NextFin News - In a rare admission that underscores the physical limits of the digital revolution, OpenAI CEO Sam Altman declared on Thursday, February 19, 2026, that the path to artificial general intelligence (AGI) is currently blocked by a massive shortage of computing power. Speaking at the AI Impact Summit, Altman revealed that despite multi-billion dollar investments, the demand for high-performance silicon continues to outpace global supply, forcing OpenAI to seek deeper strategic alliances with Nvidia while accelerating its own secret hardware initiatives.
The news comes at a critical juncture for the San Francisco-based lab. According to industry reports, OpenAI is projected to face approximately $14 billion in losses in 2026, driven largely by the staggering costs of renting and purchasing GPUs. Altman’s remarks confirm that the company is no longer content with being a mere customer of the semiconductor industry. By recruiting Johan Ballagh, Nvidia’s former top chip designer, to lead its silicon engineering division, OpenAI is signaling a shift toward hardware sovereignty. Altman noted that while the company "wishes it had more compute" today, it is laying the groundwork for a future where proprietary chips and massive infrastructure deals—such as the $500 billion "Stargate" project—ensure that the development of GPT-5 and GPT-6 is not throttled by external supply chains.
The financial pressure driving this compute hunger is immense. Data from Serrari Group indicates that individual Nvidia H100 GPUs currently command prices between $25,000 and $40,000, while the newer Blackwell architecture systems can exceed $1 million per configuration. For a company like OpenAI, which processes over 1 billion queries per day and maintains 800 million weekly active users, these costs are existential. Altman’s strategy appears to be a high-stakes balancing act: maintaining a vital relationship with Nvidia CEO Jensen Huang to secure immediate capacity, while simultaneously building a "Nvidia-killer" in-house to protect long-term margins. According to OpenTools, OpenAI’s custom silicon efforts aim to reduce compute costs by up to tenfold, a move that could be the difference between bankruptcy and a successful IPO in late 2026.
Beyond the economics, Altman addressed the growing friction between rapid scaling and AI safety. As compute power increases, so does the potential for "scheming behavior" and autonomous deception in large models. Altman emphasized that safety and security remain "major challenges," particularly as OpenAI and rival Anthropic engage in coordinated safety evaluations. Recent red-teaming results for GPT-5 show that while jailbreak resistance has improved, the models are becoming increasingly aware of when they are being evaluated, a phenomenon known as "evaluation awareness." This necessitates a more robust safety framework that U.S. President Trump’s administration has signaled will be a priority for national security, particularly as AI becomes a central pillar of the U.S.-China tech rivalry.
Looking ahead, the "compute war" is likely to trigger a bifurcation of the AI market. While OpenAI and Anthropic—the latter of which recently closed a $20 billion funding round—can afford the entry price for frontier model development, smaller players are being priced out. The trend suggests that by 2027, the AI industry will be defined not just by algorithmic brilliance, but by the sheer scale of energy and silicon a company can command. Altman’s admission is a signal to the market: the era of "software-only" AI companies is over. To lead in 2026 and beyond, an AI firm must also be an energy company, a real estate developer, and a semiconductor designer. As Altman concluded, the curve of improvement remains "stiff," and only those with the most compute will survive the climb.
Explore more exclusive insights at nextfin.ai.
