Sam Altman once spoke with confidence. In early 2025, the OpenAI chief executive posted on his blog that his team now knew how to build artificial general intelligence as traditionally understood. Time reported on those words. They carried weight across Silicon Valley and beyond.
Yet the months that followed told a different story. Progress on reasoning models like o1 and o3 delivered gains in math and coding. They did not translate easily into autonomous agents that could handle complex, multi-day tasks without constant human guidance. By mid-2025, forecasts stretched out. Optimism cooled.
That shift matters.
OpenAI had set specific internal markers. It aimed to launch AI systems capable of research intern duties before September 2026. Fully automated AI researchers would follow by March 2028. Jakub Pachocki, the company’s chief scientist, laid out the plan in public remarks. “We are trying to build a system that can independently complete research projects,” he said, according to 36Kr. He added that deep-learning systems may reach super-intelligent levels in less than a decade.
But recent assessments suggest the company sits behind even these adjusted targets. A Yahoo Finance article drawn from 24/7 Wall St. analysis shows OpenAI on pace to miss its own advertising revenue forecasts by roughly 90 percent. The piece notes the firm’s $100 billion ad revenue goal for 2030 dwarfs eMarketer’s projection of $5.41 billion for the entire chatbot advertising market that year. Yahoo Finance highlighted how this gap raises questions about the economics powering OpenAI’s compute-heavy ambitions.
Ads were supposed to become a major pillar. The company began testing them in February 2026. Projections called for $2.5 billion in ad revenue that year. Reality points to less than $1 billion across all standalone AI chatbots. ChatGPT’s share of traffic has slipped from 87 percent to 65 percent as Google’s Gemini picks up users. Those numbers surfaced in April 2026 and triggered a sharp selloff. Oracle shares dropped more than 4 percent in a single session. NVIDIA, Broadcom, AMD, Arm and SoftBank felt the ripple. Oracle remains down 37 percent year to date as of late July 2026.
The revenue pressure intersects with the technical hurdles. Daniel Kokotajlo, a former OpenAI researcher who now writes for 80,000 Hours, captured the arc in a March 2026 essay. Early excitement around o1 and o3 gave way to recognition that inference scaling hits practical limits. Ten times more thinking time often proves too expensive. Reinforcement learning shows efficiency far below pre-training. Continual learning remains elusive. 80,000 Hours detailed how Metaculus community forecasts for transformative AI stretched to November 2033 at one point.
Then came February 2026. New releases from both OpenAI and Anthropic changed the mood again. GPT-5.3-Codex and Claude Opus 4.6 demonstrated sharper coding and agentic abilities. Claude Sonnet 4.6 joined the wave. Models began handling longer, more independent workflows. Some insiders spoke of early recursive self-improvement signals. Adoption inside companies ticked higher. Yet Kokotajlo cautioned against another hype cycle. Incremental gains compound. They do not guarantee the sudden leap many once anticipated.
Altman himself has adjusted his language. He told Bloomberg that AGI will probably arrive during Donald Trump’s current term but called the term “sloppy.” In a December 2024 interview referenced in the Time article, he suggested AGI could get built sooner than most expect yet the world would continue in mostly the same way. Superintelligence, he has argued, represents the bigger focus. It could accelerate scientific discovery at rates hard to imagine.
Competitors echo parts of this view while differing on pace. Dario Amodei at Anthropic has pointed to AGI by 2027. Elon Musk has made similar claims for xAI systems. A large survey of AI researchers cited by Time put the chance of AI outperforming humans on most tasks by 2027 at 10 percent.
OpenAI’s public timeline release in late 2025 marked a philosophical turn. Altman explained the change. “We used to think that AGI would suddenly appear at some magical moment in the future,” he said. “But now we find that it is more like a process, and you are already on this journey.” The company moved away from a single definition of AGI. It embraced concrete milestones instead. That includes building platforms rather than chasing model releases alone. It restructured into a public benefit corporation alongside its nonprofit roots. It pledged funds toward AI-driven healthcare and resilience projects.
Pachocki offered a practical metric. Judge progress by how complex the tasks become and how long systems can work continuously. GPT-3 managed tens of seconds. GPT-4 stretched to minutes or as much as five hours. The goal sits at days of uninterrupted effort. That gap remains sizable.
Revenue realism now collides with these technical ambitions. OpenAI’s ad strategy must help fund enormous compute commitments. Oracle alone carries $75 billion in AI-linked performance obligations tied to the partnership. If advertising falls short, pressure grows on subscription growth, enterprise deals and fresh capital raises. The April 2026 report of missed internal user and revenue targets already showed how quickly markets punish perceived slips.
So what does this mean for the broader industry? Investors who bet on rapid AGI arrival to justify sky-high valuations face fresh scrutiny. Chipmakers and cloud providers tied to OpenAI’s expansion watch the same signals. And researchers inside and outside the labs recalibrate their own forecasts with each new model release.
Recent weeks have brought no dramatic reversal. X discussions in mid-July 2026 still reference 2027 or 2028 as plausible windows for major leaps, yet skepticism persists. One post highlighted predictions from Altman, Amodei and others aligning around superintelligence near the end of the decade. Another noted the shift from helper tools to independent scientific discoverers in fields like biomedicine and mathematics.
OpenAI continues to ship improvements. GPT-5 variants and agentic systems show genuine advances. But the distance between current capabilities and fully autonomous research agents appears wider than the 2026-2028 roadmap once suggested. The company has not issued a fresh update revising those dates. It has, however, emphasized steady process over sudden breakthroughs.
That message may comfort some. It frustrates others who expected faster transformation. Either way, the gap between aspiration and delivery has become harder to ignore. Markets noticed in April. Technical observers noticed throughout 2025. The coming quarters will test whether OpenAI can close the distance or whether its own targets become another data point in the long history of AI timeline adjustments.
Altman maintains optimism. His firm invests in nuclear power partnerships, robotics and massive data centers. The vision holds. Execution now faces the test of slower-than-hoped progress and tighter-than-expected revenue. The AI trade, as Yahoo Finance framed it, hinges on who gets the numbers right.
WebProNews is an iEntry Publication