The Large Language Model Race

Introduction

I have a recurring dream where I'm running as fast as I can, but not moving forward. When I look down, I realize I'm on a f***** treadmill, and I can't get off no matter how hard I try. With a finger on the pulse of stagnation anxiety, I took a look at the state of the large language model industry (and saw a bit of my own mania reflected back.)

We've covered the Red Queen Race before so I'll spare the full recap. The short version: since ChatGPT, every frontier lab has been sprinting just to stay in place. Player A drops a superior model, Player B leapfrogs it, Player C ( emerges and shakes them both up covered that as well with our DeepSeek piece), and the techniques get absorbed within weeks. Repeat ad nauseam with players D, E, F, etc. rotating in. The Red Queen doesn't care where the new trick comes from; she just sets the pace higher.

This piece isn't about the race itself. It's about the runners' economics when nobody can stop running. To get there, we'll talk about airlines.

The Cycle in Practice

OpenAI is spending record levels of capital to stay in the running. The company generates roughly $2 billion monthly in revenue but remains unprofitable. Elon Musk, after splitting from OpenAI, created xAI and merged it into SpaceX ahead of SpaceX's June 2026 IPO. Musk's sway over markets has kept xAI in the game, but the economics are uh not necessarily encouraging. The hyperscalers are creating their own models, funding other model creators, and doing it as off balance sheet as possible.

They’re sprinting as fast as they can to stay in the same position. Stuck on a treadmill, reaching impressive personal records, but not exactly moving the needle when everyone else is simultaneously beating their own personal records.

And there's no obvious way off. During the Web 2.0 era, the differentiator was network effects: as Facebook grew, MySpace died. Winner-take-all dynamics. AI model providers haven't found an equivalent mechanism yet. API ecosystems create some switching costs, enterprise contracts generate stickiness, and ChatGPT has over 900 million weekly active users. But a new model from a competitor can (and regularly does) pull developers and enterprise customers away, if only until the next round of releases. None of these moats approach the self-reinforcing strength of a real network effect. The treadmill keeps spinning, everyone keeps sprinting, and the entire cohort stays in place.

The Airline Parallel

What do the economics actually look like if nobody can build a moat and everybody keeps sprinting? There's an industry that already knows the answer.

We've been considering the structural similarities between the language model landscape and the airline industry. David Oks argues in the wake of Spirit Airlines' shutdown that airlines destroy investor value in the aggregate. The IATA's 2026 outlook projected a 6.8% average return on invested capital, against an 8.2% weighted average cost of capital, aka the industry as a whole earns less than it costs to fund. Oks notes the standard explanations (vulnerability to shocks like 9/11, the 2008 financial crisis, Covid-19, or current oil price instability from the war in Iran) don't necessarily track. Other industries are hit by the same shocks without being structurally unprofitable. In any case, the actual figures ended up worse than projected. IATA's June 2026 revision (reflecting the fuel price spike from the war in Iran) cut projected profits from $45 billion to $23 billion and dropped the return on invested capital to 4.3% against a cost of capital of 8.5%.

Let’s chat through the concept. Airlines carry enormous fixed costs (planes, gates, labor contracts, fuel), offer almost no fundamental product differentiation (a seat from JFK to LAX is a seat from JFK to LAX), face volatile demand, and compete fiercely on the same routes. Every time one carrier tries to raise prices or earn decent margins, rivals undercut them and push the industry back toward low or negative returns. The product keeps getting better (cheaper fares, more routes, better booking technology) but the producers keep losing money. The benefits flow to consumers, not shareholders. Like the Oracle himself noted, “the worst sort of business is one that grows rapidly, requires significant capital to engender the growth, and then earns little or no money. Think airlines. Here a durable competitive advantage has proven elusive ever since the days of the Wright Brothers. Indeed, if a farsighted capitalist had been present at Kitty Hawk, he would have done his successors a huge favor by shooting Orville down.” Hmm, does that description remind you of any other types of businesses?

