OpenAI’s $852 B valuation sparks investor doubts as GPT‑6 rollout looms

OpenAI’s latest valuation surge to $852 billion has been accompanied by a flurry of concrete technical disclosures about GPT‑6, most of which come from the company’s own public statements and tracking firms. According to The Information, the pre‑training phase for the next‑generation model wrapped on March 24, 2026, marking the first major milestone since the GPT‑5 rollout. The training run was conducted at OpenAI’s “Stargate” data center in Abilene, Texas, leveraging a cluster of more than 100,000 NVIDIA H100 GPUs—a scale that dwarfs the infrastructure used for GPT‑5 and suggests a dramatic increase in compute budget. Sam Altman, OpenAI’s CEO, told investors in a March 24 briefing that the model is “a few weeks” from a public preview, while co‑founder Greg Brockman emphasized that GPT‑6 will not be “an incremental improvement,” hinting at architectural changes that could reshape the model’s capabilities and cost profile.

The technical specifics that have been confirmed also illuminate why the valuation has raised eyebrows among venture capitalists. The same tracking reports note that OpenAI has deliberately shut down its Sora video‑generation project to reallocate GPU capacity to GPT‑6, a move that underscores the resource intensity of the new model. Analysts familiar with the internal budgeting, as cited by the tokenmix.ai newsletter, warn that the massive GPU deployment will likely inflate OpenAI’s API pricing, especially for customers who rely on high‑volume token consumption. The newsletter further points out that the model’s context window is rumored to reach 2 million tokens—far beyond the 32 k token limit of GPT‑4—but this claim remains unverified, and investors are wary of the potential impact on downstream costs and latency.

Beyond the hardware, the performance expectations are equally contentious. While an unnamed insider leak has been circulating the claim that GPT‑6 will be “40 % better than GPT‑5.4,” tokenmix.ai categorizes this as unconfirmed. The lack of independent benchmarks means that investors cannot yet assess whether the purported gains stem from larger model size, more efficient training algorithms, or novel data curation strategies. The absence of transparent metrics fuels skepticism, especially given the valuation jump that places OpenAI’s market cap above the combined worth of many legacy tech giants. Investors flagged in reports by Meyka and Storyboard18 that the strategic shift toward a single, ultra‑large model could expose the company to heightened execution risk, should the model fail to deliver the expected ROI on its $852 billion price tag.

Finally, the broader market context adds another layer of doubt. Competing firms such as Anthropic and Google have publicly committed to more modular, multi‑model ecosystems that spread risk across smaller, specialized systems. OpenAI’s decision to double down on a monolithic GPT‑6 architecture, as inferred from the reallocation of GPU resources and the lack of incremental product releases, diverges from this trend. The tokenmix.ai analysis warns that the “few weeks” timeline may compress testing cycles, potentially leading to stability or safety issues that could erode enterprise confidence. As the rollout approaches, investors will be watching closely for concrete performance data, pricing structures, and any signs that OpenAI can justify its $852 billion valuation beyond hype.