# Sam Altman Warns of 'Hiccups' as GPT-5.6 Sol Demand Strains OpenAI Infrastructure > Source: > Published: 2026-07-15 10:17:38.626677+00:00 # Sam Altman Warns of 'Hiccups' as GPT-5.6 Sol Demand Strains OpenAI Infrastructure - Sam Altman said on X that GPT-5.6 Sol's 'growth is insane' and warned of possible 'hiccups' as inference infrastructure struggles to keep up with demand [[1]](https://x.com/sama/status/2077106587307798989) - Codex and ChatGPT Work have reached 8 million active users, up from 5 million previously, forcing OpenAI to remove its five-hour rate limit and reset usage caps [[2]](https://cryptobriefing.com/codex-8-million-users-daily-resets/) - GPT-5.6 Sol launched publicly on July 9 after a 12-day government-gated safety review, with API pricing at $5 input / $30 output per million tokens [[3]](https://www.axios.com/2026/07/14/sam-altman-chat-gpt-sol-ultra-warning) - OpenAI is also deploying Sol on Cerebras hardware at up to 750 tokens per second for enterprise customers requiring real-time inference [[4]](https://cryptobriefing.com/openai-sol-model-altman-hiccups-warning/) - Microsoft Azure remains the exclusive cloud provider for OpenAI's stateless APIs, with GPT-5.6 expected to drive higher usage-based revenue [[5]](https://www.tradingkey.com/analysis/stocks/us-stocks/262018612-openai-gpt-5-6-off-balance-sheet-ipo-delay-softbank-ai-bet-tradingkey) OpenAI CEO Sam Altman warned Tuesday that the company's new flagship GPT-5.6 Sol model faces potential service disruptions as explosive user growth outpaces the company's ability to add inference capacity. "5.6 Sol growth is insane. The inference team has done heroic work to be able to support demand. We are going to move mountains to continue to scale, but it is possible there are some hiccups soon," Altman wrote on X [1]. The warning came as OpenAI disclosed that Codex and ChatGPT Work — the company's AI coding assistant and enterprise productivity agent, respectively — have reached 8 million active users, up from 5 million weekly Codex users reported earlier [2]. The surge followed Sol's July 9 public launch, which itself was delayed nearly two weeks by a government-mandated safety review under the Trump administration's voluntary AI framework . [[3]](https://www.axios.com/2026/07/14/sam-altman-chat-gpt-sol-ultra-warning)OpenAI has already taken steps to manage the load, removing a five-hour rate limit and resetting usage caps for Plus, Pro, Business, and Enterprise subscribers to expand access across both platforms [2]. The company is also running Sol on Cerebras inference hardware at speeds of up to 750 tokens per second for enterprise customers . [[4]](https://cryptobriefing.com/openai-sol-model-altman-hiccups-warning/)## What Happened Altman's post on X on July 14 was a preemptive acknowledgment that OpenAI's infrastructure may not be able to fully absorb the demand curve for its most powerful model. For developers and enterprises relying on the OpenAI API, the warning signals potential degraded performance, slower response times, or temporary service interruptions in the near term [1]. GPT-5.6 Sol is the top tier in a new three-model family that also includes Terra ($2.50/$15 per million tokens) and Luna ($1/$6 per million tokens). Sol is designed for complex reasoning, extended coding sessions, and agent-driven workflows, and delivers what OpenAI says is a 54% improvement in token efficiency for agentic coding tasks over prior models [3]. The model first entered a limited preview on June 26, available only to 20 government-approved partners, before opening to the public on July 9 . [[3]](https://www.axios.com/2026/07/14/sam-altman-chat-gpt-sol-ultra-warning)## Why It Matters The capacity warning underscores a structural tension facing all frontier AI labs: the gap between model capability and the infrastructure required to serve it at scale. Inference compute — the processing power needed to run a trained model for each user query — is now the primary bottleneck, not training. As models grow more capable and attract more users, the cost and complexity of keeping them running in real time escalates rapidly. For OpenAI's enterprise customers, particularly those integrating Sol through the API for production applications, the prospect of intermittent outages or throttling raises reliability questions. The company's Codex product has become central to many development workflows, and ChatGPT Work is being positioned as a full productivity agent capable of handling research, data analysis, documents, and presentations [2]. Microsoft, which provides Azure as the exclusive cloud platform for OpenAI's stateless APIs, is directly exposed to both the upside and the operational strain. Analysts expect GPT-5.6 to drive higher usage-based revenue through Azure AI and Copilot integrations [5]. Microsoft shares closed at $384.93 on Tuesday, down 1.5% on the day and off 20.4% year-to-date . [[6]](https://financialmodelingprep.com)## Competitive Context The infrastructure strain arrives at a moment of intensifying competition. Anthropic and other frontier labs are rolling out their own flagship models, creating a race not just for model quality but for the compute capacity to serve users reliably. The ability to scale inference without degradation has become a competitive differentiator. OpenAI's decision to deploy Sol on Cerebras hardware alongside its primary Azure infrastructure reflects an effort to diversify its compute supply chain. The Cerebras deployment targets enterprise customers needing real-time inference at up to 750 tokens per second — a speed benchmark aimed at latency-sensitive applications [4]. The broader context is a company under significant financial pressure. Reports have cited $665 billion in off-balance-sheet liabilities, a delayed IPO, and heavy dependence on Microsoft and SoftBank capital [5]. Successfully scaling Sol without major outages is a credibility test for OpenAI's ability to operate frontier AI as a reliable commercial service. ## What's Next Altman's framing was notably forward-looking — a CEO preparing users for the possibility of disruptions rather than announcing a specific outage. The practical question is whether OpenAI can add inference capacity fast enough to keep pace with adoption, or whether it will need to impose usage restrictions, queue systems, or tiered access to manage load. For enterprise customers evaluating long-term AI infrastructure commitments, the episode highlights the importance of redundancy and multi-provider strategies. As AI models become embedded in mission-critical workflows, the tolerance for downtime narrows — and the stakes of getting infrastructure scaling right continue to rise. ## Companies mentioned ## Further sources [[1] Sam Altman post on X warning of GPT-5.6 Sol infrastructure hiccups, July 14, 20… ↗](https://x.com/sama/status/2077106587307798989) [[2] CryptoBriefing: Codex reaches 8 million active users as OpenAI resets daily lim… ↗](https://cryptobriefing.com/codex-8-million-users-daily-resets/) [[3] Axios: Sam Altman warns of 'hiccups' with new flagship Sol model, July 14, 2026 ↗](https://www.axios.com/2026/07/14/sam-altman-chat-gpt-sol-ultra-warning) [[4] CryptoBriefing: OpenAI's Sol model delivers 54% token efficiency improvement, C… ↗](https://cryptobriefing.com/openai-sol-model-altman-hiccups-warning/) [[5] TradingKey: Behind OpenAI's Public Release of GPT-5.6 — $665B in off-balance-sh… ↗](https://www.tradingkey.com/analysis/stocks/us-stocks/262018612-openai-gpt-5-6-off-balance-sheet-ipo-delay-softbank-ai-bet-tradingkey) [[6] Microsoft (MSFT) real-time quote via FMP, July 15, 2026 ↗](https://financialmodelingprep.com) The stories that matter, in one email. Free — unsubscribe anytime.