An obvious objection is that AI models aren’t identical the way airline seats are. GPT-3.5, Claude Opus 4.6, and Llama 3.1 have measurably different capabilities. A developer choosing between them isn’t just picking variations of a commodity. The problem with the objection is timing. In 2023, the gap between frontier models was large enough that picking the wrong one could mean the difference between a working product and demo ware. But these days, the gap has narrowed to the point where the median enterprise use case (customer service, document processing, internal tooling) can be served adequately by any of the top five or six models with the right combination of harness and context. Differentiation exists at the frontier, for the hardest tasks, but the frontier is where the smallest share of revenue comes from.

The bulk of the market is converging toward commodity territory. Look at the shift in discourse from "token maxing" to switching to open source models and managing token spend (a new take on tokenomics). Airbnb moved parts of its AI stack onto Alibaba's Qwen, a Chinese open-source model, for customer service (cutting resolution times from three hours to six seconds thank you very much). A US House probe followed, and Chesky earned himself a visit to DC to defend the decision, eventually appearing on Bloomberg TV, telling lawmakers they were "misunderstanding" the technology. Benchmark convergence tells a similar story: the gap between the top-performing pure play lab model and the widening world of open source models on standard benchmarks has never been narrower. Airlines went through the same progression: flying used to be a noticeably different experience across carriers…and then it wasn't. The "GPT wrappers" once dismissed as undifferentiated are where value is accruing, precisely because the model layer beneath them is commoditizing. This mirrors the airline dynamic: when flights get cheap, hotels, tourism platforms, and travel agencies thrive. When models get cheap, the developers and companies building on top of them benefit. The producers of the commodity layer do not.

These brittle economics create a dependency on the state, much like airlines. Airlines need subsidies, bailouts, and protected routes to survive. AI model providers increasingly need government contracts, regulatory favor, and state-aligned capital to sustain operations while operating across borders in ways that create tension between national interests and global markets. On May 2, 2026, Spirit Airlines announced an orderly wind-down of its operations, effective immediately. Overnight, 17,000 employees lost their jobs, and thousands of passengers were stranded. Spirit hadn't turned a profit since 2019, filed for bankruptcy twice, saw a $3.8 billion merger with JetBlue blocked, and couldn't absorb the spike in fuel costs from the war in Iran. The gov't offered a last-ditch bailout of up to $500 million for a potential 90% government stake, but bondholders rejected it and the airline folded.

Spirit was the smallest, most cost-constrained player in a structurally unprofitable industry. It survived as long as it did by ruthlessly minimizing costs, but when an exogenous shock arrived, there was no margin left to absorb it. The question for AI: which provider is the Spirit Airlines of language models? Which company is operating on thin enough margins that the next compute price shock, the next shift in GPU supply, or the next regulatory curveball leaves them grounded?

The IPO Window

The second half of 2026 is providing the AI industry something it has never had before: public financial disclosures.

SpaceX (with xAI merged in) went public in June 2026 in the largest IPO in history, raising $75 billion at a $135-per-share price. The stock opened at $161, hit $225.64 within a week, and then did something airline stocks tend to do: it fell. As of mid-July, SPCX has dropped below its $135 IPO price and trades more than 40% below its peak. Five weeks from peak to below-IPO-price. The S-1 revealed that xAI's operating loss was ~$6.3 billion in 2025 on $3.2 billion in revenue. Starlink's $4.4 billion operating profit couldn't offset it. Investors got their first real look at the economics of a company with significant AI model exposure, and they do not seem to love what they saw.

Some might argue (and they'd be right) that even below its IPO price, SpaceX still has a market cap north of $1.5 trillion. Airlines don't trade at those multiples on their best days. So either the market is catastrophically mispricing these companies, or the airline analogy captures the cost structure but misses something about the revenue potential. I'm leaning toward the former, but I'll hold that thought until we see Anthropic's S-1. Anthropic and OpenAI have both filed confidential S-1s. When those go public, we'll finally learn whether the AI model industry's economics resemble a cloud platform (high margins, improving unit economics) or an airline (high revenue, higher costs, value destruction in aggregate).

