Cerebras Gimlet Labs Deal, September 2026: Infrastructure Expansion Tests the Diversification Case
Cerebras is supplying 100 megawatts of AI chip capacity to cloud startup Gimlet Labs — the latest move in an infrastructure push that spans a 165 MW Finnish data centre and a $25.4 billion backlog still dominated by a single OpenAI agreement.
The announcement that Cerebras Systems will supply 100 megawatts of AI chip capacity to cloud startup Gimlet Labs, with the first data centre due before year-end, is more than a routine supply agreement. For a company listed on the Nasdaq as CBRS and whose name has become synonymous with wafer-scale inference speed, the Gimlet Labs partnership marks a visible turn in strategy: seeding the next generation of AI cloud providers with hardware before incumbent GPU clouds can entrench their positions. The deal arrives against a backdrop of rapid expansion — a 165 MW data centre breaking ground in Mikkeli, Finland in partnership with Compute Nordic, a stated target of 200 MW of European AI compute capacity by the end of 2027, and a 10-year co-location agreement on a Minnesota AI campus. Taken together, these moves sketch the outline of a company trying to build a physical and commercial infrastructure layer that can stand independently of its largest existing customer.
Cerebras has built its commercial identity around a single, defensible claim: its Wafer-Scale Engine architecture delivers token throughput that conventional GPU clusters cannot match at comparable latency. The CS-4 system — a rack-scale configuration of three WSE-3 Turbo processors rated at a claimed 750 petaflops, 30 times the performance of its predecessor — gave that claim measurable form. Its most visible commercial proof arrived when OpenAI deployed Cerebras hardware to power GPT-5.6 Sol's Ultrafast mode, delivering approximately 750 tokens per second, roughly 14 times faster than the standard serving tier according to reports. CEO Rod Stewart has framed the implication explicitly: higher inference speed expands the total addressable market for generative AI workloads rather than simply redistributing existing demand among faster hardware.
The relationship with OpenAI is, however, also the sharpest vulnerability in the company's investment case. Cerebras has reported a $25.4 billion order backlog, with a single agreement — reportedly the OpenAI contract — accounting for the majority of that figure. Revenue surged 88 percent year-over-year, and the company's fast-inference cloud business nearly quadrupled in the second quarter of 2026. Cerebras remains unprofitable nonetheless, and CBRS shares surged 15 percent following the disclosure of the OpenAI compute deal; a separate report noted the stock extending declines overnight despite a reported $20 billion contract figure, with one analyst cautioning that customer concentration risk had not disappeared — it had merely rotated. That observation pinpoints the structural challenge that the Gimlet Labs partnership and subsequent cloud deals are designed, over time, to address.
The diversification effort is unfolding on several fronts simultaneously. A technical partnership with AMD announced in July 2026 pairs the Wafer-Scale Engine with AMD's Helios rack infrastructure for a disaggregated, ultra-low-latency inference platform, validating the wafer-scale architecture within a mainstream CPU ecosystem and opening a second route to enterprise buyers. CrowdStrike has separately deployed Cerebras hardware to power real-time AI threat detection within its Falcon platform, extending the company's reach into cybersecurity workloads. The Mikkeli construction — the initial 50 MW phase now under way — combined with the broader European capacity target represents a substantial commitment to proprietary physical infrastructure.
At Hot Chips 2026, Cerebras disclosed the Nexus system architecture — reported to triple rack-scale performance — alongside a CS-6 roadmap incorporating stacked DRAM. The market context rewards that ambition: Seeking Alpha has identified Cerebras as a key beneficiary of what analysts describe as the inference paradox — token costs across the industry have fallen roughly a thousandfold, yet aggregate AI infrastructure spending has risen because application volumes have grown faster still. Cathie Wood's ARK added approximately $17 million of CBRS stock in August 2026, with Wall Street consensus at that point indicating roughly 59 percent upside from prevailing prices; ARK's buying did not, however, prevent a share-price dip reported despite the Finland data-centre announcement.
The bull case for Cerebras rests on the premise that inference speed is a market-creating force and that no incumbent can match wafer-scale throughput for latency-sensitive workloads at scale. The bear case rests on three concrete concerns: customer concentration (OpenAI still dominates the backlog), path to profitability (revenue growth is rapid but operating margins have not turned positive), and competitive response from GPU cloud peers such as CoreWeave. Three signals are worth monitoring as 2026 closes: whether Gimlet Labs' first data centre comes online on the stated timeline, establishing proof-of-concept for Cerebras as a cloud infrastructure supplier; whether the company can add an anchor customer large enough to reduce OpenAI's visible share of its order backlog; and whether the Mikkeli build and the broader European capacity target produce any disclosed revenue or customer offtake agreements — the terms that would confirm infrastructure investment has translated into durable, diversified demand.