Cerebras’s giant IPO is really about OpenAI’s next power play

Cerebras’s giant IPO is really about OpenAI’s next power play

The headline is the $26‑billion‑ish IPO, but the real story sits one layer deeper in the stack. Cerebras going public is not just another exuberant AI listing; it’s a test of whether investors are ready to bankroll an alternative to Nvidia – and, indirectly, a new kind of power for OpenAI. With a $1 billion loan, warrants for tens of millions of shares and a multi‑year, multi‑billion compute deal, OpenAI has quietly become both Cerebras’s biggest customer and one of its most important financial backers. This IPO will tell us whether vertically aligned AI empires – from chips up to models – are now the norm, not the exception.

The news in brief

According to TechCrunch, AI chipmaker Cerebras Systems has finally set terms for its long‑anticipated IPO. The company plans to sell 28 million shares, targeting a price range of $115–$125 per share. At the top of that range, Cerebras would raise around $3.5 billion and debut with a market capitalization of roughly $26.6 billion.

The offer is already heavily oversubscribed: Bloomberg, cited by TechCrunch, reports that banks have received about $10 billion in orders for the $3.5 billion of stock on offer, suggesting pricing could rise.

Cerebras builds wafer‑scale accelerators for AI workloads, claiming advantages over GPU‑based systems for inference performance and energy efficiency. It’s backed by a long list of major VCs and hedge funds. Its customer and investor list also includes OpenAI: TechCrunch notes that in December OpenAI lent Cerebras $1 billion, secured by warrants for more than 33 million shares, and signed a multi‑year compute deal said to exceed $10 billion.

The IPO had been delayed by a U.S. federal review of a strategic investment from Abu Dhabi‑based cloud provider G42, a key Cerebras customer.

Why this matters

Strip away the IPO hype and one thing becomes clear: Cerebras is effectively a leveraged bet on the future of non‑Nvidia AI compute – and OpenAI is standing right in the middle of that bet.

Winners in the short term:

• Cerebras gets a huge capital injection in an era where building leading‑edge chips costs billions per generation. If the book really is 3x covered, management will have pricing power and a strong opening day.
• OpenAI gains optionality upstream in the supply chain. Instead of being only a voracious buyer of GPUs from Nvidia and cloud credits from Microsoft, it now has economic upside in a key alternative hardware vendor.
• Late‑stage investors who came in at an $8.1 billion valuation in September and $23 billion in February could see paper gains within months.

Potential losers:

• Nvidia and the GPU monoculture. One successful wafer‑scale IPO does not dethrone Nvidia, but it sends a signal: there is capital for alternative accelerator architectures, especially for inference where cost per token matters more than maximal flexibility.
• Smaller AI labs and startups. As OpenAI locks in multi‑year, multi‑billion commitments with chip vendors, it further crowds the market for compute capacity. The bar to “own the metal” keeps rising.

The more subtle issue is governance. Several OpenAI executives, according to TechCrunch, personally invested in Cerebras as angels, while OpenAI itself is both a lender and strategic partner. That intertwined cap table blurs the line between what’s good for OpenAI the company, what’s good for its leadership, and what’s good for the broader AI ecosystem. It revives uncomfortable questions from the Musk lawsuit: where exactly does OpenAI’s duty lie when it steers billions towards one supplier it is also financially tied to?

The bigger picture

Cerebras’s IPO slots into a broader pattern: capital markets are once again ready to fund giant, capital‑intensive bets – as long as they’re attached to AI compute.

Over the last 18 months we’ve seen:

• Cloud hyperscalers double down on custom silicon (Google’s TPU, AWS Trainium/Inferentia, Microsoft’s Maia/Cobalt).
• A crop of alternative accelerator startups (Groq, Tenstorrent, Sambanova) chase specific niches – ultra‑low‑latency inference, domain‑specific training, or sovereign deployments.
• Governments throwing money at domestic compute capacity – from the U.S. CHIPS Act to the EU Chips Act and national AI infrastructure programs.

Cerebras is different from many of its peers in two ways:

1. Wafer‑scale design. Instead of dicing wafers into individual chips, Cerebras builds effectively one giant chip per wafer. That’s a radical bet on scale, interconnect and memory locality, optimized for massive neural networks.
2. Anchor customer first, then IPO. The company secured a multi‑year, >$10 billion deal with OpenAI before going public. In that sense, this listing looks less like a risky growth story and more like securitizing a long‑term compute contract.

