Venture Capitalist Gaurav Tewari on Why “Cheaper, Faster, and Smarter” AI Models Will Define the Next Decade—and What That Means for Investors and Innovators.
Meet DeepSeek: A Provocative New AI Contender
Technology revolutions follow a familiar pattern—like the shift from mainframes to cloud computing. Once controlled by a few industry giants, breakthrough innovations rapidly become accessible to all, unleashing exponential adoption. DeepSeek is driving that shift in AI, proving that cutting-edge intelligence no longer depends on sheer computational scale.
Its arrival has set off a wave of conversations among top investors and founders. As an open-source AI model, DeepSeek rivals some of the most advanced proprietary systems—yet runs faster, costs less at inference, and challenges conventional wisdom about ever-expanding foundation models.
In a candid fireside chat hosted by Citizens Bank, I shared my perspective on DeepSeek’s emergence and its implications for the broader AI landscape. At Omega Venture Partners, we don’t wait for AI trends to emerge—we define them by backing the technologies that will shape the future before the market catches on. DeepSeek isn’t a surprise—it’s exactly the kind of AI transformation Omega has been backing for years. We don’t just follow the AI market—we shape it by investing in companies that make AI smarter, cheaper, and exponentially more scalable.
The AI Evolution
As a graduate student in Computer Science, I was absorbed in the nascent world of Software Agents at MIT and the MIT Media Lab, back when computational resources were scarce and AI evoked images of futuristic robots more than enterprise software. My journey would take me to Microsoft and McKinsey, then to Wharton, and eventually to the world of institutional tech investing at firms like Highland Capital, SAP Ventures, and Citigroup.
“AI’s true transformation isn’t in the future—it’s happening now. At Omega, we saw the inflection point early and positioned ourselves ahead of the curve, investing in companies that are turning AI’s raw power into real business results.”
At each step, I noticed a recurring pattern: as soon as technology becomes more efficient, flexible, and cost-effective, usage skyrockets. The potential for new business models expands, while entire waves of innovation take hold. A prime example is the shift from on-premise software to SaaS—once it took off, it produced multi-hundred-billion-dollar cloud giants in short order. With Omega Venture Partners, my team and I focus on AI-driven companies that solve real enterprise challenges. We saw the wave coming years ago, raising our first fund right as AI was hitting a critical inflection point—far ahead of when the broader market recognized AI’s promise.
Today, I’m more certain than ever: we stand at another evolutionary milestone. And if DeepSeek’s early performance data is any indicator, the next chapter of AI will see more efficiency, more open-source collaboration, and more innovation. But let’s break it all down, section by section.
1. Behind the Scenes of DeepSeek’s Market Disruption
When cloud computing disrupted enterprise IT, it didn’t just cut costs—it unlocked entirely new business models, enabling hypergrowth for companies like AWS and Salesforce. DeepSeek is doing the same for AI, removing cost barriers and enabling far more businesses to deploy AI at scale.
Likewise, if DeepSeek runs cheaper and faster at inference while offering comparable quality to the top proprietary models, it doesn’t kill the market for compute; in fact, it expands AI’s total addressable market. More businesses will be able to adopt and deploy AI in new ways, driving overall demand for accelerators and specialized hardware.
The real takeaway is that bigger isn’t always better in model building. DeepSeek leverages model distillation—where a large, complex model (the “teacher”) trains a smaller, optimized model (the “student”) to deliver similar high-level results with far fewer resources. The result is a smaller, nimbler system that’s proving once again that open-source efforts can catch up to heavily funded proprietary systems. This shows us that model architecture and clever engineering matter as much as, if not more than, raw compute.
2. Open Source vs. Proprietary: A Shifting Balance of Power
Open source in AI has a fundamental advantage in attracting the best researchers, fostering collaboration, and iterating faster. That has been the case for years with frameworks like PyTorch, TensorFlow, and Hugging Face. But the foundation-model layer seemed different—many believed only big tech players with bottomless pockets could field cutting-edge large language models.
DeepSeek’s rise challenges that assumption. Built on open-source practices (though with questions remaining about its training data sources), it underscores that R&D driven by open-source can yield surprisingly robust models. The AI landscape is splitting into two factions: open-source models that democratize AI innovation, and proprietary systems that must justify their price premium. Omega recognized this shift early, investing in companies that leverage open-source advantages while maintaining enterprise-grade reliability.
As open-source AI models approach parity with proprietary systems at a fraction of the cost, premium pricing models will face increasing scrutiny—forcing enterprises and investors to rethink AI’s economic landscape.
All of this is fueling talk of a split AI future: open-source models that democratize development on one side, and niche proprietary platforms that lean on scale, enterprise relationships, and brand trust on the other. Neither approach is going away, but the balance of power will likely continue to shift.
