Title: 🧠 Cerebras: Demand Is Not the Problem Source: App Economy Insights (Substack newsletter) — "How They Make Money" Premium edition Author: App Economy Insights (Bertrand) Date: 2026-JUN-30 URL: https://www.appeconomyinsights.com/p/cerebras-demand-is-not-the-problem Type: Written newsletter (business-model / earnings breakdown; no video, no timestamps). Saved for personal study via Stephen's logged-in Chrome (PAID post). Note: Key points & figures extracted from the published Premium post (the live post + its charts are the source of truth). Analytical journalism / earnings recap of Cerebras's first quarter as a public company — framing is Neutral (a recap, not a stance call). The "FROM OUR PARTNERS / BluSky AI" block is a paid advertisement (NOT editorial) and is intentionally omitted; subscription/ boilerplate trimmed. Securities referenced: Cerebras (CBRS, IPO May 2026), NVIDIA (NVDA), Amazon/ AWS (AMZN), Microsoft (MSFT), Alphabet/Google (GOOGL), Meta (META), Bell Canada (BCE), TSMC (TSM); private: OpenAI, G42, MBZUAI. Author disclosure: owns AMZN, GOOG, META, MSFT in the App Economy Portfolio. A record semiconductor IPO meets data-center reality. Cerebras pulled off the biggest semiconductor IPO. The May 2026 IPO priced at $185, surged to $386 on opening day, and closed at $311 — a 68% pop. That was the high-water mark. The stock spent the next six weeks giving back most of its gains, and its first earnings report triggered a record two-day plunge that briefly pushed it below the IPO price. The day-one pop is often short-lived, and the first earnings call is where the hype meets reality. The quarter itself was impressive. Revenue nearly doubled, Cerebras beat expectations, and management raised full-year guidance above consensus. The problem was not demand. It was the cost of keeping up with it. Management argues the market misunderstood the softer margin outlook, which is partly due to Cerebras renting back its own chips from a customer while it waits for new data centers to come online. == 🧠 How Cerebras makes money == A typical AI processor is roughly the size of a postage stamp. Cerebras builds one the size of a dinner plate: an entire silicon wafer turned into a single chip. Its Wafer-Scale Engine packs ~4 trillion transistors, against ~200 billion on NVIDIA's flagship Blackwell. Cerebras sells it inside a single machine, the CS-3, rather than as a bare chip. The point of all that silicon is speed. Keeping a model on one giant chip avoids the slow shuffle of data between thousands of smaller ones, which lets Cerebras generate tokens roughly an order of magnitude faster than a GPU cluster. On frontier models, it clears 1,000 tokens per second. The pitch: fast AI is worth a premium because it gets more done. Two revenue streams: - Hardware: Cerebras sells CS-3 systems and clusters to customers who run them in their own data centers. $111 million in Q1 FY26 (+59% Y/Y) — the legacy side of the business. - Cloud & other services: customers rent that same inference horsepower by the token from Cerebras-operated data centers. $83 million (+178% Y/Y), growing three times faster than hardware and the strategic center of gravity going forward. The moat: Cerebras sidesteps the exact chokepoints throttling everyone else. It uses no high-bandwidth memory (HBM), no CoWoS advanced packaging, and no bleeding-edge 3nm process, leaning instead on plentiful SRAM and mature 5nm. While the industry fights over the same scarce TSMC and HBM allocation, Cerebras isn't standing in that line, and it's the only AI accelerator maker building its systems entirely in the US. Customer concentration: Cerebras was, until recently, almost entirely dependent on G42, an Abu Dhabi AI group. Its risk disclosures now name four significant customers: OpenAI, G42, MBZUAI University, and AWS. The bull case rests on the first and last names — turning a UAE-heavy customer base into broader Western enterprise demand. The OpenAI deal shapes the numbers throughout: - The deal: a multi-year agreement for more than $20 billion of Cerebras compute. - The twist: to win it, Cerebras also handed OpenAI warrants (the right to buy its stock at a discount). - The accounting: those warrants count as a discount, so part of what OpenAI pays never shows up as revenue. - The upshot: the reported figure understates the true size of the OpenAI business, and that gap widens as the deal ramps. == 📊 Q1 FY26 by the numbers == Income statement: - Revenue +94% Y/Y to $193 million ($12M beat). - 🔲 Hardware: $111 million (+59%). - ☁️ Cloud & other services: $83 million (+178%). Margin trends: - Gross margin: 45% (+3pp Y/Y). - Operating margin: -8% (+21pp Y/Y). - Net loss narrowed to $14 million, or $(0.22) per share, from $(0.46) a year ago. Cash flow: - Operating cash flow was $12 million (vs an outflow of $55 million a year ago). - Free cash flow was -$120 million as CapEx rose to $132 million. Balance sheet: - Cash, cash equivalents, restricted cash, and short-term investments: $3.3 billion. - Loan from customer (OpenAI working capital loan): $1.0 billion. FY26 guidance (core/adjusted — excludes OpenAI pass-through costs, customer-warrant charges, stock comp): - Revenue $855–865 million (+69% Y/Y at the midpoint). - Gross margin 38–41%. - Operating margin -28% to -32%. Takeaways: - 🚀 Growth was never the question: revenue nearly doubled, loss per share cut in half, adjusted EBITDA flipped positive to $13 million from -$15 million a year ago. A clean beat-and-raise quarter. - 🏗️ The bottleneck is real estate: capex hit $132 million, FCF ran -$120 million, and Cerebras has ~$2.3 billion in future data-center leases not yet on the balance sheet. It can sell compute faster than it can build the buildings to house it. - 📉 Margins are the problem: Q1 core gross margin came in above plan on stronger pricing, but management guided for a sharp step-down as Cerebras temporarily rents capacity back from a customer. The market is not giving it much credit yet. - 💸 OpenAI wears three hats: marquee customer, $1 billion lender, and a warrant holder whose equity is booked as a deduction from revenue. The anchor customer is also helping finance the company it buys from. - 🇦🇪 Concentration remains a problem: two Abu Dhabi entities drove the bulk of 2025 revenue. The backlog is more diversified, but OpenAI is only starting to ramp, and AWS revenue does not begin until 2027. - 🧱 The capital stack is doing heavy lifting: in under a year, Cerebras lined up more than $10 billion in capital — late-stage rounds, an OpenAI loan, a revolver, and the IPO. Even so, the balance sheet still shows a deficit, with common equity sitting beneath billions of preferred stock. Key takeaway: Cerebras has the rarest thing in AI hardware — genuine non-Nvidia demand. The question is no longer whether customers want the product. It's whether Cerebras can build, finance, and supply fast enough to convert that demand before the margin hit and dilution catch up. == 🎙️ Key insights from the call == Andrew Feldman, Co-Founder & CEO - On why demand inflected: "Prior to 2025, AI was a parlor trick, a novelty." His demand thesis: AI only recently crossed from interesting to genuinely useful, and that shift turned compute from a nice-to-have into a bottleneck — a rebuttal to bubble talk dragging on chip stocks. - On why speed is the moat: "Once people start using a technology, speed determines its productivity." He reaches for search and broadband as precedents to argue slow inference loses every market over a long enough horizon, which is why he claims the entire inference market is addressable rather than a "fast" niche. He conceded fast tokens cost more, betting the premium holds because speed is genuinely more valuable. - On the real bottleneck: "Demand is not the constraint. Supply is not the constraint. The constraint is data centers." That reframes Cerebras as an infrastructure-buildout company as much as a chip designer — leasing capacity back from a customer while racing to add data centers across the US, Europe, and a new 120MW Bell Canada site. Bob Komin, CFO - On the pricing power behind the margin beat: "Because there's tremendous demand, we've been able to see higher pricing from existing customers." Q1 core gross margin (47%) came in above plan even as Cerebras warned of a dip ahead; rivals' HBM costs lifted the price floor, so Cerebras is raising prices rather than discounting. The margin guide-down is a capacity-cost problem, not a pricing one. == 🔭 What to watch next == - 📦 Backlog conversion: Cerebras carries roughly $25 billion in remaining performance obligations, but the OpenAI ramp is back-half-2026 loaded, and AWS revenue doesn't arrive until next year. Watch how much warrant-related contra-revenue suppresses reported OpenAI sales, and how quickly hardware revenue falls as production shifts into Cerebras Cloud. The gap between a $25 billion backlog and $865 million of 2026 revenue is the entire investment debate. - 📈 The margin round-trip: management wants investors to look past a near-term trough toward its 60% gross-margin target. Core gross margin is expected to get worse before it gets better — hinging on the rent-back rolling off and owned data centers coming online. A second drag: hardware margin should slide toward the low-30s (Q1's was inflated by one-time pricing). - 🏢 The buildout math: between $132 million of quarterly capex, $2.3 billion in future lease commitments, and an $850 million revolver drawn against a deficit balance sheet, Cerebras is spending aggressively. The question is whether internal cash flow can fund the expansion, or whether another raise (and more dilution) is coming. - 🌍 The concentration test: the cleanest sign of a healthier business is AWS and enterprise revenue showing up large enough to shrink the UAE share of the mix. The wildcard is disaggregated inference — Feldman sees an opportunity to sell Cerebras' decode (the slow one-token-at-a-time half of inference, where its speed advantage is biggest) to companies that already own GPUs. An analyst flagged a Cerebras appearance at Microsoft Build; if that lands, the addressable market opens well beyond today's handful of customers. Author's Note (Bertrand): The views expressed are solely his own and not financial advice. Disclosure: owns AMZN, GOOG, META, and MSFT in the App Economy Portfolio. Ratings (BUY/SELL/HOLD) shared with App Economy Portfolio members.