The SGI revolution, Jensen Huang’s vision, the memory shortage, and the next generation of computing
By Kevin Perelman
NOVEMBER 13, 2026 • TECHNOLOGY PERSPECTIVE
A rare convergence of pressures
NVIDIA faces two consequential developments at once: an AI-memory supply squeeze and the departure of the founding leader who guided its strategy for 33 years. On November 11, 2026, the board accepted Jensen Huang’s resignation as president, chief executive officer, and director; NVIDIA announced the transition on November 12 and appointed longtime CFO Colette Kress interim CEO while it searches for a permanent successor. [1, 2]
At the same time, manufacturers of high-bandwidth memory (HBM) and DRAM are confronting extraordinary AI demand. The resulting tension does not establish that NVIDIA is about to lose its lead. But it creates an opportunity for smaller companies able to deliver computing with less energy, less memory traffic, or an architecture fundamentally different from conventional GPUs.
Jensen Huang: the engineer and founder, not simply a hired CEO
Jensen Huang was not an executive brought in to manage an established corporation. Born in Taiwan in 1963, he moved to the United States as a child, earned electrical engineering degrees from Oregon State University and Stanford University, and worked at AMD and LSI Logic. In 1993 he co-founded NVIDIA with engineers Chris Malachowsky and Curtis Priem and became its first CEO. [3]
He did not personally design every chip or write every software library; NVIDIA’s breakthroughs required co-founders, researchers, engineers, and developers. But Huang supplied an unusual combination of engineering understanding, entrepreneurial ambition, commercial judgment, and long-term conviction. He helped shape a company willing to invest in capabilities whose markets were not yet obvious. That is a very different role from managing an already mature technology business.
NVIDIA’s existing roadmap and technical workforce remain formidable. The question after a founder’s departure is whether future leadership can identify the next market-defining transition—not merely execute investments already committed.
What NVIDIA changed: from Silicon Graphics to mainstream computing
In the 1990s, sophisticated visual effects, animation, scientific visualization, and engineering often depended on specialized systems from Silicon Graphics (SGI) and others. An Indigo2 could be configured in the tens of thousands of dollars—roughly $50,000 in some professional setups—while high-end Onyx installations could cost around $500,000 and, depending on configuration, exceed $1 million. Studio facilities with multiple machines, storage, networking, and rendering infrastructure could run into the millions. These are illustrative historical figures, not uniform list prices.
NVIDIA’s early NV1 and RIVA-era graphics boards did not instantly match every capability of an SGI Onyx. The breakthrough was the economic trajectory. As RIVA, TNT, GeForce, and Quadro generations improved, increasingly sophisticated visual computing migrated to mass-produced PC platforms. NVIDIA was a central contributor to the transformation that undermined the business model of expensive proprietary graphics workstations. Other hardware and software companies also contributed to that shift.
The revolution was never confined to 3D. NVIDIA’s GPUs and associated hardware now accelerate 2D image processing, Photoshop workflows, compositing, photographic enhancement, video effects, video encoding and decoding, color workflows, 3D rendering, simulation, games, science, engineering, and AI. Not all these functions use CUDA: dedicated NVENC and NVDEC hardware, graphics APIs, and other specialized engines also matter.
CUDA, introduced in 2006, nevertheless proved pivotal. It opened GPU parallel computation to a far broader developer ecosystem and helped establish NVIDIA across science, professional media, machine learning, and eventually generative AI. Its dominance is especially consequential in mainstream accelerated AI and many professional production workflows, even though AMD, Intel, Apple, and specialized platforms serve important portions of those markets.
The CUDA advantage—and the vulnerability inside it
NVIDIA does not compete on silicon alone. Its advantage includes CUDA, optimized libraries, frameworks, rendering software, production workflows, deployed systems, and developer expertise. For a visual-effects studio or AI developer, replacing a GPU can mean retesting an entire pipeline, not just swapping hardware. That network of software and practice is a considerable barrier to entry.
But incumbency creates a dilemma. A company built around a profitable platform can become reluctant to support technologies that might make that platform less central. A new competitor might not beat NVIDIA at conventional graphics at all. It might first establish a decisive price, energy, or latency advantage on one valuable workload, then expand as its software ecosystem develops.
