“They are not asking you to value what they earn today. They are asking you to price what they believe humanity becomes tomorrow. That is a very different proposition — and right now, Wall Street is nodding along.”
Neal Lloyd · Ground Truth, Episode 01On June 1st 2026, Anthropic — the AI safety company founded by people who left OpenAI because they thought it was moving too fast — filed confidentially for a US IPO at a valuation of $965 billion. The company had just raised $65 billion in a single Series H round led by Altimeter, Sequoia, and Dragoneer, making it briefly the most valuable private AI company in Silicon Valley. It had overtaken OpenAI. The company that exists because somebody pumped the brakes is worth nearly a trillion dollars. The brakes, it turns out, are the hottest asset class on the planet.
Two days later, OpenAI filed its own confidential S-1. SpaceX — which absorbed Elon Musk’s xAI earlier this year — priced its IPO at $135 a share, giving it a record $1.77 trillion valuation. Three companies. Combined paper value: $3.6 trillion. The GDP of France is $3.1 trillion. These three organisations, between them, are asking public markets to absorb more capital at debut than the annual economic output of the world’s seventh-largest economy. Welcome to Ground Truth. We begin where the money is.
What the Scoreboard Actually Looks Like Right Now
Let us put the numbers in one place, because the financial press has been reporting them in fragments and the full picture deserves to be seen whole.
Anthropic: $965 billion valuation at IPO filing. Revenue run rate of $47 billion, representing roughly 1,400% year-over-year growth — a figure that, if you work in finance, you will have stared at for some time before accepting it as real. Over 300,000 business and enterprise customers. Backed by Amazon ($16.8 billion in pre-tax investment gains recorded in a single quarter), Google, Sequoia, and now Altimeter. The Claude model family powering everything from consumer chat to enterprise coding to Amazon’s Bedrock cloud AI platform.
OpenAI: $852 billion valuation at private market. $25 billion in annualised revenue. 900 million weekly active ChatGPT users. Projecting losses of $14 billion in 2026. Not expecting profitability until 2029–2030. CFO Sarah Friar has flagged late 2026 or 2027 as the most likely listing window. Targeting up to $1 trillion at IPO, which would make it the most valuable company ever to debut on public markets by a significant margin.
SpaceX: $1.77 trillion valuation. $75 billion raise. Absorbed Musk’s xAI in February at a combined $1.25 trillion. $18.7 billion in 2025 consolidated revenue. Starlink now reaches over 10 million subscribers across 160 countries. Already the largest IPO in history by valuation. Nasdaq’s Fast Entry rule means it could join the Nasdaq-100 within 15 trading days of listing.
OpenAI is projecting $14 billion in losses in 2026. It will not be profitable until 2029 at the earliest. It is targeting a $1 trillion IPO valuation. That implies a price-to-sales ratio somewhere north of 40 at current revenue — in an environment where Nvidia, the most dominant chip company in history at the centre of the AI infrastructure boom, trades at 24 times sales. Wall Street is being asked to value a loss-making AI lab more richly than the company supplying the picks and shovels to every gold miner in the field. The logic requires a very specific set of assumptions about market size, competitive moats, and margin expansion. Every single one of those assumptions involves uncertainty wide enough to park a data centre in.
Everyone Is Thinking It. Let’s Actually Say It.
The comparison to the dot-com bubble is both inevitable and genuinely complicated, which is why the people making it tend to immediately qualify it in three directions at once. Here is the honest version.
What rhymes with 1999: Narratives running ahead of fundamentals. Companies projecting profitability a comfortable four years out. Capital flooding into an entire sector rather than specific companies with proven unit economics. The phrase “this time is different” being uttered with increasing conviction by the people whose compensation depends on it being different. Of the five largest IPOs in modern history, only Visa significantly outperformed markets in the years following its listing. Saudi Aramco debuted at nearly $1.7 trillion in 2019 and still trades below its issue price. Facebook’s stock fell 38% in its first six months. Large IPOs with narrative-driven valuations have a historical record that deserves attention.
What is genuinely different: The underlying technology is real and working at scale. OpenAI’s 900 million weekly users are using a product that demonstrably does something. Anthropic’s 1,400% revenue growth is actual revenue from actual enterprise customers paying actual money. The infrastructure buildout — NVIDIA GPUs, data centres, power agreements — is physical, tangible, and generating real cash flows for the companies supplying it. This is not bandwidth speculation. This is not eyeballs-as-revenue. The technology has already transformed knowledge work across every major industry. The question is not whether it is valuable. The question is whether it is this valuable, right now, at these prices, for investors buying in after the private market has already run valuations from near-zero to near-trillion.
Of the three companies heading to market, none are profitable by conventional accounting. All three are asking public investors to price a vision of the future rather than the reality of today. History has a word for that. Several words, actually. Most of them end badly.Neal Lloyd · Ground Truth, Episode 01
Follow the Money. Then Follow Who the Money Answers To.