The Shape of Consolidation

If the generative AI industry follows the airline industry’s dynamics, what can we expect? Consolidation among a select few providers that are large enough to make the economics work is virtually inevitable. These consolidated providers will have to comply with national and international regulations and differentiate through branding, scale, and financial engineering.

In the airline world, survivors tend to fall into two categories. The first are the flag carriers: state-backed airlines that enjoy subsidies, bailouts, and protected market access. Emirates (owned by the Dubai government), Qatar Airways (state-owned), Turkish Airlines (majority state-owned), and historically British Airways all benefit from governments that treat air connectivity as national infrastructure rather than an ordinary business. The second are the consolidated private carriers: American, Delta, and United, all products of decades of mergers and bankruptcy restructuring. They aren't state-owned, but they aren't independent of the state either: the federal government bailed out the entire U.S. airline industry during Covid to the tune of roughly $50 billion. Even the "private" carriers need the state when conditions deteriorate far enough.

The AI model industry appears headed toward a similar bifurcation, with one crucial addition. On one side, there are effectively flag carrier models: DeepSeek, backed by a Chinese hedge fund operating within China's broader AI ambitions, and Mistral, the Paris-based lab & European champion (also creators of the best model name, le chat). On the other side, American labs seek to become consolidated private carriers: large enough to sustain the economics, dependent on a handful of deep-pocketed corporate patrons (Microsoft for OpenAI, Google and Amazon for Anthropic), and increasingly entangled with government contracts and regulatory frameworks. Consider this soundbite from OpenAI CFO Sarah Friar, in a November 2025 interview, discussing government-backed financing for AI infrastructure: "The backstop, the guarantee, that allows the financing to happen, that can really drop the cost of the financing but also increase the loan-to-value, so the amount of debt that you can take on top of an equity portion." (Friar and Altman later walked this back, clarifying that OpenAI wasn't seeking a government bailout. My "I'm not asking for a bailout" shirt has people asking a lot of questions already answered by my shirt.)

Unfortunately for the elegance of our analogy, there's a third category the airline comparison doesn't cleanly capture: Google and Meta. These companies run frontier AI models but don't need them to be profitable as standalone businesses. Google can subsidize Gemini through advertising and cloud revenue indefinitely. Meta is looking to do the same. They're not on the treadmill in the same way because they own the gym (something SpaceX appears to have picked up on.)

Anyway, this is why our analogy still has some weight: Google and Meta aren't just building AI model businesses. They're building AI-augmented advertising and cloud businesses (Meta threw their hat in the cloud ring very recently). The standalone and pure-play AI model market (OpenAI & Anthropic's goal) resembles airlines. And it's the one where economic pressure determines who survives and on whose terms. The conglomerates set the pace of commoditization (by offering competitive models at subsidized prices), but they don't experience the consequences. The pure-play labs do.

The Conscience Problem (NOT CONSCIOUSNESS)

Okay, so the analogy is pretty good. Then we hit some complications. I’ll go deeper into those, which seems like a lot after 2,000 words on airlines. Bear with me.

In the Visible Hand trilogy, we covered an angle of this: Anthropic signed a $200 million contract with the Department of Defense through Palantir, which used Claude on classified networks. When Anthropic pushed back on the scope of military use (insisting on prohibitions against mass domestic surveillance and fully autonomous weapons), the relationship collapsed. Defense Secretary Pete Hegseth declared that AI models with "ideological tuning" had no place in the "Department of War." Anthropic's CEO responded that the company could not "in good conscience accede to their request." The result: the administration designated an American AI company a "supply chain risk" and barred federal contractors from working with it.

