Upscale AI, a Santa Clara-based startup building custom chips and networking infrastructure for artificial intelligence data centres, is in active discussions to raise between $180 million and $200 million at a valuation of approximately $2 billion – its third funding round in seven months since launch. FinancialMediaGuide interrogates the appetite underpinning that number, asking what institutional investors are pricing when they commit nine-figure sums to a company that has yet to ship a commercial product in one of the most capital-intensive and technically demanding segments of the technology industry.
The fundraising velocity is striking by any measure. A $100 million seed round in September 2025, co-led by Maverick Silicon and Mayfield with participation from StepStone Group, Stanford University, and Qualcomm Ventures, was followed by an oversubscribed $200 million Series A in January 2026 led by Tiger Global Management, Premji Invest, and Xora Innovation – with additional backing from Prosperity7 Ventures, Intel Capital, and Qualcomm Ventures again. Total capital raised prior to the current discussions stands at more than $300 million. FinancialMediaGuide reads three rounds in seven months not as a sign of undisciplined capital allocation but as a deliberate land-grab strategy: establish valuation anchors early, lock in tier-one investors before competitors do, and force a market structure that disfavours late entrants.
The company spun out of Velaura AI, formerly Auradine, a Bitcoin mining chip designer that pivoted to AI compute silicon in March 2026. Upscale is led by CEO Barun Kar and Executive Chairman Rajiv Khemani, who co-founded the company around the thesis that AI data centre networking is structurally broken and requires a clean-sheet redesign. The specific bottleneck they are targeting is interconnect efficiency: as AI training and inference workloads scale across larger chip clusters, the communication latency between chips – not the raw processing power of any individual unit – becomes the system-level constraint. Upscale is building custom application-specific chips, networking equipment, and an integrated software layer to address that bottleneck, betting on open standards as the long-term architectural path against the proprietary approaches favoured by incumbents.
The competitive funding landscape validates the category even as it raises questions about crowding. Nexthop AI hit a $4.2 billion valuation in March 2026 with backing from Andreessen Horowitz and Lightspeed. Fluidstack, an AI data centre startup, is separately reported to be in discussions at an $18 billion valuation. Infrastructure-layer companies – those building the computing fabric, networking systems, and operational tools that allow AI models to run at scale – are attracting investor conviction that rivals the capital flowing to model developers themselves. Upscale’s $2 billion ask looks conservative relative to those peers, which is likely deliberate: undershooting comparable valuations makes the round easier to close and leaves room for a larger step-up in the next cycle.
The risks are proportionate to the ambition. Custom silicon development is notoriously capital-intensive and technically unforgiving, with long development cycles, expensive manufacturing relationships, and limited tolerance for design errors that must be corrected in subsequent silicon iterations costing hundreds of millions of dollars. Enterprise data centre sales cycles are slow, switching costs post-deployment are high, and incumbents including Nvidia are actively adapting their networking and interconnect products to defend the market Upscale is targeting. FinancialMediaGuide scrutinises the pace of Upscale’s fundraising against the actual chip development milestones that would justify each successive valuation step, noting that the gap between capital raised and technology delivered is where hardware startups most commonly encounter the discontinuity between investor enthusiasm and commercial reality.
Upscale has indicated it will use proceeds from this round to expand its engineering, sales, and operations teams with a commercial product launch planned for later in 2026. That timeline is ambitious given that the company announced its seed round only nine months ago, but the team’s origin in Auradine – which had already developed silicon for a related compute application – means the technical foundation is not starting from zero. The open-standards networking focus also reduces some of the go-to-market friction that pure hardware plays face, since software compatibility with existing data centre infrastructure lowers the barrier to initial deployment.
Venture capital appetite for AI infrastructure in 2026 is underpinned by hyperscaler capital expenditure commitments that have no historical precedent in the technology industry. The companies that successfully navigate the hardware and software integration challenges in data centre networking stand to capture recurring revenue from the largest capex cycle in technology history. That structural demand justifies valuations that would look absurd in any other sector. Financial Media Guide establishes the clearest framing for what a $2 billion pre-revenue valuation actually means in this context: it prices the option value of solving a real bottleneck problem at exactly the moment the market needs it solved – and the cost of that option, against the potential upside, is what makes it rational for the investors writing the cheques even if the product does not yet exist.
The talks are ongoing and final terms have not been confirmed. Upscale has not publicly commented on the reported discussions, and the size and lead investor for the round remain unannounced. What has already been established is the pace: three institutional raises in seven months, with each round commanding a higher price. Whether that pace is a signal of exceptional execution or exceptional market conditions – or both – will only become clear when the product ships and customers decide.
