DeepSeek Just Changed AI - But the Real Lesson Is Bigger Than Tech
By Bryan Kaus
DeepSeek has emerged as one of the most intriguing developments in AI—shocking NVIDIA and shaking up the broader market. As an NVIDIA investor since 2020, I’ve watched its meteoric rise with both excitement and caution. The market can be irrationally exuberant, and what climbs rapidly often faces corrections. That’s why I rarely put all my eggs in one basket.
Beyond the market reaction, DeepSeek’s emergence raises critical questions about AI’s future—on everything from training costs and hardware demand to energy consumption and the risks of unchecked automation. These are the kinds of inflection points where hype and reality collide, and where the real winners are those who can see past the noise.
1. A Paradigm Shift in AI Training Costs
Reports suggest DeepSeek trained its model for around $6 million—a fraction of the hundreds of millions typically required for massive AI models. If this holds, it could mark a significant shift, particularly if DeepSeek can run efficiently on more affordable chips without sacrificing too much performance. This lower entry point could drive wider adoption of AI across industries, from consumer applications to enterprise solutions in medicine, engineering, and beyond.
But here’s the bigger question: what does this mean for AI’s energy demand? A more efficient training process could suggest reduced power needs, but history tells us that efficiency gains often lead to more widespread usage - ultimately driving demand higher, not lower.
The AI boom is already reshaping energy projections, and DeepSeek throws another variable into the equation. If energy forecasts assume a steady, linear rise in power demand based on traditional training models, a shift to lower-cost, lower-power AI could lead to miscalculations. At the same time, AI’s expansion across industries will create entirely new energy requirements.
This is where we need to be careful. We’ve seen industries misread demand curves before. Take the early 2000s fiberoptic boom—telecom companies raced to lay down massive infrastructure expecting exponential data growth, only to end up with extreme overcapacity when that growth didn’t materialize as quickly as expected. The lesson? Infrastructure investment needs to be built on adaptable, reality-checked projections, not just momentum. AI-driven energy demand could follow a similar trajectory—either overshooting or undershooting real needs, depending on how the technology scales.
Bottom line: AI’s energy needs will grow, but how fast and where remains an open question. Reliable electricity sources will be more critical than ever, and companies that get their energy forecasts wrong—whether overbuilding or underestimating demand—will be exposed to major risks.
2. Market Implications & the Unfinished AI Race
Despite the headlines, DeepSeek doesn’t spell the end for major chip players like NVIDIA. AI is still in its early days, and the demand for high-performance computing isn’t going away. While cheaper AI models could challenge certain assumptions about the market, specialized chips will remain crucial for cutting-edge applications.
That said, AI cycles—like all tech cycles—don’t move in a straight line. Breakthroughs create waves of excitement, but every boom comes with eventual recalibrations. Investors and businesses that plan for multiple scenarios, rather than blindly betting on one outcome, will be in the best position to adapt.
For AI companies and chipmakers, this means avoiding overcommitment to any single model of the future. For businesses investing in AI capabilities, it means keeping an eye on flexibility—ensuring that their strategies can pivot as new developments emerge. The winners won’t necessarily be those who move the fastest, but those who can stay nimble as the landscape shifts.
3. Bias, Governance & the Human Factor
One major criticism of DeepSeek is its apparent bias—certain topics seem off-limits, and the model reportedly avoids discussing politically sensitive issues. But let’s be honest: every AI model has biases, including the ones built in the West. The issue isn’t just that bias exists—it’s that AI models can create feedback loops that reinforce narrow perspectives, often without users even realizing it.
The real risk isn’t just biased models, but the human tendency to blindly trust AI outputs without questioning them. I’ve seen this firsthand.
Years ago, we rolled out an automated sales model that claimed a 45% ROI. Looked great—until we double-checked the math and found it was actually -20%. The problem? Some reps trusted the system completely, failing to verify whether the numbers made sense. AI is only as reliable as the people interpreting it, and if we hand over decision-making authority without scrutiny, we’re setting ourselves up for costly mistakes.
This applies far beyond sales models. AI is increasingly making decisions in areas like healthcare, finance, hiring, and even national security. If organizations treat AI recommendations as gospel rather than as tools to be challenged and verified, the risks compound. That’s why governance, transparency, and continuous oversight aren’t just “nice to have” features—they’re essential.
4. Stay Nimble, Stay Thoughtful
DeepSeek is a reminder that AI’s future is far from settled. No one knows for certain how demand for AI will evolve, which models will dominate, or how power requirements will shift. That’s why agility, diversification, and critical thinking matter.
For investors, it’s a reminder to stay diversified and avoid getting swept up in market euphoria. The AI race is still early, and there will be corrections along the way.
For business leaders, it’s a signal to focus on scenario planning - ensuring that AI investments align with adaptable strategies rather than rigid assumptions. Betting everything on a single technological path is a high-risk move.
For everyone engaging with AI, it’s a call to remain engaged, skeptical, and proactive. AI can be a transformative tool, but it doesn’t replace human judgment. The more advanced AI becomes, the more critical it is that we maintain our ability to think independently, challenge assumptions, and verify the numbers ourselves.
The AI landscape is evolving fast, but we should never evolve past the need for real scrutiny -that's how we win, and do it right.