Alex A.C.
Value/GARP Investor, Semiconductor & Financial Analysis, Electronics MEng
Cerebras's yield problem is not solved, because the fault tolerance it claims comes at a price. Wafer-scale chips need testing, retesting of repairable defects and repair, and long bypass routes around defects can fail higher-speed specifications. @damnang2 lays out the details in a yield model for the WSE-3 built from Cerebras's public disclosures, though by his own account the figures rest on his assumptions, not Cerebras's actual yield. x.com
Cerebras looks increasingly threatened as new chip-packaging advances erode the edge of its wafer-scale design. Applied Materials and Besi are jointly developing die-to-wafer hybrid bonding, which applied to logic would be very bearish for Cerebras, and custom HBM4 could free room for more on-chip SRAM in rival designs. A quoted post also notes that OpenAI's new ultrafast model runs on Nvidia rather than Cerebras, while its COO sold $78 million of stock. x.com x.com x.com
Cerebras's June 2026 earnings report is the latest checkpoint for the bearish case on the stock. x.com
Cerebras is not worth owning because its proprietary stack faces limited broad adoption. The attached analysis says NVIDIA-aligned cloud operators are unlikely to deploy Cerebras hardware, while Groq LPUs fit NVIDIA's software ecosystem more easily. It also flags Cerebras' scaling and deployment friction, despite solid product-market fit with marquee customers. x.com
Cerebras looks attractive because the short report's two-year wafer-life thesis rests on unvalidated reliability assumptions, not representative failure data. Energy per useful token, lower warranty utilization and ordinary repair accounting also undercut its financial conclusions. OpenAI has integrated Cerebras into production; AMD and AWS have announced integrations, while Oracle plans 2027 availability for ultra-fast inference, adding commercial validation despite the report. x.com x.com
Cerebras faces brutal competition from SpaceX, and competing in AI models or data centers requires cash at scale. SpaceX is already trashing Cerebras and CoreWeave. x.com
Cerebras is an unattractive stock to hold now, even though ultrafast speed is a feature companies would happily pay for. Controversy over OpenAI's ultrafast running on Nvidia chips, the share lockup expiration and insider selling outweigh the chance that OpenAI's next 6.1 Sol ultrafast actually runs on Cerebras chips. x.com
Better power delivery appears to be helping Cerebras work through wafer-scale power and thermal limits more than expected. CS-4 gets higher clocks and improved I/O from WSE-3T, despite the same core count and SRAM capacity as WSE-3. Oracle, OpenAI and AWS choosing to work with Cerebras carries weight. But higher clocks do not answer Culper Research's allegation that its wafers fail after about two years; how representative those failures are of current systems remains unclear. x.com
Cerebras' first buyable price, near a $100 billion market cap after its IPO price was raised 50% and the stock doubled at open, buys an uncertain call option on an architecture with hard trade-offs. Its wafer-scale chip gives each core only 48 KB of SRAM, 44 GB in total versus 13.5 TB of HBM3E in Nvidia's GB200 NVL72 rack, and its compiler complexity means it can never match Nvidia's software optimization. Severe memory constraints and limited I/O bandwidth are its key limits. x.com x.com
Cerebras is heavily overvalued at 62.7x EV/sales, even though it can re-rate as HBM stays rationed across the GPU world. Its wafer carries 44GB of SRAM, skipping the HBM queue, the latest quarter grew 74% with a $25.4B backlog, and its customer list should broaden beyond two UAE-linked buyers. Adding makes sense only near $186. If G42 and MBZUAI slow their orders, that multiple has nothing under it. x.com
Culper Research's wafer-failure complaints do not justify shorting Cerebras at $100. The customer complaints sound illogical, although quality issues would not be surprising while the chips are in the early-adopter phase. x.com x.com
Cerebras is too expensive to buy at current levels, even though its technology is fully legitimate. Wafer-scale SRAM could be a game-changer for AI inference, and the hybrid-bonded DRAM planned for the CS-6 (expected in 2029, per the quoted post) addresses its main bottleneck, memory capacity. The company checks many boxes and offers more optionality, but it is not yet a threat to Nvidia. x.com
