To The Moon

Inference became theatre. Investors arrived before the final act

The IPO That Arrived Exactly on Cue

Cerebras Systems did not quietly tiptoe onto the public markets. It marched in wearing a brass band, carrying a wafer-sized silicon dinner plate, and demanding Wall Street’s full attention. At one point, investors valued the company at roughly $95 billion following its explosive debut, briefly treating it less like a semiconductor firm and more like the AI equivalent of discovering fire.

What fascinates me is not merely the technology. The real story is the timing.

Cerebras went public at the precise moment the AI narrative flipped from training models to running them. For the past two years, investors obsessed over who could build the biggest large language model. Now the market cares about inference: the daily, relentless processing of billions of AI requests from agents, copilots, enterprise systems, and consumer applications.

That transition matters enormously because inference economics are very different from training economics. Training rewards brute-force scale. Inference rewards efficiency, latency, and power consumption. Cerebras’ wafer-scale architecture suddenly looks perfectly designed for this new world — which is exactly why I think investors may be underestimating the danger.

Markets are remarkably good at overpaying for companies that arrive at exactly the right macro moment. Cisco experienced it. Sun Microsystems experienced it. Entire fibre-optic businesses experienced it before discovering that narrative momentum has the lifespan of supermarket sushi.

The challenge for $Cerebras Systems(CBRS)$ is not proving the 'Inference Flip' exists. The challenge is proving that today’s economics survive competition tomorrow.

CBRS since its IPO debut on 14 May 2026 — 30-minute intervals capturing the market’s attempt to price architectural ambition against execution reality. Source: TradingView

One Giant Chip, One Giant Gamble

Cerebras’ core innovation is genuinely extraordinary. Rather than linking together thousands of smaller GPUs, the company builds a single Wafer-Scale Engine that keeps compute, memory, and communication on one giant slab of silicon.

It is an elegant solution to one of AI’s ugliest problems: latency.

Every time traditional GPU clusters communicate across networking layers, efficiency deteriorates. Cerebras attempts to bypass that bottleneck entirely. The result is extremely fast inference performance and simplified scaling for certain AI workloads.

This is where the bullish case becomes seductive. If AI agents become persistent digital workers operating continuously across enterprises, then inference demand could eventually dwarf training demand. In that scenario, specialised architectures may outperform general-purpose GPUs economically.

Wall Street clearly believes this possibility is real.

At roughly $63 billion to $65 billion in market capitalisation, Cerebras trades at more than 120 times trailing sales despite generating only around $510 million in trailing revenue. Its enterprise-value-to-revenue ratio sits above 123, placing it among the most aggressively valued infrastructure plays in the AI market.

The market is effectively assuming Cerebras becomes foundational AI infrastructure before competitors compress margins, replicate functionality, or undercut pricing.

That is an enormous assumption for a business still generating negative EBITDA and levered free cash flow of roughly minus $623 million.

The OpenAI Paradox

One detail continues to bother me more than the valuation itself.

Most investors view Cerebras’ relationship with OpenAI as pure validation. I think it is simultaneously validation and vulnerability.

OpenAI committed to a massive multi-year compute arrangement extending through 2028, transforming Cerebras from speculative outsider into strategic AI supplier almost overnight. The agreement gave investors confidence that Cerebras’ hardware is commercially relevant at hyperscale.

But there is a catch hidden beneath the excitement.

OpenAI reportedly holds warrants for up to 10% of the business. That creates an embedded conflict between Cerebras’ need to protect margins and its dependence on the very partner that benefits most from squeezing them.

In simpler terms, Cerebras needs pricing power. OpenAI needs leverage.

That tension rarely resolves elegantly in technology history.

One underappreciated risk is that hyperscale AI customers increasingly behave less like customers and more like infrastructure architects themselves. Amazon Web Services is already developing its own Trainium accelerators, while Google has spent years deploying proprietary TPUs internally. The largest AI customers are steadily becoming semiconductor strategists themselves.

The strategic implications are significant because these companies no longer simply purchase chips; they influence design, deployment, pricing, and supply allocation across the AI stack.

Independent chip firms can gradually find themselves reduced from strategic partners to highly specialised suppliers serving platforms with vastly greater negotiating leverage.

That is not always where long-term shareholder economics remain strongest.

Foundry Physics

Most analyses frame Cerebras versus Nvidia as an architectural war. I think that misses the real battlefield entirely.

The true choke point is manufacturing allocation.

Cerebras’ wafer-scale architecture is undeniably brilliant, but it also creates a commercial problem almost nobody is discussing properly. The company is not simply competing against Nvidia for customers; it is competing against $NVIDIA(NVDA)$, $Apple(AAPL)$, $Advanced Micro Devices(AMD)$, and Broadcom for access to the same advanced manufacturing ecosystem at $Taiwan Semiconductor Manufacturing(TSM)$.

