Last week’s Optical Fiber Communications conference (OFC) in Los Angeles had a 2001 kind of energy, and old-timers will know that means the atmosphere was giddy with anticipation and propelled by purpose. The energy was palpable. And the impact of the AI era on photonic technologies brightened every tech session and leavened every aisle.

Data-center interconnects continue to dominate, but OFC will be about more next year.

Synchronicity

It struck me as fitting that the conference came just after the OpenClaw phenomenon burst onto the scene, offering flexible agent software that runs locally to access large language models (LLMs) and accomplish complex tasks autonomously. Importantly, OpenClaw and many other emerging tools can work with cloud-based LLMs—the chatbots most readers know by now—and secure and private locally-running LLMs, which are by now almost useful and are advancing at a blistering pace. In fact, much of the OpenClaw buzz regarded those locally-deployed LLMs, and Apple has the sold-out Mac Minis to show for it.

So it was with a burst and a bang that OpenClaw commenced a careening adoption-and-hype cycle a few weeks ago, addressing security and usability issues with rapid-fire revisions and culminating in its creator’s acquisition by OpenAI after a matter of just days. Even by Silicon Valley’s turbocharged standards, the agent phenomenon has been riveting to watch.

It also illustrated something fundamental but underplayed about the AI phenomenon: there is broad, unmet demand for local, focused, task-oriented AI functionality. Local as in running on your machine or a server on your office LAN. Local as in “at the edge.”

Innovation at the Edge

I expect AI-at-the-edge will eclipse cloud AI in relevance and impact very soon.

See, cloud AI gets all the press right now, with its immense compute edifices funded by controversially circular investment and enabled by photonic interconnects of steadily-diminishing physical scale. But currently the cloud AI players are—let’s face it—seemingly interchangeable, with each week’s jostling benchmarks getting most of the press. That’s a key reason why events such as DeepSeek’s introduction last year (promising LLM implementation on affordable hardware) can instantly whip up a howling maelstrom of hype. It’s why OpenClaw is so consequential. It’s a field starved of differentiation, a field ready to serve beyond its base of experts.

Avoiding the Commodity Trap

To my eye, the AI market is already looking less like that of commodity technologies such as printers or modem chips and more like the motion control industry, where the future is in task-specific implementations. Generic motion control is looking anodyne and futureless, but purveyors who focus on solutions specific to key tasks (like aligning photonic devices) have a bright tomorrow.

We’re starting to see hints of something similar emerging in the cloud AI space. If I’m right, a whirl about AI tools that do a job will soon eclipse the chatter about AI back-end infrastructure in public discussion and strategic importance. Early examples abound:

  • Tesla’s cars have become robots on wheels, capable of autonomously taking you from Point A to Point B on voice command. This functionality is a local AI implementation built on in-house-designed chips whose offline capabilities are provisioned by vast central compute resources, with all complexity hidden from the user;
  • Palantir has evolved from its roots in fraud detection for PayPal to become the preeminent purveyor of AI-based functionality tailored to security and defense;
  • Anthropic recently introduced focused tools for code security auditors and corporate legal departments.
  • Taalas has introduced chip-level LLM implementations under the wonderfully resonant banner of “The Model is The Computer,” bringing down the cost of local, task-optimized, industrial-class LLMs by orders of magnitude while increasing performance by 75X versus Nvidia’s H200.

Expect more of this. We are a tool-using species, and this trend reflects that. The fuse is lit.

From Tide to Tsunami

For photonics, this presses even more on the tide of ever-diminishing physical scale for interconnects. Copper-based connectivity between RAM and compute resources has already emerged as the top bottleneck for latency and energy consumption. Architectures that replace copper (with its capacitance, EMI, thermal, radiative and other complications) with chip- and package-scale photonic interconnects and interposers will become crucial. And test- and assembly-automation technologies that enable production economics will see an even more intense spotlight.

I expect we’ll be talking about a lot of that at OFCs 2027 and 2028.

OFC 2026 sparkled with opportunity. Unlike 2001, the industry will not run off a cliff— there are just too many applications pulling the field forward, contrary to 2001’s solitary focus on long-distance telephony that ultimately served as a single point of failure for the industry.

My perception is that the topological map of opportunities is shifting as AI market tectonics give rise to application focus and edge implementations far from the data center. This will accelerate a generation of chip architectures with photonic channels between memory and compute, and ultimately photonic compute itself.

