Is OpenAI Losing the AI War to Google and Anthropic? The $100 Billion Nvidia Deal on Ice — What Now
The artificial intelligence frontier is no longer a one-company story. For much of the post-ChatGPT era, OpenAI defined the pace of progress. Its models set the benchmarks, its consumer product dominated mindshare, and its partnerships secured the compute needed to stay ahead. By mid-2026 that picture has changed. Anthropic’s Claude models now lead multiple independent intelligence and agentic rankings. Google continues to leverage unmatched distribution even as its highest-end Gemini releases face delays. And the headline $100 billion Nvidia partnership that was supposed to lock in OpenAI’s next wave of infrastructure has largely stalled in its original form.
The result is a genuine multipolar contest rather than a runaway lead. OpenAI remains a powerhouse by revenue and user base, but the question of whether it is losing relative ground is no longer fringe.
Where the Models Stand
Independent evaluations in August 2026 consistently place Anthropic’s latest systems at or near the top of composite intelligence indexes. Claude Opus 5 and Claude Fable 5 occupy the leading positions on several Artificial Analysis and BenchLM-style leaderboards that aggregate coding, reasoning, knowledge work, and agentic performance. OpenAI’s GPT-5.6 Sol sits close behind—often within a few points—and remains highly competitive on coding-agent tasks and multimodal work. Google’s Gemini family trails further on pure capability rankings, though its long-context and multimodal strengths remain relevant for certain enterprise workflows.
This is not a permanent ranking. Model leadership has flipped multiple times since 2023. What matters is the pattern: Anthropic has converted technical credibility into enterprise adoption. Multiple data sources show Anthropic capturing the largest share of paid enterprise LLM API spend, while OpenAI retains a clear majority of consumer web traffic through ChatGPT. Google benefits from default placement across Search, Android, and Workspace, yet quality shortfalls and delayed frontier releases have limited its ability to monetize that distribution at the highest tier.
Revenue figures reflect the same split. OpenAI’s annualized revenue run rate has surpassed $40 billion, roughly doubling from late 2025 levels, driven by subscriptions, coding tools, enterprise contracts, and early advertising experiments. Anthropic has posted its own aggressive growth trajectory, with some reports placing its run rate in a similar or higher range depending on methodology. Both companies are preparing for public markets. The competitive dynamic is therefore not collapse versus dominance; it is two well-funded labs trading blows while a third player with superior distribution tries to close the quality gap.
The Nvidia Partnership That Did Not Materialize as Announced
In September 2025, OpenAI and Nvidia announced a letter of intent for what was described as the largest AI infrastructure deployment in history. The plan called for at least 10 gigawatts of Nvidia systems, with Nvidia intending to invest up to $100 billion in OpenAI progressively as capacity came online. The first gigawatt was targeted for the second half of 2026 on the Vera Rubin platform.
By early 2026 that equity-style commitment had effectively evaporated. Nvidia’s chief executive publicly stated that the $100 billion figure was never a firm commitment. Reports indicated OpenAI had explored alternative chip suppliers and expressed dissatisfaction with certain inference performance characteristics. The original structure—large-scale equity investment tied directly to successive gigawatts—did not close.
What has emerged instead is a different, still-massive arrangement centered on an Ohio data-center campus developed with SoftBank’s SB Energy. Nvidia has agreed to provide residual-value guarantees capped around $105 billion for the initial phase (approximately 4.25 gigawatts of an eventual 8-gigawatt project), with exclusivity on its chips for that phase. Capacity is not expected online until 2028. The numbers remain enormous, but the risk profile, timeline, and structure differ sharply from the 2025 letter of intent. The original $100 billion equity deal is, for practical purposes, on ice.
Compute is the binding constraint for frontier training. Any slippage or restructuring of supply pipelines forces harder prioritization among research, productization, and cost control. OpenAI still has access to substantial capacity through Microsoft and other partners, yet the loss of the cleanest, largest single Nvidia commitment reduces optionality at a moment when rivals are also racing to secure power and silicon.
Additional Headwinds and Industry Pressures
OpenAI has also slowed work on its next major internal model, referred to as Astra. In early August 2026 the company disclosed that preliminary evaluations could not rule out “Critical” cybersecurity capabilities under its own Preparedness Framework—the first time an OpenAI system had approached that threshold. Portions of internal development and large-scale reinforcement-learning runs were paused or placed under stricter isolation and monitoring while safeguards are strengthened. No public release date has been set. Prediction markets currently lean toward a late-2026 launch at earliest.
Meanwhile the broader market is experiencing price pressure. Both OpenAI and Anthropic have cut prices on mid-tier and high-volume models in response to cheaper Chinese open-weight systems that have gained meaningful token-usage share. Token prices have softened even as GPU rental costs remain firm. Capital intensity continues to rise. Training and inference costs are measured in tens of billions annually, and circular financing concerns around chipmakers and labs have drawn investor scrutiny.
Google faces its own challenges—talent departures, delayed high-end Gemini releases, and the difficulty of disrupting its profitable search business while still competing at the frontier. Its advantages in custom silicon (TPUs), data-center scale, and product distribution remain real, but they have not yet translated into consistent leadership on the models that enterprises pay premium prices for.
What Comes Next
OpenAI is not disappearing. It still commands the largest consumer brand in AI, generates tens of billions in annualized revenue, and retains deep technical talent and Microsoft backing. Anthropic’s lead on certain benchmarks and enterprise metrics is real but narrow and potentially reversible with the next model cycle. Google’s distribution edge is structural. The original $100 billion Nvidia vision has been replaced by a more cautious, phased infrastructure commitment with a later timeline.
The practical implications are straightforward. Frontier progress will remain capital- and power-constrained. Labs that secure reliable next-generation silicon earliest will train and iterate faster. Safety frameworks are beginning to impose real schedule costs, as OpenAI’s Astra pause demonstrates. Price competition will continue to compress margins on commodity inference while the highest-capability tiers retain pricing power for complex agentic and reasoning workloads.
For OpenAI the path forward requires executing on the next model generation under tighter safety constraints, converting its consumer scale into durable enterprise revenue, and locking in diversified compute supply beyond any single partnership. For the industry as a whole, the era of unchallenged single-lab dominance has ended. The AI race in 2026 is a three-way contest of models, money, and megawatts—and no participant can afford to assume the lead is permanent.
About The Author