The implications are far more serious than most investors appreciate because wafer-scale silicon is not an ordinary production workload.

Cerebras effectively asks TSMC to dedicate enormous advanced-node capacity to a highly specialised product that generates a fraction of the aggregate revenue produced by larger hyperscale customers. In practical terms, every advanced wafer allocated to Cerebras is capacity that cannot simultaneously be used for Nvidia’s GPUs, Apple’s iPhones, or custom AI accelerators ordered at dramatically larger scale.

Foundries optimise economics, throughput, yield, and long-term strategic relationships. They do not optimise for technological elegance.

That creates an uncomfortable asymmetry for Cerebras investors. The company’s architectural advantage may simultaneously be its greatest industrial vulnerability. Wafer-scale systems are extraordinarily difficult to manufacture efficiently, and they consume disproportionate engineering attention within a supply chain already operating near the limits of advanced packaging and fabrication capacity.

In a tightening semiconductor cycle, TSMC has every commercial incentive to prioritise customers capable of absorbing millions of units annually rather than niche, capacity-intensive deployments tied to a comparatively small revenue base.

This is where Nvidia’s position becomes especially dangerous for Cerebras.

Nvidia’s moat is no longer merely CUDA software or GPU performance. Its scale now gives it gravitational pull across the manufacturing ecosystem itself. The company’s demand profile makes it strategically indispensable to suppliers, packaging partners, and foundries. That level of industrial leverage is extraordinarily difficult for a newly public company to replicate, regardless of architectural ingenuity.

In other words, Cerebras may not simply need superior technology to win. It may need continued preferential access to the world’s most constrained manufacturing infrastructure at precisely the moment AI demand is exploding globally.

That is a far harder challenge than beating benchmark scores in a laboratory demonstration.

Financials Beneath the Fireworks

To Cerebras’ credit, the revenue trajectory is undeniably impressive.

Quarterly revenue growth exceeded 1,400% year-on-year, while trailing revenue approached $510 million. Gross profit reached roughly $199 million, suggesting the underlying economics of the hardware are not fundamentally broken.

The balance sheet is also healthier than many high-growth hardware peers. Cerebras carries only about $28 million in debt against over $209 million in cash, with a current ratio above 2. That provides some operational breathing room.

Yet the income statement still reveals a business in aggressive expansion mode rather than sustainable profitability mode.

Operating margins remain deeply negative at roughly minus 31%, EBITDA sits below zero, and levered free cash flow burn remains severe at approximately minus $623 million. Interestingly, the company’s reported net income appears positive partly because of accounting and financing dynamics tied to its IPO structure rather than mature operating efficiency.

More importantly, the cash-flow profile raises a question the market currently seems willing to postpone: infrastructure businesses consuming more than $600 million in annual free cash flow do not usually self-fund expansion for very long. With roughly $209 million in cash on the balance sheet, Cerebras may eventually require either dramatically improved operating leverage, continued access to favourable capital markets, or significantly larger commercial scale to sustain its current trajectory comfortably.

And that nuance becomes critically important once valuation enters the discussion.

At more than 120 times trailing sales and over 123 times enterprise value to revenue, investors are effectively pricing Cerebras as though the inference market structure has already been decided — and decided permanently in its favour.

History suggests technology markets rarely remain that cooperative for long.

Brilliance alone rarely escapes the gravity of industrial scale

My Verdict: Brilliant Technology, Dangerous Timing

I genuinely admire what Cerebras has achieved technologically. The Wafer-Scale Engine is not a gimmick. It addresses real AI bottlenecks in ways competitors cannot easily dismiss.

But I also think the company may have entered public markets at the exact moment enthusiasm around inference economics reached peak intensity.

That creates a dangerous setup where investors are simultaneously underwriting flawless execution, sustained hyperscale demand, expanding margins, stable manufacturing access, and long-term architectural superiority before the competitive landscape has properly stabilised.

Could Cerebras eventually justify today’s valuation? Absolutely.

But at this stage, I suspect investors are no longer paying for possibility. They are paying for inevitability.

History suggests infrastructure transitions create extraordinary winners. It also suggests markets routinely overestimate how long early advantages remain structurally protected.

Cerebras may ultimately become foundational AI infrastructure. Yet at its current valuation, even a brilliant company may discover that being early, admired, and technologically superior is not always enough when the economics of the ecosystem begin asserting themselves.

@TigerStars @Daily_Discussion @Tiger_comments @Tiger_SG @Tiger_Earnings @TigerClub @TigerWire