Liquid Silicon: How Morphing Machines’ shape-shifting chips will unfreeze a high-performance computing bottleneck

The name alone should give you a hint. Morphing Machines. Shapeshifters. And in the world of silicon, that is precisely what they are — a Bengaluru-based fabless semiconductor startup building processors that physically reconfigure themselves in real-time to match the demands of a workload. After nearly two decades incubating at the Indian Institute of Science (IISc), Morphing Machines has crossed a critical threshold: a Rs. 80 crore (approximately $11.6 million) Series A round, led by the IAN Alpha Fund, that funds a fully capitalised 24-month roadmap to silicon.

For those of us at Speciale Invest who backed this company in its earlier stages, this raise is more than a funding milestone. It is an industry signal — evidence that the Indian venture market is finally willing to absorb the long gestation cycles of deep-tech hardware in exchange for a category-defining prize: a processor architecture that is domestically designed, strategically sovereign, and technically ahead of its time.

The Architecture: A Polymorphic ASIC

To understand why this matters, you need to appreciate the fundamental trade-off that has defined chip design for decades. Traditional hardware forces engineers to choose between two extremes: the flexibility of FPGAs (Field-Programmable Gate Arrays), which are reconfigurable but power-hungry and relatively slow; or the raw efficiency of ASICs (Application-Specific Integrated Circuits), which are fast and lean but completely fixed once manufactured.

Morphing Machines has spent twenty years building a third way: the Coarse-Grained Reconfigurable Architecture, or CGRA, commercialised under the REDEFINE™ brand. At its core, REDEFINE is a homogeneous grid of compute elements whose interconnects and arithmetic pipelines physically restructure in real-time to match the mathematical shape of any given workload. The result is a processor that delivers ASIC-grade performance — with FPGA-grade adaptability. A polymorphic ASIC, if you will.

“Like a Swiss knife, REDEFINE™ can dynamically switch between CPU and GPU capacity cores, making it ideal for the diverse, demanding and yet unpredictable workloads of modern applications that are increasingly AI-driven.” — Rajnish Kapur, Managing Partner, IAN Alpha Fund

This is achieved through spatial processing: rather than executing a sequence of instructions in the conventional Von Neumann fashion — fetch instruction, decode, execute, repeat — REDEFINE maps an application’s dataflow directly onto the physical fabric of the chip. The instruction-fetch overhead is eliminated entirely. Benchmarks show 3–80x improvements in performance-per-watt over standard processors in specific workloads such as digital beamforming — the kind of compute-intensive signal processing that underpins 5G radios, radar systems, and satellite communications.

The Moment: A $1.3 Trillion Race and Its Discontents

The global semiconductor landscape is currently in the grip of one of the most consequential technology races in history. As AI workloads scale exponentially, hyperscalers — Amazon, Meta, Google, Microsoft — are scrambling for alternatives to Nvidia’s dominant but increasingly costly and power-hungry GPU stack. A single Nvidia GB200 NVL72 rack dissipates over 100 kilowatts: approximately the average annual power consumption of a hundred Indian households, running continuously.

Procurement cycles for Nvidia hardware already stretch beyond a year — made worse by a critical upstream bottleneck: High Bandwidth Memory (HBM), essential to packaging these systems, remains in acute short supply, constraining the entire GPU stack regardless of fab capacity.

The industry’s response has been a proliferation of specialised silicon: custom TPUs, NPUs, and a wave of AI accelerator startups. What almost every one of these designs shares is a fundamental problem — they are static. Once taped out, the architecture is frozen. If the workload evolves — and in an AI-defined world, it always does — you either over-provision (wasteful) or build again (expensive). REDEFINE is designed to resolve exactly this problem.

Its ability to reshape its compute fabric in real-time makes it particularly well-suited for AI inferencing — where irregular, bursty workloads demand adaptability that static silicon simply cannot deliver.

