OpenAI and Anthropic pursue trillion dollar IPOs

OpenAI and Anthropic are moving closer to initial public offerings that could become defining tests for the artificial intelligence boom, with private-market valuations, revenue projections and computing costs now drawing heavier scrutiny from public-market traders.

The possible listings are being discussed at a moment when AI companies are reporting extraordinary growth, but also facing rising questions about how much of that growth can turn into durable profit. Reports cited by market participants suggest Anthropic could generate more than $100 billion in annual revenue this year and seek a market value as high as $3 trillion. OpenAI is also said to be preparing for a later public filing, with sales estimates for this year reaching about $70 billion.

If those figures hold, the two companies could enter public markets at valuations rarely seen outside the largest technology groups. SpaceX, which previously listed with a valuation of about $1.75 trillion, has been cited by market watchers as a precedent for large, highly anticipated technology offerings. According to comments attributed to Brad Gerstner, SpaceX raised $75 billion in its IPO and has since traded roughly 25% above its listing price, strengthening the argument that demand for category-leading private technology companies remains strong.

But the road to public markets may also expose a core tension in the AI trade: revenue is rising quickly, while the cost of serving and improving large models remains high. Once companies file public documents, traders will receive a clearer view of how much these firms spend on computing infrastructure, model training, data, inference and token usage. That level of disclosure could reshape debate over whether today’s AI leaders deserve trillion-dollar valuations.

AI listings move closer

The expected IPO push by OpenAI and Anthropic would mark a major shift for the AI sector. Until now, much of the value created by generative AI has been held in private companies backed by venture funds, strategic technology partners and early shareholders. Public listings would allow a broader group of traders to buy into the sector directly, while also forcing the companies to publish financial statements, risk factors and operating metrics.

That move could change the tone of the market. In private funding rounds, fast-growing AI companies have been valued largely on revenue growth, market share, strategic importance and expectations for future dominance. Public markets usually apply a broader test. They tend to focus not only on sales growth, but also on gross margins, cash flow, capital spending, customer concentration and the path to sustained profitability.

Chamath Palihapitiya has warned that the window for these companies to go public may remain open only before the wider market fully understands the gap between AI spending and measurable productivity gains. He said operational spending on model tokens is doubling roughly every 45 days, while productivity gains in downstream businesses have grown by less than 5%.

That argument strikes at one of the central questions facing the AI economy. Companies are spending heavily to integrate AI into customer service, software development, sales, marketing, legal work and data analysis. Yet the financial return from those integrations remains uneven. Some firms report meaningful efficiency gains, while others are still experimenting or discovering that high model usage can raise costs faster than expected.

Revenue projections meet cost concerns

Palihapitiya’s review of the S&P 493, which excludes the seven largest U.S. technology companies, found earnings-per-share growth of about 9%. He said much of that increase came from inflation-linked pricing power and share buybacks, leaving AI-driven returns between 0% and 2%.

That finding, if reflected in broader market data, would challenge the idea that generative AI has already delivered a large earnings boost across corporate America. While AI tools may be improving workflows inside certain teams, the gains have not yet shown up clearly in aggregate profit growth for a wide range of listed companies.

For OpenAI and Anthropic, the issue is not only whether demand is strong. Demand appears substantial. The issue is whether high demand can be served at margins that justify valuations of $1 trillion, $2 trillion or even $3 trillion. Large models require expensive graphics processors, data centers, energy, networking equipment and technical staff. Inference costs — the cost of running models when users send prompts and receive answers — can become significant as usage scales.

Gerstner has taken a more optimistic view, suggesting that Anthropic’s public debut could attract strong demand similar to SpaceX. He has also said OpenAI’s sales may reach $70 billion this year and that GPT-6 could arrive within one month. In his view, revenue scale and demand growth could allow both OpenAI and Anthropic to surpass the trillion-dollar mark in public markets.

The difference between the optimistic and cautious views is not about whether AI is important. Both sides accept that AI is likely to remain one of the most important technology trends of the decade. The disagreement is about timing, margins and how quickly customers can turn AI spending into measurable output.

Compute bills shape adoption

Rising token and compute costs are already affecting how companies use large language models. Chief technology officers at many firms have reported escalating bills as employees and customers rely more heavily on AI tools. Some companies are building internal systems to route queries to cheaper models when tasks do not require the most advanced systems.

This routing strategy can reduce costs. A simple classification, summarization or formatting task may not need a frontier model. A complex legal, coding, financial or scientific task may require a higher-end system with stronger reasoning and accuracy. The challenge is building infrastructure that can identify the right model for each task without harming reliability.

David Sacks has said many corporate users want to reduce dependence on proprietary AI systems but often lack the technical ability to migrate. He noted that enterprise spending on open-source tools fell from 19% last year to 11% this year, while use of closed models expanded. That shift suggests that ease of use, performance and commercial support may still matter more than lower software costs for many business users.

Sacks also said request routing between model types can reduce compute consumption, though only firms with advanced infrastructure have managed it effectively. He cited companies such as Coinbase and DoorDash as examples of businesses with the technical capacity to manage these systems across large-scale operations.