3. Training vs. Inference—and Why Efficiency Matters
When discussing the hardware implications of DeepSeek and other advanced models, it’s instructive to separate the cost of training from the cost of inference. Training a large model can be astronomically expensive, often requiring vast amounts of specialized hardware such as Nvidia GPUs at scale. But for many businesses deploying AI in daily workflows, the bigger concern is inference—i.e., the cost and speed of generating responses or outputs.
“Compute costs have been the gatekeeper of AI adoption. As inference becomes radically cheaper, AI is shifting from an elite technology to an everyday business necessity.”
DeepSeek reduces inference costs so much that it would erode the advantage of some top commercial models. Yet that efficiency does not necessarily mean an overall decline in data-center capital expenditure. Quite the opposite: once you lower the barrier to entry, companies that would never have built custom AI solutions may now do so. Just as cloud computing slashed IT infrastructure costs and fueled SaaS dominance, cost savings at the per-inference level can spark even heavier aggregate usage across thousands of new AI-powered applications.
In turn, big infrastructure players—from Microsoft and Amazon to Oracle and SoftBank—are doubling down on building out robust GPU and ASIC-based data centers. They see a market that’s far from saturating. Innovations in AI architecture, whether open-source or not, look poised to accelerate this expansion, not reduce it.
4. Enterprise Applications: Real ROI Over Hype
For the venture capital community, the question isn’t which foundation model is bigger, better, or more novel. It’s how quickly AI can deliver tangible ROI to customers. Early on, robotic process automation (RPA) teased the idea of slashing repetitive workflow costs, but proved brittle at scale. AI-driven applications can go further, automating tasks like analyzing call center transcripts, scanning invoices, or summarizing contracts.
That’s the sweet spot for near-term monetization: harnessing AI to handle repetitive, data-intensive tasks and free up human capital for higher-value work. Whether it’s a global insurance giant analyzing thousands of documents daily or a consumer tech platform extracting insights from user interactions, the principle is the same: AI is best deployed as an ‘Iron Man suit’—not a ‘Terminator replacement.’
“AI isn’t just an efficiency play—it’s a profit center. The next market leaders won’t be the ones using AI to cut costs; they’ll be the ones using it to create entirely new revenue streams.”
Companies that can seamlessly layer AI onto existing workflows, ensuring reliability, security, and compliance, stand to generate immediate, measurable returns. Meanwhile, larger incumbents—Salesforce, ServiceNow, or other established SaaS players—will likely incorporate AI-based enhancements to keep customers in their ecosystem. But that’s no guarantee they’ll stave off newer “AI-native” challengers, just as on-premise software could not prevent the rise of modern SaaS giants.
5. Security and Geopolitical Implications
Of course, no conversation about foreign-developed AI is complete without addressing privacy and security concerns. The chat raised a pivotal question about potential data exfiltration, especially if a model or its code is developed in a jurisdiction where state intelligence could demand compliance. This remains a legitimate obstacle to widespread adoption of certain open-source models, particularly if government or highly regulated enterprises fear hidden vulnerabilities.
And let’s not forget the broader geopolitical dimension. Governments worldwide recognize that leading in AI is tantamount to wielding the next generation of economic and strategic clout. Academia and the research community may be relatively agnostic to border politics, but regulators and national-security bodies do not share that view. For many American businesses, trusting a foreign open-source model is likely a non-starter, no matter how attractive the performance metrics.
Final Thoughts: The Era of Democratized AI
From my vantage point as a longtime AI investor and domain expert, one thing is abundantly clear: the DeepSeek phenomenon isn’t a mere novelty, but rather a sign of a deeper trend. As open-source research drives down costs and reduces barriers, we’ll see more organizations adopting AI and more emergent players staking their claim.
Yet for all the talk of disruption, the real action is in enterprise applications. The next wave of trillion-dollar value creation will come from turning AI’s raw horsepower into frictionless user experiences with quantifiable business outcomes. Whether that involves summarizing complex calls, parsing mountains of financial data, or delivering entirely new agent-based interactions, it all boils down to one metric: ROI.
The AI industry is reaching an inflection point—where efficiency breakthroughs drive mass adoption at an unprecedented scale. For those of us who have followed AI’s twisty path from niche research to mainstream game-changer, the script remains familiar: make it easier, make it cheaper, and everyone will want it. The difference now is speed. DeepSeek has shown what’s possible when the community rallies behind new architectures and open-source breakthroughs.
AI is shifting from theory to mass enterprise adoption. The winners of the next decade are being decided today. At Omega, we don’t just invest in AI—we invest in the companies defining its future. The time to act isn’t next year or next quarter—it’s now.
#########
