The Networking Bottleneck Gets Its Biggest Bet Yet

The AI infrastructure boom has a new flashpoint — and it's not GPUs. On March 10, 2026, Nexthop AI closed an oversubscribed $500 million Series B round at a $4.2 billion post-money valuation, signaling that the investment community has firmly identified networking as the next critical frontier in AI data center buildouts.

The round was led by Lightspeed Venture Partners, with Andreessen Horowitz (a16z) and Altimeter Capital joining as major participants alongside all existing investors. For a company that launched just a year ago with $110 million, the velocity of this raise — and the caliber of investors lining up — tells a compelling story about just how urgent the AI networking problem has become.

From Arista's COO to AI Networking's Most-Funded Founder

Nexthop AI was founded by Anshul Sadana, a networking industry veteran with a pedigree few can match. Sadana spent 17 years as Chief Operating Officer at Arista Networks, helping build the company into a data center networking powerhouse now on track to surpass $10 billion in annual revenue. Before Arista, he spent eight years at Cisco.

Sadana saw the AI wave coming and identified a fundamental gap in the market. "AI is causing a massive disruption at the infrastructure level," he explained when launching the company. "We are no longer just building a separate pizza box — it's an integrated solution that gets to the customer the way they want it."

Founded in 2024, Nexthop AI is headquartered in Santa Clara, California, with offices in Seattle, Vancouver, Dublin, and Bengaluru. The team currently numbers approximately 100 employees and is scaling rapidly with this new capital infusion.

Inside the Round: $610M Total Raised in 12 Months

Nexthop AI's funding trajectory has been nothing short of extraordinary:

• March 2025: $110M seed/Series A — Lightspeed Venture Partners, Kleiner Perkins, WestBridge Capital, Battery Ventures, Emergent Ventures
• March 2026: $500M Series B — Led by Lightspeed Venture Partners, with a16z, Altimeter Capital, and all existing investors

Total funding to date stands at approximately $610 million. The round was oversubscribed, meaning investor demand exceeded the capital the company was willing to take — a strong signal of market conviction.

The timing aligns with Lightspeed's own aggressive AI infrastructure thesis. The firm closed over $9 billion in new funds in December 2025 and has already deployed more than $5.5 billion across 165+ AI-native startups. Guru Chahal of Lightspeed called Nexthop a potential "$100B+ company" and described AI as "forcing a fundamental rethink of data center network architecture."

Andreessen Horowitz, which raised $15 billion across five new funds in January 2026, published a dedicated investment memo framing networking as a generational opportunity. "Every major platform shift in computing has produced a new networking giant," the firm wrote. "Today, as hyperscalers rebuild their infrastructure around AI, networking is once again up for grabs."

The Product: Purpose-Built Hardware for AI Traffic

Nexthop AI's core thesis is deceptively simple: existing networking equipment was designed for enterprise and cloud workloads, not for the unique demands of AI training and inference. AI clusters require thousands of GPUs communicating simultaneously, exchanging gradients, synchronizing weights, and moving data at extraordinary speeds. A single dropped packet or microsecond of latency can cascade through an entire training job.

Alongside the Series B announcement, Nexthop unveiled three new switch platforms:

• NH-4010: 51.2 terabits per second throughput; up to 20% more power-efficient than competitors, potentially saving dozens of megawatts per deployment
• NH-4220: Double the capacity of the NH-4010
• NH-5010: Designed for disaggregated spine architecture, separating internal data center traffic from inter-facility packet orchestration

All switches are built on Broadcom networking silicon and support 1.6 terabits per second per port — performance that was once reserved for massive telecom core routers. Key technical capabilities include:

• RoCEv2 (RDMA over Converged Ethernet v2): Enables direct GPU-to-GPU communication bypassing CPUs entirely
• DCQCN protocol: Automatically detects and resolves network congestion
• Linear LPOs/LROs: Optical modules that reduce costs by minimizing digital signal processors
• Nexthop NOS: A custom network operating system built on Microsoft's open-source SONiC platform with enhanced cybersecurity optimizations

This open-source-native approach is central to Nexthop's strategy. As a16z noted in its investment thesis, the company "inverts the traditional vendor model — engineering world-class hardware around open-source from the start" rather than retrofitting proprietary systems.