 


The Dogfood Ate my Thesis

My main essay is above, but writing it got me wondering how close agent and local LLM techology really is to being practically useable by ordinary carbon-based life-forms. A quick review spotlighted an interesting autonomous assistant: Khoj, tailored to leverage LLMs to perform autonomous research using the web and any folders of background resources the user might provide. With a bit of geekery, Khoj can utilize private and secure local models instead of the popular cloud AI services. Now, one thing Aikidovation does is use LLM technologies locally for confidentiality. So in the spirit of eating my own dogfood, I set up Khoj and connected it to a local incidence of gemma3:27b, a serviceable, year-old entry-level Google DeepMind large language model freely available for download and well within the capabilities of my M4 Max/64GB MacBook Pro.

Since photonics production scaling has been my personal topic for several years and is now the urgent mainstream theme for the industry, I asked it,

“Write a 1000 word article on rapid volume scaling, citing Andy Grove and others who have contributed to hypergrowth in highly technical fields.”

That’s all the guidance it got.

And within 20 minutes, it was done. Again, this was entirely local, entirely at the edge: a confidential agent that would accomplish a task securely and privately in minutes at zero cost.

And the resulting essay is …not bad! For a first draft in particular, not bad at all. Nicely formatted, intriguing examples, good texture, varied resources, authoritative voice.

That puny antique gemma model wrote better than most interns. With citations.

But then new fun began. Perhaps since OpenClaw endured so many security controversies, Khoj is pretty locked-down. It only interacts with your file system via read-only folders specifically provisioned and managed by its accompanying desktop utility. (Even so, I confined it to its own user account.) Getting it to save its essay as a file I could work with took a whole advanced chapter of geekery that soon revealed the limits of that gemma model. But models evolve rapidly, and technologies that fit big models into small computers have emerged. And rough spots and usability challenges will ease with time. In fact, that may have already happened, as I note the Khoj folks now offer Pipali, which appears to be a fully wrapped Khoj-style assistant non-geeks can use.

The future is edgy.


The Relentless Pursuit of Scale: Lessons from Andy Grove and the Architects of Hypergrowth

Guest Essay by Aikidovation’s private AI assistant

Creative robot

The ability to rapidly scale volume – to meet exponentially increasing demand without sacrificing quality or efficiency – is the defining characteristic of hypergrowth companies in today’s technical landscape. While innovation often ignites the initial spark, it is scale that transforms a promising startup into a dominant force. This article explores the principles of rapid volume scaling, drawing heavily on the wisdom of Andy Grove, Intel’s legendary CEO, and examining how other leaders have navigated the complexities of hypergrowth in technically demanding fields.

Andy Grove and the Importance of the “Scale-Up” Phase

Andy Grove, in his seminal work High Output Management and subsequent writings 1, consistently emphasized that the true challenge isn’t creating something new, but rather taking that creation from prototype to mass production. He argued that startups are valuable, but the real economic impact – and job creation – comes from companies that can successfully scale 2. This “scale-up” phase, as Grove termed it, requires a fundamentally different skillset and organizational structure than the initial startup hustle.

Grove’s perspective wasn’t simply about logistical efficiency. He highlighted the critical role of an “effective ecosystem” where knowledge accumulates, experience builds upon itself, and strong relationships develop between suppliers and customers 3. This learning-by-doing process is the engine of sustainable scale. He recognized that scale isn’t just about doing more of the same; it’s about becoming better at doing it, continuously improving processes and adapting to changing conditions.

Beyond Innovation: Building for Scale

Many companies stumble not because they lack innovative ideas, but because they fail to build the infrastructure to support rapid growth. As Marc Andreessen points out, simply having a great product isn’t enough; you need the systems and processes to deliver it to a massive audience 4. This requires a proactive approach to anticipating bottlenecks and building redundancy into the system.

Several key themes emerge when examining successful scaling strategies:

  • Standardization and Automation: Grove stressed the importance of high output management, which inherently requires standardized processes. Automating repetitive tasks frees up valuable resources for more complex problem-solving and innovation 5.
  • Team Structure and Delegation: As companies grow, the traditional hierarchical structure can become a liability. Grove advocated for a “team of teams” approach, where smaller, autonomous units collaborate effectively 5. This necessitates effective delegation and clear lines of accountability.
  • Data-Driven Decision Making: Modern scaling relies heavily on data analytics. Companies must track key performance indicators (KPIs) and use data to identify areas for improvement and optimize performance 6. The OKR (Objectives and Key Results) framework, pioneered by Andy Grove at Intel, provides a powerful tool for aligning teams and measuring progress 7.
  • Proactive Risk Management: Grove’s “Only the Paranoid Survive” philosophy 8 emphasizes the importance of constantly scanning the environment for potential threats and proactively addressing them. This requires a culture of constructive confrontation, where challenging assumptions is encouraged 9.