The memory crisis: why it changes the competitive equation
Large AI systems demand not just processing power but enormous memory capacity and bandwidth. Model weights, KV caches, and intermediate values must be stored and moved. For many inference workloads, repeatedly moving data consumes as much strategic attention as adding more arithmetic capability. Enlarging HBM capacity, adding GPU systems, and building more memory factories are obvious responses—but not the only ones.
In its October 28, 2026 earnings call, SK hynix said that volumes secured by long-term agreements together with specific customer negotiations meant its DRAM and HBM output was effectively committed through 2029. The company clarified that signed long-term agreements represented more than half of expected capacity in 2027 and 2028; the broader “sold out” characterization also included advanced purchase discussions, not solely binding contracts. [4, 5]
This distinction matters. The announcement is evidence of unusually strong demand visibility, not a guarantee that every memory product will remain unavailable for three years. Nonetheless, constrained supply and elevated costs strengthen the case for approaches that reduce how frequently data must cross a memory bus.
A future photonic or neuromorphic system would still need to represent model parameters somehow. Its potential advantage would be keeping state closer to computation, exploiting optical parallelism, or processing sparse events instead of constantly transferring huge arrays of weights. That could improve the economics of selected AI workloads without eliminating memory outright.
Where smaller companies could break through
Photonic computing. Q.ANT is developing a photonic Native Processing Unit that performs selected mathematical operations with light. In June 2026 the company reported running a diffusion-model image-to-image demonstration and a recurrent neural network on its second-generation photonic hardware. This is a meaningful demonstration of workloads beyond toy arithmetic, although it is company-reported and not evidence that the platform replaces CUDA across general-purpose graphics or AI. [6]
Neuromorphic computing. Intel’s Loihi 2 research, including Hala Point, explores distributed, event-driven spiking neural computation. Companies such as BrainChip and Innatera are pursuing specialized low-power neural processing for sensors and other embedded applications. The strongest use cases today differ from rendering a Hollywood production or training a frontier-size transformer. [7]
Photonic-neuromorphic and memory-centric architectures. Experimental systems combine optical pathways with neural-style dynamics, local state, and adaptation. They could eventually reduce costly data movement or enable different types of intelligent processing. They remain less mature than current GPU platforms, and no broadly available hybrid system has demonstrated general-purpose replacement of a CUDA workstation.
The competitive opportunity does not require an immediate all-purpose replacement. A startup might win AI inference at the edge, a particular scientific calculation, low-power sensing, or another specialized task. If its advantage proves durable, programmers can build tools around it, manufacturers can scale production, customers can move more workloads onto it, and the market can widen. That is how disruption can begin at the margins of an incumbent’s business.
Why the SGI Lesson Matters Again
Silicon Graphics possessed extraordinary engineering talent, accomplished customers, advanced technology, and an established position in professional computing. Yet those advantages ultimately failed to protect it from a technological transformation driven by smaller companies developing dramatically more affordable computing hardware.
NVIDIA was one of the companies that helped bring about that transformation.
SGI wasn’t displaced overnight by a single inexpensive RIVA graphics card. Rather, successive generations of increasingly powerful PC graphics hardware gradually undermined the economics of its specialized workstation business. Technologies once requiring hundreds of thousands or even millions of dollars became accessible through comparatively inexpensive computers.
Today, NVIDIA faces the possibility of becoming the very company it once helped displace.
For more than three decades, Jensen Huang helped guide NVIDIA through technological changes that repeatedly expanded its markets. Under his leadership, NVIDIA evolved from a graphics-chip startup into a dominant platform spanning professional graphics, Hollywood visual effects, video processing, scientific computing, and artificial intelligence.
But NVIDIA’s position is now potentially weakened by two major developments occurring at the same time: the growing global shortage of high-performance memory and the departure of the founder who helped establish its technological direction.
These are not minor concerns. They strike at both NVIDIA’s existing computing architecture and the leadership responsible for anticipating what comes next.
A Memory Crisis and a Leadership Vacuum at the Worst Possible Time
Modern NVIDIA AI systems depend heavily on increasingly large quantities of high-bandwidth memory. As artificial intelligence models expand, the demand for memory capacity, bandwidth, electricity, and cooling continues to place enormous pressure on the industry.
