The ledger remembers what the algorithm forgets. In the quiet hum of data centers, the most critical battles are not fought over which chip has the most teraflops, but over how to keep that chip fed with clean, stable power. As AI training clusters push single GPU power consumption past 700 watts—and the next generation of NVIDIA’s B200 and Rubin architectures will likely exceed 1,000 watts—the power delivery infrastructure has become a bottleneck. This is not a problem of raw capacity; it is a problem of control, efficiency, and integration. On September 23, 2024, Navitas Semiconductor announced its intention to acquire Claros Technologies for up to $232.8 million. The move is a direct bet on the coming surge in AI server power demand and the shift toward 48-volt architectures. But as with any bet on infrastructure, the real story is in the details of execution, not the headline.
Context: The Players and the Landscape Navitas is a fabless GaN (gallium nitride) power semiconductor company, widely recognized as a top-tier player in the GaN space alongside Power Integrations and EPC. Its core technology—GaN ICs that monolithically integrate driver, control, and power stages—has enabled faster switching, higher efficiency, and smaller form factors in applications ranging from fast chargers to data center power supplies. The company has ridden the wave of AI server demand, but its revenue base remains heavily weighted toward consumer electronics (25-30% of revenue) and the broader data center segment (15-20%). Its AI-specific power supply revenue is estimated at $20-30 million in 2024, a fraction of what is possible.
Claros, on the other hand, is a smaller, privately held company specializing in digital power control. Its core IP includes digital control loop algorithms, firmware, and architectures that allow precise regulation of power delivery. Unlike analog controllers, digital controllers can adapt in real time to changing load conditions, optimize efficiency curves, and communicate with the host system. For AI servers, where workloads fluctuate wildly, digital control is not a luxury—it is a necessity. The acquisition price of up to $232.8 million suggests Claros already has some revenue base (likely $20-40 million annually) and a strong team of engineers with deep expertise in 48V architectures.
The broader market context is critical. The AI server power supply market is projected to grow from roughly $5 billion in 2024 to $15-20 billion by 2028, a compound annual growth rate of over 30%. This growth is driven by the insatiable appetite of AI accelerators. Traditional 12V power architectures, which have served the industry for decades, are hitting their limits. At 1,000 watts per GPU, the current required at 12V is over 80 amps, leading to unacceptable I²R losses and voltage droop. The industry is moving toward a 48V bus architecture, which reduces current by a factor of four and cuts losses by 16 times. This transition requires digital control to manage the tighter voltage tolerances and dynamic load changes. Claros’s technology sits squarely at the center of this transition.
Core: The Technical and Strategic Rationale The acquisition is a classic case of buying adjacency. Navitas has the GaN power stage, but it lacked a sophisticated digital control layer. Historically, the power stage and the controller have been separate chips from different vendors. The controller company (like Texas Instruments or MPS) would design the digital loop, and the GaN company would supply the power FETs. This separation creates inefficiencies: the two parts must communicate over a bus, and the overall solution is larger, more complex, and less optimized than a fully integrated alternative.
By acquiring Claros, Navitas can now offer a single-chip solution that integrates the GaN power stage with the digital controller. This is not just a convenience; it unlocks performance benefits. The digital control loop can be tuned specifically to the characteristics of the GaN FETs, enabling faster transient response, higher efficiency at light loads, and better thermal management. The integration also reduces the number of external components, saving board space and cost—both critical in the dense, power-hungry environment of an AI server.
From a financial perspective, the deal is aggressive. With Navitas’s market capitalization around $1-1.5 billion, the $232.8 million price tag represents roughly 15-20% of its market cap. The company likely used a mix of cash and stock to finance the deal, but the precise structure is not yet public. Based on my experience during the 2024 Spot ETF integration, where I modeled the impact of large capital flows into emerging markets, I see a similar pattern here: a large capital outlay undertaken for a strategic asset that may take 12-18 months to bear fruit. The amortization of the acquired IP, estimated at $30-40 million per year over 5-7 years, will suppress gross margins by 2-3 percentage points initially. To break even on the acquisition, the combined GaN-plus-digital-control products must generate at least $100-150 million in annual revenue. That is a high bar, but the AI power market is growing fast enough to support it.
Technologically, the acquisition positions Navitas to compete directly with the incumbents in the broader power management space: Texas Instruments, MPS, and Infineon. These companies have their own digital control IP and have been investing heavily in GaN. However, Navitas’s GaN IC technology is arguably more advanced in terms of monolithic integration. By adding Claros’s digital control, Navitas can leapfrog the competition in the specific niche of AI server power supplies. The 48V architecture is the key battleground. I noted in my earlier analysis of the 2026 AI-agent economic modeling that the power delivery network for AI clusters is becoming as critical as the compute itself. The acquisition of Claros is a direct acknowledgment of that reality.
Contrarian: The Hidden Risks and the Decoupling Thesis The narrative around the acquisition is overwhelmingly positive: Navitas is acquiring a missing piece, the market is growing, and the integration is natural. But the contrarian view is that the deal may be overvalued and the integration risk is underappreciated. First, the $232.8 million price tag is at the high end of reasonable estimates. Claros’s annual revenue is likely in the $20-40 million range, implying a price-to-sales multiple of 6-10x. While that is not unreasonable for a high-growth tech company, it leaves little room for error. If the integration takes longer than expected, or if the AI power market growth slows, the goodwill could be impaired.
Second, the competitive landscape is not static. Texas Instruments is investing heavily in its own GaN and digital control roadmap. MPS is already a leader in digital power controllers for servers. Infineon has both SiC and GaN portfolios. These companies have far larger R&D budgets and sales forces. Navitas, with its relatively small team, may struggle to win design-ins at the largest cloud service providers (CSPs) like Amazon, Google, and Microsoft. These CSPs are increasingly developing their own power management ICs in-house, posing a threat to external suppliers. The acquisition may be a necessary step, but it is not sufficient.
Third, the 48V architecture transition is still in its early stages. While NVIDIA’s B200 platform is expected to use 48V, the broader ecosystem—including power distribution, connectors, and backup systems—will take years to standardize. The full revenue potential from the 48V shift may not materialize until 2027 or later. In the meantime, Navitas must continue to serve the existing 12V market, where competition is fierce and margins are lower.
Finally, there is the risk of talent retention. The value of Claros lies not just in its IP, but in its engineering team. Digital power control is a specialized field, and the engineers are in high demand. If key personnel leave during the acquisition process, the value of the deal plummets. The acquisition structure likely includes earn-out clauses tied to retention, but those are not guarantees. Trust is borrowed; trust is never owned.
Takeaway: Positioning for the Next Cycle The Navitas-Claros acquisition is a microcosm of the larger infrastructure race that is unfolding beneath the AI hype. The market is focused on the front-end AI chips, but the back-end power delivery is where the real bottlenecks lie. As I wrote in my 2022 post-Terra analysis, “Safety is the only yield that compounds over time.” In the context of power infrastructure, reliability and efficiency are the safety. Navitas has placed a bet that integrating digital control with GaN will create a safer, more efficient power delivery system for AI servers. The next 12 months will tell us whether that bet pays off.
The question is not whether the acquisition is strategically sound—it is. The question is whether Navitas can execute. The ledger remembers what the algorithm forgets. In the end, the market will judge the deal not by the press release, but by the quarterly revenue numbers from the Claros-based products. The cycle is shifting, and the winners will be those who build the infrastructure that enables the next generation of computation. Navitas is building that infrastructure. Now we wait to see if the foundation holds.