20% of US power supply into orbit: a dense pack of satellites “cooled by space,” avoiding terrestrial coolant logistics, power-price inflation and communities that do not want data centers nearby.
GoodAlexander puts the proposed prize at $300–600 billion annually and sees the Anthropic and Colossus deals as foreshadowing the pitch: Musk built terrestrial AI data centers, rented them to major users and now says he can repeat the model in space. One unnamed top investor captured the distinction: “I don’t know about SpaceX as a company, but as a stock, it’s one hell of a stock.”
The story’s advantage is its long verification window. GoodAlexander says Musk can spend nine months “running around saying how big the sun is,” while Thread Guy argues that failed reusable-rocketry economics or inadequate orbital structural integrity would be the kinds of evidence that could break the story. Both see near-term falsification as difficult.
3. The cleaner trade is against legacy telecom
GoodAlexander is short telecom because no listed “space beta” can compete with Musk’s version of the story. If SpaceX can beam connectivity and AI compute from orbit, incumbents such as T-Mobile face a supplier with “everyone by the balls”; slow revenue growth and shareholder-destructive management leave little fundamental cushion.
Thread Guy proposed the crypto analogy of a developer launching a second token: short Tesla when SpaceX arrives. GoodAlexander initially saw the appeal, then rejected it—Tesla remains one of the only liquid public expressions of robotics and autonomous driving, while telecom offers aging management teams, roughly 6% growth and stocks that had run as much as 180%.
His short-selling scar is Plug Power. After the CEO left, he tripled a short he believed must finally go to zero; the market welcomed the departure and the stock rose roughly 600% in a straight line. The lesson from his “five whys” review: “If this has been a scam for 20 years, what stops it from being a scam for 23 years?”
4. OpenAI has Amazon’s once-mispriced advertising option
The market frames Anthropic as the safer enterprise company, promising free-cash-flow positivity in 2027, while OpenAI is the frightening cash burner that needs an IPO before it runs out of money. GoodAlexander prefers OpenAI precisely because that consensus discounts the strategic value of its consumer audience.
His analogy comes from trading Amazon when investors owned it for AWS and assigned the retail business no value. Amazon later turned 300 million active shoppers into a $69 billion annual advertising run rate as of Q1 2026; OpenAI, he says, already reaches 900 million ChatGPT DAUs. “It’s really hard to get 900 million users.”
A six-week OpenAI advertising pilot reportedly printed $100 million amid heavy oversubscription, while the stated target is only $2.4 billion. GoodAlexander also praises OpenAI’s GPT-5.5 Instant search experience: unlike traditional search, it can consume ten pages simultaneously and synthesize them without forcing the user through ten sites and their ads.
The unresolved risk is trust. Anthropic can promise an ad-free product, while OpenAI must show that recommendations remain useful rather than becoming disguised shilling; GoodAlexander’s favorable specimen is Codex automatically choosing a Cloudflare tunnel for security—a commercially valuable recommendation that still feels justified by the task.
5. Anthropic’s product is stronger than its accounting story
GoodAlexander calls Anthropic’s profitability presentation “misleading,” because its press releases emphasize a non-GAAP operating-profit measure while the company remains deeply free-cash-flow negative. Model-training capex is not analogous to optional Coca-Cola capex: stop spending, and the next model upgrade can destroy today’s unit economics.
He therefore treats that investment as structural R&D, not a one-off adjustment. Anthropic still deserves credit for structurally better margins than OpenAI because most of its revenue is enterprise, but its claim to current profitability should not be confused with actual cash generation.
On the product, GoodAlexander reverses his initial annoyance with Fable. It refused a requested wallet-security analysis, yet once prompted correctly it proactively found issues such as a possible XSS exploit. Thread Guy says Fable found many more bugs in GPT-5.5 code than Opus 4.8 did, and GoodAlexander says GPT-5.5 generally agreed with its diagnoses.
The cost is severe: a typical engineer running two agents may spend about $100 an hour, or several hundred thousand dollars annually—roughly the price of another engineer. GoodAlexander’s honest tension is that Fable is “the alpha model” for deployment even though the economics invite cheaper Chinese replication and tighter corporate usage caps.
6. Enterprise AI may cannibalize itself while frontier science expands demand
GoodAlexander’s enterprise objection is causal: increasingly expensive models make software easier to reproduce, which lowers demand for packaged software and games. If everyone can customize an application, the models are “cannibaliz[ing] their own end market” even as vendors celebrate how much code AI generated.