The current memory shortage exposes a fundamental weakness in the economics of conventional GPU-based AI: enormous quantities of information must be stored, moved, and repeatedly processed.
NVIDIA can respond by engineering faster processors, integrating additional memory, improving bandwidth, and increasing manufacturing capacity. But those approaches largely continue along the technological path the company has followed for decades.
What happens if the better solution is to abandon significant portions of that architecture altogether?
This is precisely where Jensen Huang’s departure becomes so consequential.
Huang wasn’t simply a corporate administrator overseeing quarterly earnings. He was a founding engineer and entrepreneur who helped envision technological opportunities long before they became established markets. His leadership was central to NVIDIA’s expansion from graphics acceleration into general-purpose GPU computing and AI infrastructure.
Now, at a moment when the industry may need another fundamental architectural transformation, NVIDIA is undergoing a major leadership transition.
Its engineering teams can continue developing planned products. Its suppliers can continue fulfilling agreements. Its software developers can continue improving CUDA.
But maintaining those operations is not the same as having the visionary leadership necessary to identify the next computing revolution.
NVIDIA may find itself with tremendous resources, an extraordinary engineering organization, and a long-term roadmap, yet without the same leadership that helped determine where that roadmap should lead.
That does not mean the company is incapable of innovation without Huang. It means the leadership transition introduces substantial strategic uncertainty precisely when the industry faces pressures that could reward fundamentally different approaches to computing.
The danger is not necessarily an immediate collapse. It is that a company accustomed to leading technological change could become increasingly focused on protecting and refining its existing architecture while smaller innovators move in an entirely new direction.
The Door Opens for Smaller Companies to Challenge NVIDIA
This combination of circumstances could create an extraordinary opportunity for emerging technology companies.
Photonic computing, neuromorphic processing, processing-in-memory architectures, and experimental photonic-neuromorphic systems are investigating alternatives to the conventional separation of processing and memory.
Rather than continually increasing GPU performance and memory capacity, these technologies could eventually perform certain computations with dramatically less data movement, lower energy consumption, and fundamentally different processing structures.
Companies such as Q.ANT, along with neuromorphic developers and other emerging research organizations, are exploring technologies that could challenge the assumptions underlying today’s GPU-dominated computing model.
Their current products are not complete replacements for NVIDIA’s graphics or AI platforms. Many remain specialized or experimental. But that is exactly why the historical comparison with Silicon Graphics matters.
The technology that eventually displaces an established market leader rarely begins as a complete replacement for everything that leader produces.
It often begins as a smaller, less expensive, more efficient solution to a particular problem.
If one of these emerging companies develops a photonic or neuromorphic computing architecture that performs an important class of AI workloads at a fraction of the memory, electricity, or hardware cost required by conventional GPUs, the consequences could be enormous.
Developers would begin adapting software. Manufacturers would invest in production. Customers would begin evaluating entirely new computing platforms. Other companies would expand the technology into additional applications.
Eventually, those innovations could spread beyond their original specialized markets.
What begins as an alternative AI processor could become the foundation of an entirely different generation of computing.
And NVIDIA, despite its enormous advantages, could find itself competing against technologies that make significant portions of its existing architecture less economical.
This is not a prediction that photonic or neuromorphic processors will inevitably replace NVIDIA. It is a recognition that the opportunity for such disruption could become substantial if these technologies fulfill their technical promises.
The memory crisis only strengthens that opportunity by making efficiency, memory integration, and alternative computational approaches increasingly valuable.
Retool, Reinvent, or Risk Falling Behind
NVIDIA has the financial resources and engineering talent to respond to these emerging technologies.
It could develop its own photonic processors, acquire smaller innovators, integrate neuromorphic computing into future systems, or design entirely new architectures that move beyond the traditional CPU, GPU, and external-memory model.
But doing so may require more than incremental product development.
NVIDIA may eventually need to challenge the very technologies and business models that made it successful.
That is a difficult decision for any established company.