The financial story and the power story are inseparable and the press tends to tell them separately. Here is why that matters.
Amazon holds a position in Anthropic significant enough that it recorded $16.8 billion in pre-tax investment gains from that stake in a single quarter. It is also Anthropic’s primary cloud provider through AWS Bedrock. It is simultaneously the investor, the infrastructure, and the distribution channel. Google holds a similar position — investor, cloud competitor, and distribution partner simultaneously. Microsoft has built 11,000 AI models inside Azure, funded OpenAI to the tune of tens of billions, and now also runs Claude on the same platform. NVIDIA supplies the chips to everyone and has invested in several of the labs it supplies.
This vertical integration is not coincidence and it is not conspiracy. It is rational capital allocation by technology companies that correctly identified AI infrastructure as the most consequential platform shift since mobile. But it has a structural consequence: the companies that own the cloud, the chips, and the capital are positioned to capture value from every outcome in the AI race. Whether OpenAI wins or Anthropic wins or a third player emerges, the infrastructure companies win. The question for public market investors is not whether AI creates enormous value. It already has. The question is where in the stack that value accrues — and whether the companies going public are the ones who capture it or the ones who generate it for others.
Both OpenAI and Anthropic were explicitly founded with safety-first missions. OpenAI as a non-profit ensuring AGI benefits humanity. Anthropic because its founders believed safety was being deprioritised at OpenAI. Both are now preparing to be accountable to public shareholders whose fiduciary interest is return on capital. This is not an impossible combination — public companies can hold genuine non-financial commitments. But the structural pressure of quarterly earnings, competitive positioning, and deployment speed has reshaped organisations with far more robust institutional defences than two-decade-old AI labs. Whether the safety mission survives contact with the earnings call is one of the most important governance questions in technology. The IPO prospectuses will be required to address it. Read them carefully when they arrive.
The Domino Nobody Wants to Name
Perplexity AI CEO Aravind Srinivas told CNBC that his company is sticking to its 2028 IPO plan — but there is, in his words, no sugar-coating the ripple effects if the 2026 listings fail to perform well. That is a careful statement from a CEO who understands that the entire next generation of AI IPOs is downstream of how this wave lands.
A weak debut from any of the three would force a repricing conversation across the entire AI sector. Private valuations for every AI company downstream — the thousands of startups that have raised at elevated multiples on the assumption that the public market will eventually absorb them at similar or higher prices — would face mark-to-market pressure. Venture portfolios would reprice. The secondary markets where employees and early investors have been selling AI company shares would seize up. The capital flowing into AI infrastructure at current rates depends, in part, on the assumption that the public markets will validate the private valuations. If they do not, the feedback loop runs in reverse.
This is not a prediction of collapse. It is a description of risk. The base case for all three IPOs, among institutional investors currently, is a successful debut followed by a period of consolidation as the companies grow into their valuations. Polymarket prices the SpaceX June listing at 65.5% probability of success. The optimistic case is that these are genuinely extraordinary businesses — and they are — and that public markets correctly price them at extraordinary valuations. History, as noted, has views on this. But history also did not anticipate a company growing revenue at 1,400% year over year. Both things can be true simultaneously.
The S-1 Documents Will Tell You More Than Any Press Release
Read the prospectuses. When Anthropic and OpenAI publish their public S-1 filings, they will be legally required to disclose things that private companies never have to: revenue concentration, customer dependency, regulatory exposure, the actual cost structure of safety research, the governance arrangements that determine who controls the company, and the specific risk factors that their lawyers have identified as material. These documents will be among the most consequential in the history of technology. They will contain more honest information about the state of the AI industry than everything published in the press over the preceding five years combined. They will be long, dense, and extraordinarily revealing. Read them.
Watch the lock-up expiry. Insider lock-up periods — the windows during which early employees and investors cannot sell their shares — typically expire 90 to 180 days after an IPO. The behaviour of insiders at lock-up expiry is one of the most reliable signals of how people actually close to the company assess its prospects. When the people who know the most sell aggressively at the earliest opportunity, that is information. When they hold, that is also information.
Track the revenue quality.** Not just the growth rate — the concentration. If 30% of Anthropic’s revenue comes from Amazon, and Amazon is both its biggest customer and its biggest infrastructure provider and a major shareholder, the quality of that revenue is structurally different from diversified enterprise revenue. Concentration risk of this kind needs to be understood in context.
And follow Ground Truth. This series exists precisely for this moment — to track what is actually happening as it happens, anchor it in context, and give you the analysis without the hype or the panic. Episode 02 is already in the works.
The companies that were founded because people were worried about AI moving too fast are now the most expensive assets in the history of technology. At some point the irony stops being ironic and starts being the actual story. We are there now.Neal Lloyd · Ground Truth, Episode 01
Ground Truth, Episode 01 · June 2026
Neal Lloyd covers the real-world impact of AI — money, power, geopolitics, and the stories behind the headlines. Ground Truth is his live news and analysis companion to the Inside The Machine theory series.
