OpenAI and Anthropic are moving toward what could become some of the largest IPOs in history, and that alone forces a different level of scrutiny. An IPO is not about the moment of takeoff. It is about trajectory. A rocket can launch with momentum, but what matters is whether it stays on course once gravity, cost, and friction begin to accumulate.
Underneath the current AI boom, there are three issues that are now becoming central to that trajectory. The first is “vibe slop.” The second is overreliance on government as a customer. The third is the cost of scaling through chips and energy. These are not separate themes. They are different expressions of the same underlying economic pressure.
The concern around “vibe slop” is coming from inside the system itself, not from outside critics. Engineers involved in building modern AI coding tools describe a pattern where software is increasingly generated through prompts rather than carefully designed and tested through traditional engineering discipline.
Software can be created quickly, sometimes in minutes now with AI but they are not always built to last. They accumulate hidden bugs, inconsistent behavior, and technical debt that only shows up later when systems scale or break under real-world load. The issue is not whether AI can write code but whether AI-generated code produces systems that remain stable over time.
READ: Sreedhar Potarazu | The Norton Principles: Why the future of AI in healthcare requires wisdom of the past (May 19, 2026)
This is already visible inside large organizations. Even where AI is generating a significant portion of new code, human engineers are still required to review, debug, and stabilize production systems. The gap between creation speed and system durability is becoming an operational problem, not just a theoretical one.
This matters for companies like OpenAI and Anthropic because long-term value will not be determined by how much they can generate, but by how much of that generation survives at scale without creating rising maintenance and failure costs.
The second issue is cost, which is increasingly tied to physical constraints rather than software abstractions. AI is not just code. It is computation at industrial scale. That means chips, data centers, and electricity are not supporting inputs. They are the core cost structure.
A large portion of that structure depends on advanced semiconductors, especially from NVIDIA. That creates a concentrated dependency at the most critical layer of the system. It limits pricing flexibility and makes long-term margin expansion harder to assume.
Data center expansion adds another layer. Training and running frontier models requires massive infrastructure buildouts that resemble utilities more than traditional software businesses. That changes the economics from scalable software margins to capital-intensive infrastructure scaling.
The third issue is demand structure, and this is where government involvement matters in a very specific way.
A growing share of the AI ecosystem is indirectly tied to the government as a customer, not just as a regulator or policymaker. Governments are becoming early and sometimes large-scale buyers of AI systems for defense, security, public services, and administrative automation. In some cases, they are also funding or anchoring infrastructure expansion tied to AI capacity.
This creates a structural dependency that is not usually present in traditional software companies. Instead of purely private enterprise demand driving growth, part of the demand base is anchored in public sector procurement cycles and policy-driven spending priorities. That can be stable in some periods, but it also introduces concentration risk that depends on political cycles, budget constraints, and shifting policy priorities.
The important point is not whether government demand is good or bad. The point is that it becomes a meaningful customer category at scale. And in an IPO context, customer mix matters as much as customer size.
The revenue picture today still looks strong in absolute terms. OpenAI is widely estimated to be operating at an annualized revenue run rate in the range of roughly $3 to $5 billion. Anthropic is generally estimated at roughly $1 to $2 billion. Both are growing quickly, and both are expanding enterprise and API usage.
READ: Sreedhar Potarazu | The Gods of Silicon Valley go to court: Altman, Musk, and the dangerous psychology behind AI and wealth (May 16, 2026)
But revenue growth alone does not answer the sustainability question.
A meaningful portion of that revenue is concentrated in enterprise customers and high-usage developers. These customers are valuable, but they are also flexible. They multi-source providers, test competing models, and build internal alternatives when pricing or performance shifts. That creates a base of revenue that is real, but not fully stable.
At the same time, the cost side of the equation is moving in the opposite direction. Every increase in usage increases compute demand. Every increase in compute demand increases dependence on chips, energy, and large-scale infrastructure. Costs scale with growth rather than decoupling from it.
So both sides of the model are under pressure at the same time. Revenue is growing but concentrated. Costs are rising and tied to physical constraints. And the gap between the two is still being supported by capital markets rather than durable operating margins.
That is the gap public markets will eventually close.
Historically, major technology transitions follow a familiar pattern. Early growth is driven by capability and adoption. Later stages are defined by margin pressure, consolidation, and a reassessment of what actually produces durable value. The internet, cloud, and social media all went through versions of this cycle.
AI is different in scale and speed, but not necessarily different in structure.
The deeper issue is that AI makes creation easier, but does not automatically make value more durable. In some cases, it may do the opposite by increasing the volume of output faster than the system’s ability to filter what actually matters.
That is what “vibe slop” ultimately points to. Not low-quality output in isolation, but a system where creation is abundant and durability is uncertain.
And that is exactly the kind of imbalance that IPO markets are designed to correct.
Because once these companies are public, the question will no longer be what they can generate.
It will be what actually holds value after it is generated.