Airlines don't face this problem. An airline doesn't have ethical objections to where it flies. It doesn't refuse routes. It doesn't push back on passenger behavior after landing. The flag carrier model works for airlines because the product is politically inert: a seat is a seat. AI is not politically inert. An AI model that can write code can write malware. An AI model that can analyze intelligence can enable surveillance. An AI model that can generate persuasive text can generate propaganda. The terms of state sponsorship come with obligations about what the model can and cannot do, and those obligations may conflict with the company's stated values & strategy, its customers' expectations, or its developers' willingness to show up to work.

This scenario has interesting analytical ramifications outside of the drama: after all, Anthropic didn't exactly collapse after losing the DOD relationship. Less than four months later, it filed for an IPO at a $965 billion valuation, backed by a $65 billion Series H. It found a different path, not flag carrier status, but private capital at scale. That path is riskier. Private capital is more impatient than a government backstop. Venture investors and public-market shareholders expect returns on a timeline that a sovereign sponsor does not. Anthropic is betting that commercial revenue and corporate partnerships (circular financing if you prefer) can sustain the economics without accepting every state demand. Whether that bet pays off depends on whether the economics of AI models ever improve enough to generate real margins or whether today's structural unprofitability is permanent.

This is where the airline analogy breaks and reveals a fun, new twist to the analysis. The AI model industry faces a trilemma that airlines never did.

Option one: accept flag carrier status with all its constraints. Build your business around government requirements even when they conflict with your public strategy or values (you know, like Anthropic’s strange values of not having their product used for autonomous weapons), like defense contractors and state-operated corporations. This is the path of least financial resistance and most strategic friction. The economics work, but the product becomes inseparable from the state’s interests.
Option two: refuse flag carrier status, rely on private capital, and operate on thinner margins with less protection. This is Anthropic’s current bet — a $965 billion valuation backed by a $65 billion Series H, with an irrevocably altered relationship with the gov’t it walked away from. The economics are harder. Private capital is more impatient than a sovereign sponsor. Venture investors and public-market shareholders expect returns on a timeline that a government backstop does not. But the product remains yours to define.
Option three: the tension proves unresolvable, and the industry fractures along political lines like defense supply chains. Some models get approved for government use, others don't. Companies sort themselves into camps based on what compromises they're willing to make. We're arguably already watching this happen: the DOD's designation of Anthropic as a "supply chain risk" was the first visible crack, and the Fable episode widened it. In June, the Commerce Department opened the floodgates and issued an export-control directive forcing Anthropic to pull its two most advanced models offline globally aka the first time the US government caused a leading AI company to retract its systems from public use. The ban lasted two weeks before being partially lifted. If you're an enterprise customer choosing a model provider, congrats! You now have to account for geopolitical risk in areas you never even considered.

Okay Cool, But What If You're Not an AI Lab?

If you're an organization using AI models, the dynamics described here aren't academic. They're procurement risks. The model your business depends on today may be priced differently in six months, deprecated in twelve, or operated by a company that no longer exists in twenty-four.

When Spirit Airlines shut down, hotels and travel agencies that built packages around Spirit routes scrambled overnight. The AI equivalent: if your product hardcodes a single model provider's API, you're one terms-of-service change away from a scramble of your own.

Build abstraction layers.

Evaluate concentration risk, not just which provider you use, but whether that provider's economics are sustainable and whether their macro positioning creates exposure for your business. Treat your AI infrastructure the way we argue you should treat any digital supply chain dependency: with diversification, contingency planning, and a clear understanding of what happens when a key supplier goes under (or uh gets labeled a supply chain risk by the DOD).

(If you'd like help doing some of this analysis and infra work, please reach out: diego @ wayspire.com and we can chat)

Conclusion

The economics of the AI model industry look increasingly like the airline industry's. A small set of state-adjacent companies will survive bruising competition, enormous fixed costs, and regulatory scrutiny, while consumers and application-layer businesses capture most of the value. Cheap, powerful AI will likely be underwritten by states and deep-pocketed corporate patrons who view these systems as critical infrastructure rather than stand-alone businesses.