CUDA, NVIDIA’s extensive software libraries, its developer community, and its relationships with major technology companies provide enormous competitive advantages. But they also create powerful incentives to continue investing in the architecture that supports those advantages.
The risk is that NVIDIA becomes so committed to its existing ecosystem that it underestimates a smaller competitor offering a fundamentally different approach.
This was part of the historical challenge faced by established workstation manufacturers as commodity graphics technology improved.
A new technology does not need to outperform NVIDIA at everything immediately. It needs to become sufficiently superior in an important market that customers have a compelling economic reason to adopt it.
If that happens, today’s CUDA advantage may become less decisive as developers build software for the emerging architecture.
NVIDIA would then face a choice: reinvent significant parts of its computing platform or risk surrendering technological leadership in those markets.
The Question That Defines NVIDIA’s Next Era
For 33 years, NVIDIA developed under the leadership of one of the computing industry’s most influential technology entrepreneurs.
Jensen Huang helped establish the company that transformed professional graphics, challenged the economics of Silicon Graphics workstations, and expanded GPU computing far beyond 3D rendering into image processing, video production, scientific computing, and artificial intelligence.
Now, at a time when the computing industry is facing enormous memory constraints and potentially revolutionary alternative processor technologies, NVIDIA is confronting the challenge of continuing without its founding visionary at the helm.
The concern is not whether NVIDIA can survive without Jensen Huang. It almost certainly has the resources and established business to continue operating and competing. The concern is whether it can remain the technological leader without the person who repeatedly helped anticipate where computing was heading.
A company can have billions of dollars, thousands of talented engineers, a sophisticated manufacturing network, and decades of planned products—and still miss the next major technological transformation.
Meanwhile, a small company developing technology that initially appears experimental or commercially insignificant could discover a fundamentally superior approach.
That is the opportunity being created.
The memory shortage is exposing weaknesses in the existing computing model. Emerging technologies are offering possible alternatives. And NVIDIA is facing a leadership transition at precisely the moment when recognizing those alternatives may become increasingly important.
NVIDIA may be entering a period in which its greatest vulnerability is not a shortage of financial resources, engineering talent, or customers, but uncertainty over who will provide the next generation of technological vision.
The company that once helped make multimillion-dollar computing systems economically obsolete could eventually find itself challenged by a much smaller innovator armed with a technology that changes computing all over again.
And that would be the ultimate irony.
NVIDIA revolutionized computing by helping replace the technologies that came before it. Its greatest challenge now may be preventing another company from doing exactly the same thing to NVIDIA.
Sources and perspective
[1] NVIDIA Newsroom, “NVIDIA Announces CEO Transition,” November 12, 2026. https://nvidianews.nvidia.com/news/nvidia-announces-ceo-transition/
[2] NVIDIA, SEC Form 8-K concerning Jensen Huang’s resignation, November 12, 2026. https://www.sec.gov/Archives/edgar/data/1045810/000104581026000082/nvda-20261111.htm
[3] NVIDIA Newsroom, “Jensen Huang,” founder biography. https://nvidianews.nvidia.com/bios/jensen-huang
[4] Reuters, “Chipmaker SK Hynix to return $75 billion to shareholders after record profit,” October 27/28, 2026. https://www.reuters.com/world/asia-pacific/sk-hynix-q3-profit-jumps-more-than-7-fold-beats-forecasts-2026-10-27/
[5] SK hynix Q3 2026 earnings transcript, October 28, 2026. https://www.marketbeat.com/earnings/reports/2026-10-28-sk-hynix-inc-stock/
[6] Q.ANT, “Q.ANT Runs Generative AI on Photonic Hardware,” June 23, 2026. https://qant.com/press-releases/q-ant-runs-generative-ai-on-photonic-hardware/
[7] Intel, “Intel Builds World’s Largest Neuromorphic System,” April 17, 2024. https://www.intel.com/content/www/us/en/newsroom/news/intel-builds-worlds-largest-neuromorphic-system.html
Perspective note: The industry opportunities, long-term leadership concerns, and disruption scenarios are analysis, not predictions. SGI pricing is illustrative and dependent on model, date, and configuration. Emerging technology demonstrations do not establish commercial superiority over general-purpose NVIDIA systems.