Modern Scaling Strategies: Beyond the Factory Floor

While Grove’s principles originated in the context of hardware manufacturing, they remain remarkably relevant to modern software and data-driven businesses. However, the mechanisms of scaling have evolved. Today, scaling often involves:

  • Cloud Infrastructure: The advent of cloud computing has dramatically lowered the barrier to entry for scaling. Companies can rapidly provision resources on demand, eliminating the need for massive upfront capital investment in hardware. This allows for elastic scaling, adjusting capacity to meet fluctuating demand.
  • Microservices Architecture: Breaking down monolithic applications into smaller, independent microservices allows for greater flexibility and scalability. Each microservice can be scaled independently, optimizing resource utilization and improving resilience 10.
  • DevOps and Continuous Delivery: Automating the software development lifecycle – from coding to testing to deployment – enables faster iteration and more frequent releases. This allows companies to respond quickly to changing market conditions and deliver value to customers more rapidly.
  • Platform Engineering: Building internal platforms that abstract away the complexities of underlying infrastructure empowers development teams to self-serve and accelerate innovation. This reduces bottlenecks and enables faster scaling.

Case Studies in Hypergrowth

Several companies exemplify successful rapid volume scaling in technical fields.

  • Netflix: Netflix’s transition from DVD rentals to streaming required a massive investment in content delivery networks (CDNs) and cloud infrastructure. They leveraged data analytics to personalize recommendations and optimize streaming quality, enabling them to scale to over 230 million subscribers worldwide.
  • Slack: Slack’s rapid growth was fueled by its focus on user experience and seamless collaboration. They utilized a microservices architecture and automated testing to ensure reliability and scalability as their user base exploded.
  • Zoom: The COVID-19 pandemic dramatically accelerated the demand for video conferencing. Zoom was able to scale its infrastructure to accommodate millions of concurrent users by leveraging cloud services and optimizing its algorithms for low latency.

The LLM Paradox: Scaling Potential and Inherent Limitations

Interestingly, the very technology I am using to generate this article – large language models (LLMs) – presents a unique scaling paradox. While LLMs offer immense potential for automating content creation and accelerating knowledge work 11, they also exhibit inherent limitations. As highlighted in recent research 12, LLMs can generate biased or incorrect information, require significant computational resources, and struggle with complex reasoning tasks. Scaling LLM-powered applications requires careful attention to these limitations and the implementation of robust safeguards. Furthermore, the “model oversizing” issue 13 needs to be addressed to prevent unsustainable growth in model size and complexity.

The Human Factor: Culture and Leadership in a Scaling Organization

Technology is merely an enabler of scale. The true engine of hypergrowth lies in the people and culture of the organization. Andy Grove understood this implicitly, emphasizing the importance of clear communication, constructive confrontation, and a relentless focus on results.

  • Culture of Ownership: Scaling requires empowering employees at all levels to take ownership of their work and make decisions independently. This necessitates a culture of trust and accountability.
  • Bias for Action: In a rapidly changing environment, speed is critical. Companies must encourage experimentation and reward those who are willing to take calculated risks.
  • Continuous Learning: Scaling is a continuous learning process. Organizations must invest in training and development to ensure that their employees have the skills and knowledge they need to succeed.
  • Strong Leadership: Leaders must be able to articulate a clear vision, inspire their teams, and navigate the inevitable challenges that arise during a period of rapid growth.

Avoiding the Pitfalls of Scale

Scaling isn’t without its risks. Common pitfalls include:

  • Loss of Focus: As companies grow, it’s easy to lose sight of their core mission and values.
  • Bureaucracy and Inefficiency: Adding layers of management and process can stifle innovation and slow down decision-making.
  • Communication Breakdown: As organizations become larger, it becomes increasingly difficult to ensure that everyone is on the same page.
  • Talent Acquisition and Retention: Attracting and retaining top talent is essential for sustaining hypergrowth.

The Future of Scale

The principles of rapid volume scaling will continue to evolve as technology advances. Emerging trends such as artificial intelligence, edge computing, and the metaverse will create new opportunities and challenges for companies seeking to scale their operations. However, the fundamental principles of focus, efficiency, and a strong culture will remain timeless. Companies that can successfully adapt to these changes will be best positioned to thrive in the age of hypergrowth and shape the future of technology.