Market Context: A $100 Billion Opportunity Taking Shape

The AI data center networking market is experiencing explosive growth. Current estimates put the market at $10.3 billion in 2025, growing to $12.8 billion in 2026 (24.2% CAGR). But the longer-term projections are what's truly driving investor excitement: SemiAnalysis forecasts the market reaching $100 billion by 2031.

The growth drivers are structural and accelerating:

• Hyperscaler capex: Alphabet, Amazon, Meta, and Microsoft are expected to spend approximately $650 billion on AI data centers in 2026 alone
• GPU cluster scaling: As clusters expand to hundreds of thousands of processors, networking becomes the binding constraint on system performance
• Hardware cycle acceleration: The industry is rapidly transitioning from 400G to 800G to 1.6T switches, compressing deployment cycles

The competitive landscape is in flux. Nvidia has emerged as the surprising market leader in backend AI networking with 25.9% share ($2.3B in quarterly sales, up 647% YoY) through its Spectrum-X Ethernet technology. Arista Networks continues to ride the AI wave toward $10B+ in annual revenue. Cisco secured over $2 billion in AI infrastructure orders in FY2025. Meanwhile, the Ultra Ethernet Consortium released its UEC Specification 1.0, establishing new standards for AI and HPC networking.

Nexthop differentiates by targeting hyperscalers and NeoClouds exclusively with a JDM (Joint Development Manufacturing) model — custom solutions co-developed with customers rather than off-the-shelf products. The company claims it can compress customer development cycles by 6-12 months and help evaluate 4-6 technological alternatives versus the 1-2 options customers can explore internally.

Strategic Implications: Where the Money Goes

The $500 million will be deployed across three primary vectors:

Product development acceleration: With 400G-to-800G-to-1.6T transitions happening faster than ever, Nexthop needs to stay ahead of the hardware curve. The three new switches announced alongside the funding are just the beginning.

Team scaling: From approximately 100 employees today, Nexthop plans significant hiring across hardware engineering, software development, and customer-facing roles. The company's five-office global footprint suggests an aggressive international talent strategy.

Manufacturing and supply chain: Building custom hardware at hyperscaler scale requires deep investment in production capacity and component supply chain management — especially as supply constraints in chips, memory, and optical components remain a persistent industry challenge.

Why Investors Are Betting Big

The investment thesis converges on a few key points that make Nexthop AI compelling despite competing against entrenched giants:

Founder-market fit: Anshul Sadana doesn't just understand networking — he helped build the company that currently dominates data center switching. His 17 years at Arista and deep hyperscaler relationships give Nexthop instant credibility that no other startup in this space can claim.

Architectural timing: AI workloads are fundamentally different from the cloud and enterprise traffic that existing switches were optimized for. This creates a genuine platform-shift moment where purpose-built solutions have a structural advantage.

Open-source leverage: By building on SONiC and embracing open-source from day one, Nexthop aligns with the procurement preferences of hyperscalers who increasingly reject vendor lock-in. This is the opposite approach from incumbents who are trying to adapt proprietary stacks.

Power efficiency: With energy infrastructure emerging as the primary bottleneck constraining AI data center expansion, Nexthop's claim of 20% better power efficiency is not a nice-to-have — it's a competitive weapon. At hyperscaler scale, that translates to millions of dollars in operational savings.

Raghu Raghuram of a16z put it directly: "Networking is the bottleneck between GPU capacity and GPU output." In a market where the biggest technology companies on Earth are spending hundreds of billions on AI infrastructure, solving that bottleneck is worth a very large bet.

What to Watch

Nexthop AI's $500M raise marks a watershed moment for AI networking infrastructure. The company faces formidable competitors in Nvidia, Arista, and Cisco — all of which have massive scale advantages and existing customer relationships. But with a proven founder, purpose-built technology, top-tier investor backing, and a market that analysts project could reach $100 billion within five years, Nexthop has assembled the ingredients for a genuine challenge to the incumbents. The key milestones to watch: customer wins with major hyperscalers, revenue trajectory toward a potential IPO path, and whether the company's power efficiency claims hold up at production scale. In the AI infrastructure race, the networking layer has officially become a main event.