Consider the quiet desperation of a market starving for alternatives. When a single company, Cerebras, announces a $25 billion order backlog, the entire AI infrastructure ecosystem holds its breath. But as someone who spent 600 hours auditing Aave V2’s interest rate models during the DeFi summer, I learned a simple truth: numbers that sound too good to be true usually are. The claim, reported by Crypto Briefing, arrives just as whispers of Cerebras’ IPO circulate. It is a perfect bull-market story: demand so insatiable that a startup with less than $1 billion in cumulative revenue is suddenly a $25 billion giant. But beneath the headline lies a far more interesting tale—one of IPO theater, intentional ambiguity, and the fragility of trust in a hype-driven market.
At its core, this is not about Cerebras versus NVIDIA. It is about what happens when we confuse aspiration with execution. The $25 billion figure, if taken at face value, implies that Cerebras has secured orders equivalent to half of NVIDIA’s entire data center revenue in FY2024—a company that ships millions of GPUs annually. Cerebras, by contrast, shipped roughly 100 CS-2 systems in 2023. The math does not add up. Based on my experience analyzing tokenomics and order-book manipulation in DeFi protocols, I recognize this pattern: an inflated metric designed to anchor investor expectations pre-IPO. The backlog likely includes non-binding letters of intent, multi-year framework agreements, and even hypothetical demand projections—all rolled into a single dramatic number. Transparency isn’t the oxygen of trust.
To understand the real story, we must deconstruct three layers. First, the commercial context. Cerebras’ known customers—G42, the U.S. Department of Energy—have signed contracts in the tens of millions, not billions. G42’s $100 million deal was celebrated. To reach $25 billion, Cerebras would need 250 such customers or a single customer committing $10 billion+. The latter is possible (a sovereign AI fund, perhaps), but the absence of named backers is telling. In crypto, we call that “trust me, bro.” Second, the competitive dynamics. Cerebras’ WSE-3 chip excels at training monolithic LLMs, but its ecosystem is minuscule compared to CUDA. Clients who sign large pre-orders are betting on a platform that may not see major software adoption for years. Third, the infrastructure reality. A $25 billion order implies ~50,000 WSE-3 chips, demanding over 1GW of power—equivalent to a nuclear reactor. Where are these data centers? Cerebras has not announced any new 500MW facilities. The claim is infrastructure vaporware.
Yet, I do not dismiss the possibility that some portion of this backlog is real. In 2021, I curated an NFT exhibition rejecting speculation, and learned that even small communities can create outsized signals. If even $5 billion of this backlog is genuine—say, from a Middle Eastern sovereign wealth fund building a national AI grid—it represents a legitimate shift in the compute supply chain. Cerebras is the most credible non-NVIDIA alternative for large-scale training. But the gap between a credible alternative and a $25 billion backlog is the difference between a seed and a forest. Code is law, but ethics is soul. The ethics of this announcement matter because it sets false expectations for hardware markets, influences VC allocation, and potentially misleads regulators about the true state of AI hardware competition.
Now, let me offer a contrarian angle that might surprise you. The exaggeration itself is a rational strategy for a company facing an existential technology window. NVIDIA’s B100 and GB200 architectures, due in 2025, will narrow the interconnect advantage that Cerebras currently enjoys. If Cerebras does not capture massive orders now—before NVIDIA’s next generation locks in customers—its window closes. So the $25 billion number is a competitive weapon, not just a PR stunt. It signals to potential clients: “We have momentum, join the migration.” This is similar to how Solana’s “500k TPS” claim in 2021 was technically achievable under ideal conditions, but never in production. The market rewarded the narrative until reality caught up. Cerebras is playing the same game, but with hardware that actually works—just not at the scale suggested.
What does this mean for us—builders, ethicists, and open-source advocates? First, verify before amplifying. I wrote a 15,000-word manifesto on GitHub titled “Trustless but Not Careless” after Aave’s near-miss. The same principle applies here: demand third-party audits of these backlog claims. Where are the SEC filings? Where are the customer confirmations? Second, recognize that hype has consequences. In the crypto space, exaggerated throughput claims led to misallocated capital in infrastructure tokens. In AI hardware, inflated order books could distort capex decisions for hyperscalers, leading to overbuilding of specific chip architectures while neglecting software diversity. Third, focus on the signal within the noise. The real insight from this announcement is not the $25 billion, but the desperation for alternatives. Enterprises are so NVIDIA-dependent that they are willing to sign speculative LOIs with a startup. That demand is real, and it creates opportunities for other open-source hardware efforts like RISC-V AI accelerators. But we must ensure these opportunities are built on honest foundations, not inflated promises.
As I sit here in Lisbon, reflecting on the Ethereum whitepaper translation I did in 2017—distributing 5,000 physical copies at Web Summit—I remember why I left academia for this space: because decentralization promised a truthful ledger. But ledgers only work if we commit to honest inputs. A $25 billion backlog that is actually $2 billion in signed deals and $23 billion in wishful thinking is not a ledger; it is a fairy tale. We need to guard the commons of honest infrastructure by demanding rigor, even—especially—from the projects we admire. Guard the commons, or lose the future. Let us not let bull-market euphoria blind us to the technical and ethical gaps that remain. The real story is not Cerebras’ order book; it is our collective capacity to see through the noise and build infrastructure that earns trust through transparency, not scale.
The takeaway is a question: In a market that rewards narratives over reality, what responsibility do we—as evangelists, analysts, and engineers—bear to call out the gap between code and claim? The answer determines whether the next decade of AI infrastructure is built on sand or stone.