But the airline analogy, for all its explanatory power, breaks down at exactly the point that matters most. Airlines are infrastructure that doesn't think. AI models are infrastructure that does, or at least mimics thinking well enough that the distinction is politically irrelevant (debating consciousness is for a different piece entirely). That difference means the flag carrier model comes with a cost that air travel never imposed: the obligation to ideologically bake in your sponsor's interests into your product's capabilities, even when those interests make your engineers uncomfortable, your customers nervous, and your acceptable-use policies incoherent.

Anthropic's bet that private capital can replace state sponsorship is the most consequential experiment in the industry, and we will evaluate it in public once its S-1 goes live. If Anthropic can sustain itself commercially without compromising its position, it proves a third path exists between flag carrier obedience and extinction. If it can't...if the economics force it back to the negotiating table with the same governments it walked away from, then the airline parallel is far more than an analogy, it's uh…well it's a forecast.

Credits

Huge thank you to Noelia Veras for her help with research and editing . Outside of patiently listening to my rants about these industries, she put in a massive shift looking into the different companies & players as well as contributing both her writing and editorial voice to the piece.

References

Al Jazeera. (2026, July 1). US lifts restrictions on Anthropic's powerful AI models Fable and Mythos. Al Jazeera. Retrieved July 22, 2026 from [https://www.aljazeera.com/economy/2026/7/1/us-lifts-restrictions-on-powerful-ai-models-fable-mythos-anthropic-says](https://www.aljazeera.com/economy/2026/7/1/us-lifts-restrictions-on-powerful-ai-models-fable-mythos-anthropic-says) (Al Jazeera)
Buffett, W. (2008, February). Berkshire Hathaway 2007 Annual Letter to Shareholders. Reproduced in Yale School of Management Module 8 Reading. Retrieved July 22, 2026 from [https://som.yale.edu/sites/default/files/2021-12/Module8-Readng.pdf](https://som.yale.edu/sites/default/files/2021-12/Module8-Readng.pdf) (Yale SOM / Berkshire Hathaway)
CNBC. (2026, May 5). Spirit Airlines bankruptcy costs. CNBC. Retrieved July 22, 2026 from [https://www.cnbc.com/2026/05/05/spirit-airlines-bankruptcy-costs.html](https://www.cnbc.com/2026/05/05/spirit-airlines-bankruptcy-costs.html) (CNBC)
Duffy, K. (2026, July 1). Meta stock rises on cloud AI compute announcement. CNBC. Retrieved July 22, 2026 from [https://www.cnbc.com/2026/07/01/meta-stock-cloud-ai-compute.html](https://www.cnbc.com/2026/07/01/meta-stock-cloud-ai-compute.html) (CNBC)
International Air Transport Association. (2025, December 9). Airline Profitability Stabilizes with 3.9% Net Margin Expected in 2026. IATA Pressroom. Retrieved July 22, 2026 from [https://www.iata.org/en/pressroom/2025-releases/2025-12-09-01/](https://www.iata.org/en/pressroom/2025-releases/2025-12-09-01/) (IATA)
International Air Transport Association. (2026, June 7). Middle East Disruptions and High Fuel Prices Halve Airline Industry Profitability. IATA Pressroom. Retrieved July 22, 2026 from [https://www.iata.org/en/pressroom/2026-releases/06-07-middle-east-disruptions-high-fuel-prices-halve-airline-industry-profitability/](https://www.iata.org/en/pressroom/2026-releases/06-07-middle-east-disruptions-high-fuel-prices-halve-airline-industry-profitability/) (IATA)
