The Original Sin: A Calculated Risk Gone Wrong
Dario's decision two years ago to make an overly conservative investment in GPU capacity now haunts Anthropic to this very day. Despite boasting Fable, arguably the smartest and biggest model on the planet, they simply cannot serve it, constrained by their past strategic misstep. This initial frugality created a bottleneck directly impacting customer access and model utility, a critical flaw OpenAI keeps exploiting.
OpenAI, in stark contrast, pursued an aggressive, forward-looking compute acquisition strategy during the same period. They continue to offer the most generous quotas as part of paid subscriptions, resetting them seemingly every other day, showcasing a commitment to user experience Anthropic cannot match. Look at Sam Altman's tweet, Throwing salt on the wound: "Come for the best model, stay because we don't treat you with contempt." This is a direct shot, revealing the stark difference in customer treatment.
This was no mere oversight; it was a critical miscalculation in the face of exponential AI model scaling and insatiable demand. Anthropic's compute deficit renders their superior model less accessible—they might not even give Fable as part of a subscription. Meanwhile, OpenAI's GPT 5.6 Sol, almost as good and much more efficient, is widely usable due to abundant resources. The original sin of compute conservatism now dictates their market position, a self-inflicted wound.
The Inference Tax: Fable's Achilles' Heel
Anthropic may possess the smartest model on the planet, Fable, But their current reality is a cruel paradox: they simply cannot serve it. Dario's conservative compute investment Two years ago now imposes a crippling inference tax on users, creating a competitive chasm that OpenAI relentlessly exploits. Anthropic has the best model, And it also happens to be the biggest model, yet its creators struggle to deploy it at scale.
Users bear the brunt of this capacity crisis. Strict rate limits hamstring productivity, And high operational costs translate directly into premium pricing for access. Worse, Anthropic signals deep uncertainty about Fable's inclusion in standard subscriptions, leaving customers in limbo regarding access to their flagship model. This situation forces users to pay more for less certainty, a frustrating proposition for enterprise clients and developers alike.
This poor customer experience is Anthropic's Achilles' heel, making them acutely vulnerable to rivals. Sam Altman’s pointed tweet, "Come for the best model, stay because we don't treat you with contempt," perfectly encapsulates the market sentiment, a direct shot at Anthropic's perceived disregard for its user base. While Fable excels in raw intelligence, OpenAI's GPT 5.6 Sol offers a nearly equivalent experience, But with superior efficiency, much lower costs, And far more generous usage quotas. This rival model, almost as good as Fable, allows users to leverage it so much more, effectively exploiting Anthropic's self-inflicted wound and driving customers away.
OpenAI's Salt-in-the-Wound Strategy
OpenAI, ever the opportunist, has not merely observed Anthropic’s self-inflicted wounds; it has actively poured salt into them. Competitors are leveraging Anthropic's capacity crisis by offering generous quotas for powerful models like GPT 5.6 Sol, resetting access seemingly every other day as part of their paid subscriptions. This aggressive accessibility stands in stark contrast to Anthropic’s current predicament, where Claude Fable 5, despite its prowess, struggles to find adequate compute to serve its users.
Sam Altman, never one to miss a competitive opening, delivered a public broadside that resonated deeply. He famously tweeted: "Come for the best model, stay because we don't treat you with contempt." This was a direct, unmistakable jab at Anthropic, exploiting their strained customer experience and perceived disdain for users amid the ongoing compute crunch. It framed the entire debate as a matter of respect.
This cunning strategy has paid dividends, winning over developers and users frustrated by Anthropic's limitations. While Claude Fable 5 may be technically the "best model," GPT 5.6 Sol is a very good, highly accessible alternative: almost as capable, much more efficient, and significantly less expensive. OpenAI’s generous quotas mean users can leverage it far more often, making it the practical choice for those who need reliable, scalable AI access. Anthropic's desperate search for compute continues, as detailed in Anthropic's desperate search for compute - Runtime, highlighting the stark contrast in operational realities.
The Long Game: Can Billions in GPUs Fix This?
Anthropic's leadership, having miscalculated two years ago, now executes a colossal course correction. Deals with Amazon and Google secure up to 10 GW of compute power by 2027, a staggering investment designed to close the capacity gap. This aggressive, multi-billion-dollar scramble, however, starkly highlights the severity of Dario's initial, conservative GPU bet, a misstep now costing them dearly in market share and developer trust.
Enjoying this? Get one like it in your inbox each morning.
one email a day · unsubscribe in two clicks · no third-party tracking
Perhaps a saving grace emerges, a long shot But a plausible one: Recursive Self-Improvement (RSI). If Fable 5's architectural lead is truly profound, it might autonomously compound its advantage, generating new data and insights to iteratively improve itself. This inherent momentum, fueled by its own superior intelligence, could potentially allow Fable to outpace competitors, even those like OpenAI, who currently wield superior immediate compute access and offer generous quotas for models like GPT 5.6 Sol.
Look, the generational race for AGI hinges on this. Was Dario's initial compute blunder merely a temporary stumble, albeit an expensive one, that Anthropic can overcome with billions in new infrastructure by 2027? Or did that critical, two-year delay prove a fatal flaw, granting Sam Altman's OpenAI an insurmountable head start and defining whether Anthropic truly had the "smartest model on the planet" But failed to capitalize, or if their long game ultimately prevails?
Frequently Asked Questions
What was Anthropic's big mistake regarding compute?
Approximately two years ago, Anthropic's leadership was reportedly 'overly conservative' with investments in GPU capacity. This miscalculation led to a significant compute shortage as their models grew larger and more powerful.
How is OpenAI exploiting Anthropic's compute shortage?
OpenAI is offering more generous API quotas and a better user experience for its powerful models like GPT 5.6 Sol. CEO Sam Altman has also publicly criticized Anthropic's customer treatment, aiming to attract users frustrated with Anthropic's access limitations.
What is 'Fable' and why is its access limited?
Claude Fable 5 is Anthropic's most powerful, Mythos-class model. Its immense size requires vast computational resources, which Anthropic currently struggles to provide at scale, leading to restrictive quotas and temporary service suspensions.
Can Anthropic still win the AI race?
Despite the setback, Anthropic might still succeed. They are aggressively securing new compute capacity (up to 10 GW). Furthermore, if their model's lead in quality is substantial, it could achieve Recursive Self-Improvement (RSI) first, potentially creating an insurmountable advantage.