Market signal:  The global semiconductor market is projected to exceed $1.3 trillion by the end of this decade, with AI-specific silicon capturing a growing share of design investment and foundry capacity.

Against this backdrop, India’s Rs. 76,000 crore ($10 billion) semiconductor mission is attempting to re-position the country as an IP originator, not merely a services provider. The Morphing Machines raise is one of the most tangible proofs of that thesis: a domestically-incubated processor architecture, funded by Indian venture capital, targeting global compute markets.

The Dividend of Patient Capital

Venture investing in deep-tech hardware requires a different mental model for time. Software compounds over months. Chips mature over decades. Morphing Machines is an unusually instructive case study in how to preserve intellectual property value through the slow, expensive, and unforgiving process of hardware development.

For nearly seventeen years, founders Dr. Ranjani Narayan and Professor S.K. Nandy funded the company’s development almost entirely through academic grants and sovereign research contracts. The result was a remarkable achievement: two decades of engineering iteration, financed at effectively zero equity dilution. The intellectual property entered the commercial phase intact, unencumbered, and with a substantial head start over any would-be competitor attempting to replicate the architecture from scratch.

When Deepak Shapeti joined as CEO in 2021, the company had the technology. What it needed was commercial discipline — a clear product roadmap, a path to revenue, and a story that could attract growth-stage capital. In the threefive years since, the team has demonstrably delivered on all three counts, culminating in a fully subscribed Series A that funds a concrete engineering milestone: a production-grade proof-of-silicon chip and pilot deployments with cloud providers.

Why Speciale Backed This

At Speciale Invest, our core thesis is that India’s deep-tech edge is built on design talent, intellectual capital, and the ability to originate breakthrough IP — not on manufacturing scale, which the country does not yet possess. Morphing Machines is this thesis in concentrated form.

REDEFINE emerged from IISc, one of Asia’s foremost engineering research institutions, through a two-decade gestation that produced an architecture genuinely different from anything produced in the United States or Taiwan. It is not a derivative of an existing paradigm. It is a new one. That is rare. It is the kind of differentiation that cannot be easily replicated, and it is the kind of moat that makes a hardware company defensible over a multi-decade horizon.

The defence and national security dimensions amplify the investment case. By developing a fully domestic processor architecture, Morphing Machines provides the Indian government with a verified, trustworthy platform for applications that cannot run on foreign silicon — radar signal processing, electronic warfare, secure communications, and critical infrastructure control. The architecture also enables encrypted compute — allowing operations on protected data without ever exposing proprietary data — a capability increasingly central to sovereign and classified workloads. In an era of escalating export controls and supply-chain weaponisation, this is not a niche consideration. It is a central procurement criterion.

Speciale view:  Domestic IP ownership in semiconductor architecture is a 30-year asset. The moat is not just technical — it is institutional, regulatory, and geopolitical.

The Road Ahead: Formidable but Funded

Honesty requires acknowledging that the path from funded research to commercial silicon is not straightforward. Tape-out cycles are expensive and unforgiving. Manufacturing partnerships with foundries — whether TSMC, GlobalFoundries, or domestic alternatives being developed under India’s semiconductor mission — require careful negotiation and technical alignment. And the hyperscaler pilots that represent the commercial prize will demand rigorous performance validation before any volume commitment is made.

None of this is insurmountable. The Series A funds the full 24-month runway needed to reach the proof-of-silicon milestone and initiate those pilot conversations. The architecture’s performance benchmarks are already established in simulation and pre-silicon validation. The founding team has demonstrated the patience and technical rigour that deep-tech hardware demands.

Our view is straightforward: if the engineering continues to match the ambition — and the track record suggests it will — Morphing Machines will do more than produce a chip. It will produce a signal. A signal that India is no longer content to be a consumer of the world’s most strategically important technology. It is now in the business of originating it.

---

Speciale Invest is an early-stage deep-tech fund investing in Indian founders building at the frontier of space, semiconductors, defence, quantum, and advanced materials.

speciale.in // @SpecialeInvest