Thread Guy’s Uber example becomes the reality check. The company can spend an enormous sum on model credits, and Salesforce can say AI wrote 45% of its app, but GoodAlexander asks whether either product is detectably better than two years ago: “I can’t tell, man.”
The better case is work that previously required scarce specialists. Orchard, Zcash’s complex cryptography system released in 2022, allegedly carried bugs for years because so few people could understand it; GoodAlexander says a frontier model can inspect the implementation, identify the exact flaw and potentially improve the underlying algorithm.
That logic extends to biotechnology and mathematics, where AI might enlarge the discovery frontier. It does not make public biotech an easy trade: Ginkgo Bioworks, ticker DNA, is his cautionary example—down roughly 99%—and the best private longevity assets are often bought early by ultra-rich investors who value health more than returns.
7. Political support may be real without being easy to monetize
GoodAlexander labels his own claim controversial: he speculates that US equity and oil futures have seen intervention, possibly involving Scott Bessent. His evidence is behavioral—during geopolitical escalation, unusually large buyers repeatedly appeared around a 0.5% S&P decline, preventing the gaps and cascades familiar from Ukraine in 2022 or COVID.
His inference remains explicitly a conspiracy theory, not proven fact: Bessent trained under Stanley Druckenmiller, Trump is intensely focused on the market, and the price action looked unlike prior risk-off episodes. GoodAlexander expects later administrations may investigate “what the hell is going on.”
Even obvious political information is not automatically profitable. He cites Hegseth buying defense stocks that subsequently collapsed, while Thread Guy compares the game to chasing CZ-endorsed memecoins: each successive endorsement may create a weaker, shorter-lived pump.
The harder warning is stablecoin supply, stuck near $180 billion when policy advocates expected hundreds of billions. Thread Guy recalls a possible $10 trillion Scott Bessent plan, but GoodAlexander’s stated concern is that USDT and USDC have flatlined and the Clarity Act has not passed. If supply does not resume growth, he sees that as a “canary in the coal mine” for crypto and non-AI risk assets.
8. Anthropic could attack Palantir’s ontology moat
Anthropic’s Goldman Sachs language caught GoodAlexander’s attention because it describes placing “forward-deployed” engineers inside banks—Palantir’s own terminology. After Alex Karp publicly attacked Dario Amodei’s employment warnings, GoodAlexander sees motivation for Anthropic to challenge a company whose deployments it may already help power.
Palantir’s ontology originated in a real intelligence problem: agencies could not share information, and even shared descriptions—“a guy in a white outfit with a Jeep in Afghanistan”—matched too many people. Its software attached identities and relationships to data, then transplanted that structured layer into commercial AI deployments.
Karp’s current claim is that LLMs do not work reliably out of the box; they need Palantir’s ontology. GoodAlexander’s open question is whether stronger models, vector databases and automated tagging make that layer unnecessary: “What if you just need a Snowflake database” that Fable or Mythos can organize in a day? If so, he imagines an 80% stock decline.
Having worked at Palantir for two years, GoodAlexander still calls Karp a fantastic “CEO whisperer”—interesting enough to captivate executives afraid of being boring, and skilled at combining “spike talents” who excel at one thing. That same combative charisma becomes dangerous if it starts a war with the company’s most important AI supplier.
9. SpaceX needs the dream; OpenAI needs proof that users like the ads
Into the listing, GoodAlexander watches whether retail can understand, articulate and amplify the orbital-compute story. At roughly 90 times sales, the offering needs Musk, Jamie Dimon and sell-side research to sustain a credible dream; Thread Guy says evidence against reusable rockets or space data centers would be the clearest way to break it.
His base case is near-term strength but poor long-term value: execution risk is already priced in, current growth is too slow, and performance becomes path-dependent on xAI. If subsequent Composer models combine state-of-the-art quality with speed and privileged compute, he would hold; if xAI “is not a thing,” SpaceX could drift lower.
He remains broadly bullish on both AI labs because Fable and Codex are “some of the best tech products I’ve ever seen,” growing at perhaps 8× despite 30–40-times-sales valuations. His stronger preference remains OpenAI, contingent on repeatable ad pilots and evidence that advertising does not erode DAUs or user trust.
Google is the relative fade: it is the consensus buy-side long because TPUs, cash flow and cloud backlog look obvious, yet the stock is not trading well and Thread Guy says the backlog “looks like fraud” and Gemini is poor for coding. Thread Guy closes with an $80 billion Google equity raise—larger, he says, than Anthropic’s last round, OpenAI’s last round and the SpaceX IPO raise—as another major source of equity supply.