The broader market is still trying to determine whether open-source AI will put lasting pressure on proprietary model makers. Meta intensified pricing competition after releasing a low-cost model priced at roughly one-hundredth of some comparable systems. That kind of price gap creates pressure on closed-model providers to prove that their systems are materially better for high-value work.

For customers, the decision is practical. Cheaper models may be sufficient for routine tasks. More expensive models may be necessary when errors are costly, such as in engineering, medicine, law, finance or high-stakes customer operations. Palihapitiya compared the cycle to smartphone adoption, where performance eventually reached a “good enough” level for many users. If AI follows a similar path, premium model providers may need to show continuous improvement to defend pricing power.

China and U.S. policy risks

The AI boom is also becoming more closely tied to national security policy. China is reportedly considering restrictions on foreign access to its most advanced AI models and exploring penalties for leaks of research abroad. The measures would reflect a growing belief that frontier AI systems are strategic assets, not merely commercial products.

Sacks said China’s approach resembles the shift made by some U.S. developers, which initially promoted open access but moved toward closed systems as competition intensified. He said American policymakers broadly agree that maintaining technological superiority over China in AI is a central policy goal.

Washington is also examining whether China’s GLM-5.2 system includes distilled elements from U.S. frontier models. Model distillation generally refers to using outputs from a larger or more advanced model to train a smaller or competing model. If policymakers conclude that foreign systems are being trained on restricted U.S. model outputs or advanced datasets, new limits could follow.

Potential legislation may focus on limiting unauthorized model distillation, controlling exports of advanced datasets and restricting access to certain frontier capabilities. Sacks said such measures would not necessarily hurt U.S. firms capable of developing their own open-source models and infrastructure.

Policy risk now sits alongside financial risk for AI companies preparing to list. Public traders will have to weigh not only revenue growth and margin pressure, but also export controls, data rules, intellectual property disputes, energy availability and international competition.

Trump Accounts add a new market force

At the same time, a new savings initiative known as Trump Accounts has launched nationally, adding another major development for U.S. markets. The program provides each American newborn with $1,000 placed into an S&P 500 index fund. Within 24 hours of launch, 1.5 million accounts were opened, attracting more than $1 billion in deposits, according to figures cited by supporters of the program.

Families can contribute up to $5,000 per year, with tax-free compounding for 18 years. After that period, partial withdrawals are allowed for education, housing or business creation. The structure is designed to give children early exposure to long-term market growth and to expand household participation in broad equity funds.

Financial planners cited by program supporters estimate that fully funded accounts could reach $1 million by age 28 if market returns resemble those of the past 30 years. Sacks has calculated that a balance held by an 18-year-old could grow beyond $10 million by retirement under long-term compounding assumptions.

Those projections depend heavily on future market returns, contribution levels and withdrawal behavior. Past equity performance does not guarantee similar results, and the S&P 500 can experience long periods of volatility. Still, the program could create a meaningful new source of demand for broad-market index funds if participation remains high.

Gerstner has said the initiative may eventually create more than 100 million personal market accounts and channel $2 trillion to $4 trillion into household savings over the next 15 years. Developers including Joe Gebbia and corporate contributors from the technology sector helped design the accompanying mobile platform, making the rollout one of the largest technology-led collaborations with government to date.

Philanthropic donors, corporate leaders and charitable foundations have also pledged billions of dollars to expand the program for lower-income families. If those pledges are fulfilled, the accounts could become a major savings vehicle for households that historically had limited access to long-term market compounding.

Market implications widen

The combination of AI mega-IPOs and a government-backed stock savings program could reshape capital flows. Large AI listings would offer new public exposure to companies that have so far been mostly private. Trump Accounts could increase long-term demand for broad index funds that hold the largest listed U.S. companies.

That dynamic may benefit established public technology groups if index contributions rise steadily over time. Broad-market funds typically allocate capital according to market capitalization, meaning the largest companies receive the largest share of passive inflows. If AI companies eventually join major indexes, they could also become beneficiaries of those flows, provided they meet eligibility rules.

Some market commentators have argued that this shift could reduce speculative demand in riskier assets, including digital tokens. Their view is that federal policy encouraging broad stock ownership may pull capital toward index funds and away from more volatile markets. Others caution that the relationship is not automatic. Digital asset markets are driven by liquidity, regulation, adoption, monetary policy, network activity and trader sentiment, not only by stock-market programs.

What is clear is that public disclosure will become more important. AI companies that go public will need to show how revenue translates into margins. Firms adopting AI will need to prove that high compute spending leads to productivity gains. Policymakers will need to balance innovation, national security and market access. Traders will need to assess whether current valuations reflect lasting earnings power or expectations that remain ahead of reality.

For now, OpenAI and Anthropic stand at the center of that debate. Their expected public offerings could validate the AI boom, challenge it, or do both at the same time. The companies may arrive in public markets with extraordinary revenue growth and unmatched strategic importance. They may also arrive with cost structures that demand close inspection.

The next phase of the AI market will be less about excitement alone and more about evidence: how much businesses are willing to pay, how fast model costs decline, how productivity improves and how regulators define the rules of global competition. Public markets are preparing to put those questions in full view.