Menlo Ventures. (2025, December 9). 2025: The State of Generative AI in the Enterprise. Menlo Ventures. Retrieved July 22, 2026 from [https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/](https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/) (Menlo Ventures)
Oks, D. (2026, May 4). Why airlines are always going bankrupt. David Oks Blog. Retrieved July 22, 2026 from [https://davidoks.blog/p/why-airlines-are-always-going-bankrupt](https://davidoks.blog/p/why-airlines-are-always-going-bankrupt) (David Oks)
Sacra. (2026, June 15). OpenAI revenue, valuation & funding. Sacra. Retrieved July 22, 2026 from [https://sacra.com/c/openai/](https://sacra.com/c/openai/) (Sacra)
Sitebolts. (2024, September 20). The Red Queen's Race: Lessons from Biology for IT. Sitebolts. Retrieved July 22, 2026 from [https://sitebolts.com/the-red-queens-race-lessons-from-biology-for-it/](https://sitebolts.com/the-red-queens-race-lessons-from-biology-for-it/) (Sitebolts)
Sitebolts. (2025, January). DeepSeek AI: A Deep Dive. Sitebolts. Retrieved July 22, 2026 from [https://sitebolts.com/deepseek-deep-dive/](https://sitebolts.com/deepseek-deep-dive/) (Sitebolts)
Sitebolts. (2025). The Visible Hand: When Digital Supply Chains Meet Geopolitical Reality. Sitebolts. Retrieved July 22, 2026 from [https://sitebolts.com/the-visible-hand-when-digital-supply-chains-meet-geopolitical-reality/](https://sitebolts.com/the-visible-hand-when-digital-supply-chains-meet-geopolitical-reality/) (Sitebolts)
Sitebolts. (2025). When Digital Supply Chains Meet Geopolitical Reality Part II: Electric Boogaloo. Sitebolts. Retrieved July 22, 2026 from [https://sitebolts.com/when-digital-supply-chains-meet-geopolitical-reality-part-ii-electric-boogaloo/](https://sitebolts.com/when-digital-supply-chains-meet-geopolitical-reality-part-ii-electric-boogaloo/) (Sitebolts)
Sitebolts. (2025). When Digital Supply Chains Meet Geopolitical Reality Part III: The Trilogy. Retrieved July 22, 2026 from [https://sitebolts.com/when-digital-supply-chains-meet-geopolitical-reality-part-iii-the-trilogy/](https://sitebolts.com/when-digital-supply-chains-meet-geopolitical-reality-part-iii-the-trilogy/) (Sitebolts)
Sitebolts. (2025, December 25). Sitebolts Predictions 2025-2026. Sitebolts. Retrieved July 22, 2026 from [https://sitebolts.com/predictions25/](https://sitebolts.com/predictions25/) (Sitebolts)
Sircar, A. (2026, May 21). Airbnb CEO Brian Chesky Called Chinese AI Fast And Cheap. Now, Congress Wants Answers. Forbes. Retrieved July 22, 2026 from [https://www.forbes.com/sites/anishasircar/2026/05/21/airbnb-ceo-brian-chesky-called-chinese-ai-fast-and-cheap-now-congress-wants-answers/](https://www.forbes.com/sites/anishasircar/2026/05/21/airbnb-ceo-brian-chesky-called-chinese-ai-fast-and-cheap-now-congress-wants-answers/) (Forbes)
Snyk. (2026, June 14). When a Government Pulls an AI Model: Security Takeaways from the Fable and Mythos Suspension. Snyk Blog. Retrieved July 22, 2026 from [https://snyk.io/blog/fable-mythos-suspension-security-takeaways/](https://snyk.io/blog/fable-mythos-suspension-security-takeaways/) (Snyk)