Historically, we’ve seen similar patterns: ARM’s IPO in 2023 and Mobileye’s in 2022 were both about crystallizing value in foundational infrastructure for the next decade of computing (mobile and ADAS, respectively). Cerebras is the AI‑era sequel: if mobile was about battery‑efficient CPUs, the foundation model era is about energy‑efficient inference at scale.

The competitive question is whether Cerebras can avoid the fate of earlier AI chip hopefuls like Graphcore, which struggled despite impressive technology because Nvidia’s software ecosystem (CUDA, cuDNN, libraries) proved harder to dislodge than its silicon. Cerebras is implicitly betting that anchor customers like OpenAI, plus an integrated hardware‑software stack, can overcome that barrier.

The European / regional angle

For Europe, Cerebras’s IPO is not merely a Wall Street curiosity. It touches on three strategic European concerns: compute sovereignty, regulatory leverage and competitive diversity.

First, European AI startups like Mistral, Aleph Alpha and Stability AI routinely complain about access to affordable, large‑scale compute. Today, their options are dominated by U.S. hyperscalers renting Nvidia GPUs. A credible, power‑efficient alternative like Cerebras – especially if deployed in regional clouds or on‑premise HPC centres – could give European players more bargaining power.

Second, Brussels is tightening the screws on both AI and cloud. The EU AI Act, Digital Markets Act (DMA) and upcoming cloud interoperability rules all aim, in different ways, to prevent a small group of U.S. gatekeepers from controlling the entire stack. If OpenAI plus a handful of U.S. chip vendors become a new vertically integrated gatekeeper, expect EU regulators to take a keen interest.

Third, European cloud providers – OVHcloud, Scaleway, Deutsche Telekom, smaller regional data centre operators – are actively looking for differentiators against AWS, Azure and Google Cloud. Partnering with alternative accelerators like Cerebras fits nicely into that story: “We can’t match the scale of AWS, but we can offer sovereign, energy‑efficient AI compute on non‑Nvidia hardware.”

There is also a risk. If Cerebras becomes financially and strategically dependent on OpenAI and U.S. policy, it may end up as “Nvidia 2.0” from a European perspective: another critical dependency, outside EU jurisdiction, subject to U.S. export controls and geopolitical swings.

Looking ahead

Three questions will determine whether Cerebras’s IPO becomes a footnote or a turning point.

1. Can it escape customer concentration risk? Right now, OpenAI looks like an outstanding asset – a creditworthy anchor tenant buying a lot of compute. But dependence cuts both ways. If OpenAI’s strategy changes, if regulators force it to diversify suppliers, or if Microsoft pulls more of OpenAI’s workloads into its own custom silicon, Cerebras could suddenly look exposed.

2. Can it build a true ecosystem? Nvidia’s moat is less the GPU and more CUDA plus years of optimization work by thousands of developers. Cerebras needs more than a few flagship customers; it needs toolchains, libraries, cloud integrations and a generation of engineers who treat its hardware as a first‑class target. That requires long, patient investment after the IPO glow fades.

3. How will regulators react to the OpenAI entanglement? The mix of loans, warrants and executive angel stakes is the sort of thing competition and securities regulators scrutinize after, not before, blockbuster listings. If OpenAI later moves to deepen the integration – say, through a strategic takeover – it could trigger serious antitrust and conflict‑of‑interest debates on both sides of the Atlantic.

In the near term, expect a volatile but strong trading debut if current demand holds. Over the next 12–24 months, watch for:

• New, non‑OpenAI marquee customers (especially in cloud and national AI centres).
• Announcements of European deployments or partnerships.
• Any hint that OpenAI wants board seats, governance rights or tighter control.

If all three converge, Cerebras may shift from “interesting chip IPO” to “critical node in the global AI power structure.”

The bottom line

Cerebras’s planned IPO is about more than one company cashing in on AI mania. It’s a live experiment in whether capital markets will back alternatives to Nvidia – and in how far OpenAI can extend its influence down into the hardware layer. If Cerebras executes, it will give the AI ecosystem badly needed diversity in compute. If it stumbles, the lesson will be harsh: in the age of foundation models, even billion‑dollar wafers may not be enough to crack Nvidia’s grip. As investors pile in, users should be asking: who do we want to own the engines of intelligence?