SpaceX, Inc. (2026, May 20). S-1 Registration Statement. U.S. Securities and Exchange Commission. Retrieved July 22, 2026 from [https://finance.yahoo.com/markets/article/spacex-files-ipo-prospectus-offering-a-peek-into-its-finances-205406189.html](https://finance.yahoo.com/markets/article/spacex-files-ipo-prospectus-offering-a-peek-into-its-finances-205406189.html) (Yahoo Finance / CNBC)
Spirit Airlines. (2026, May 2). Spirit Airlines Begins Orderly Wind-Down of Operations. Spirit Restructuring. Retrieved July 22, 2026 from [https://www.spiritrestructuring.com/resources/Spirit-Airlines-Begins-Orderly-Wind-Down-of-Operations.pdf](https://www.spiritrestructuring.com/resources/Spirit-Airlines-Begins-Orderly-Wind-Down-of-Operations.pdf) (Spirit Airlines)
Stempel, J. & Kaye, B. (2025, November 6). OpenAI backtracks on government support for chip investments. CNN. Retrieved July 22, 2026 from [https://www.cnn.com/2025/11/06/tech/openai-backtracks-government-support-chip-investments](https://www.cnn.com/2025/11/06/tech/openai-backtracks-government-support-chip-investments) (CNN)
TradingKey. (2026, July). SpaceX IPO Makes History, But Breaks Below Offer Price a Month Later. TradingKey. Retrieved July 22, 2026 from [https://www.tradingkey.com/analysis/stocks/us-stocks/262038991-nvda-spacex-musk-spcx-sndk-tradingkey](https://www.tradingkey.com/analysis/stocks/us-stocks/262038991-nvda-spacex-musk-spcx-sndk-tradingkey) (TradingKey)
Value Add VC. (2026). OpenAI Revenue 2026: $25B ARR and a -122% Operating Margin. Value Add VC. Retrieved July 22, 2026 from [https://valueaddvc.com/blog/openai-revenue-2026-25b-arr-2b-month-and-the-path-to-profitability](https://valueaddvc.com/blog/openai-revenue-2026-25b-arr-2b-month-and-the-path-to-profitability) (Value Add VC)
Wile, R. (2026, May 21). 'Financials look reckless': Lifting the xAI hood in the SpaceX IPO. Yahoo Finance / Morningstar. Retrieved July 22, 2026 from [https://finance.yahoo.com/markets/stocks/articles/financials-look-reckless-lifting-xai-000608351.html](https://finance.yahoo.com/markets/stocks/articles/financials-look-reckless-lifting-xai-000608351.html) (Yahoo Finance / Morningstar)
World Airline Strategy. (2026, May 3). Why Spirit Airlines Failed: A Strategic Autopsy. World Airline Strategy (Substack). Retrieved July 22, 2026 from [https://worldairlinestrategy.substack.com/p/spirit-airlines-failure-strategic-autopsy](https://worldairlinestrategy.substack.com/p/spirit-airlines-failure-strategic-autopsy) (World Airline Strategy)
Wu, S. (2026, May 28). Anthropic raises $65 billion, nears $1T valuation ahead of IPO. TechCrunch. Retrieved July 22, 2026 from [https://techcrunch.com/2026/05/28/anthropic-raises-65-billion-nears-1t-valuation-ahead-of-ipo/](https://techcrunch.com/2026/05/28/anthropic-raises-65-billion-nears-1t-valuation-ahead-of-ipo/) (TechCrunch)
Yuen, E. (2026, May 21). SpaceX SPCX IPO S-1 Full Teardown: $1.75 Trillion Valuation, Starlink, xAI, and the Anthropic Deal. The VC Corner. Retrieved July 22, 2026 from [https://www.thevccorner.com/p/spacex-spcx-ipo-s1-teardown-valuation-2026](https://www.thevccorner.com/p/spacex-spcx-ipo-s1-teardown-valuation-2026) (The VC Corner)
Zhang, C. (2026, May 20). Airbnb's Chesky Says US 'Misunderstanding' Use of Chinese Open-Source AI Models. Bloomberg. Retrieved July 22, 2026 from [https://www.bloomberg.com/news/articles/2026-05-20/airbnb-s-chesky-says-us-misunderstanding-use-of-chinese-open-source-ai-models](https://www.bloomberg.com/news/articles/2026-05-20/airbnb-s-chesky-says-us-misunderstanding-use-of-chinese-open-source-ai-models) (Bloomberg)