- The BRRR
- Posts
- DeepSeek’s AI Breakthrough Sends Shockwaves Through Tech Markets
DeepSeek’s AI Breakthrough Sends Shockwaves Through Tech Markets
PLUS: Implications for crypto?
TheBRRR’s Thoughts
Editor’s note: the podcast is back! We generate a full conversation between two AIs covering the day’s most interesting stories. Subscribe on Spotify and Apple to get notifications when a new episode drops.
GM.
Markets are violent today as the world prices in some shocking developments and performance of an LLM funded by the rainy-day fund of some Chinese hedge fund.
Deepseek’s R1 model has matched OpenAI’s latest model in performance while spending a mere fraction of what OpenAI spent on GPUs and training.
This has called into question whether or not the hyperscalers in big tech actually need to spend tens of billions on Nvidia equipment to reach artificial super intelligence.
In isolation, it’s a wildly optimistic development that the world might only have to spend a fraction of what it expected to reach artificial super intelligence. This should have deflationary affects across the global economy, given time.
Markets are fearful today because this level of disruption and market-wide expectations gives people the perception that they have no idea what is actually going to happen anymore.
With confusion and uncertainty comes selling of speculative assets.
In a world with abundant intelligence and automation, how do you preserve or grow your wealth? We believe that digital-native money continues to be a sound bet, as swarms of AI agents will use blockchain rails because of its permissionless nature and programmability.
Additionally, access to compute may no longer represent a massive moat in the business world, but the case for proprietary data (for model training purposes) as a strong moat has strengthened.
This would bode well for companies like Meta, Apple, and Google at the expense of companies like Nvidia and AMD.
DeepSeek’s AI Breakthrough Sends Shockwaves Through Tech Markets, Challenging U.S. Dominance
Synopsis:
Chinese AI startup DeepSeek is disrupting global markets with its R1 model, claiming top-tier performance while using lower-cost resources, including NVIDIA's export-restricted H800 chips. Developed for just $5.6 million, R1 rivals OpenAI’s o1 in benchmarks for reasoning, coding, and mathematics, sparking fears of a new AI arms race. Following the announcement, tech stocks plummeted, with NVIDIA losing $465 billion in market value, marking a historic drop. This development exposes vulnerabilities in U.S. semiconductor export controls and raises geopolitical and market risks.
The Details:
Technological Advancements:
DeepSeek's R1 Model:
Developed at $5.6 million, a fraction of OpenAI’s model costs ($100M–$1B estimates).
Trained on NVIDIA H800 chips, which are less advanced than A100 or H100 chips restricted by U.S. export bans.
Performance benchmarks:
MATH-500: R1 scored 97.3%, outperforming OpenAI’s o1 at 96.4%.
Codeforces coding benchmark: R1 achieved 96.3%, just behind OpenAI's 96.6%.
Ranked 3rd globally on Chatbot Arena, tied with OpenAI’s premium o1 model.
Market Impact:
DeepSeek’s emergence caused a sharp sell-off in global tech stocks, underscoring investor fears:
NVIDIA: Dropped 15%, wiping out $465 billion in market value—the largest single-day value loss in history.
Broadcom: Fell 16%, while Microsoft (-3.7%) and Alphabet (-3.5%) also declined.
Nasdaq: Closed down 3.5%, marking its 3rd-worst day in two years.
Geopolitical Ramifications:
U.S. Export Controls Under Scrutiny:
DeepSeek’s use of H800 chips, despite export bans, suggests workarounds or inefficiencies in the U.S. sanctions regime.
Analysts speculate DeepSeek may have rented or indirectly acquired restricted chips, raising questions about enforcement.
Demonstrates China’s ability to deliver cutting-edge AI with constrained resources, potentially challenging U.S. dominance in AI innovation.
AI Arms Race Intensifies:
With Trump endorsing OpenAI's $500 billion Stargate project to maintain U.S. AI supremacy, DeepSeek’s rapid ascent highlights the fierce competition between the U.S. and China.
Security and Governance Challenges:
Cyberattack:
A cyberattack forced DeepSeek to temporarily limit new user registrations during its app's surge to #1 on Apple’s App Store in the U.S. and several other markets.
This highlights growing security concerns as the company scales rapidly.
Censorship:
While DeepSeek performs well on general knowledge, it avoids or self-censors sensitive topics like Tiananmen Square or China’s human rights record, showcasing government influence over AI outputs.
DeepSeek’s Vision:
Open Source Advantage:
R1 is fully open source, available for anyone to access, modify, and deploy.
DeepSeek claims minimal profit motives, pricing its products just above production costs to undercut competitors.
WHY IT MATTERS:
Market Disruption:
NVIDIA’s Dominance Threatened: With DeepSeek achieving comparable AI performance using mid-tier chips, demand for premium hardware like H100s may soften, pressuring NVIDIA’s margins.
AI Democratization: Open-source models like R1 could commoditize AI, threatening high-cost, proprietary platforms like OpenAI.
Geopolitical Implications:
DeepSeek highlights how China is closing the AI gap, even under export restrictions.
A successful AI arms race may bolster China’s influence in tech and military applications, escalating tensions with the U.S.
Investor Opportunities and Risks:
Short-Term Risks: High-growth tech stocks like NVIDIA and Broadcom remain vulnerable to further volatility as AI competition intensifies.
Long-Term Themes: Look for opportunities in second-tier chipmakers (e.g., AMD, TSMC) or cloud infrastructure providers that can support open-source models.
Regulatory Questions:
U.S. export controls need revisiting, as DeepSeek demonstrates China’s ability to circumvent restrictions and innovate with limited resources.
Key Bullet Points:
DeepSeek R1 Performance: Rivals OpenAI’s o1; scored 97.3% on MATH-500, outperforming o1.
Low-Cost Development: Trained for $5.6M using mid-tier NVIDIA H800 chips.
Market Fallout:
NVIDIA lost $465 billion in market value, dropping 15%.
Nasdaq fell 3.5%, marking its 3rd-worst day in two years.
Security Challenges: Cyberattack forced DeepSeek to limit registrations during app’s surge to #1 on Apple’s App Store.
Censorship: Self-censors sensitive topics like Tiananmen Square, reflecting Chinese government influence.
Open Source Impact: Challenges high-cost proprietary AI models like OpenAI, paving the way for democratized AI tools.
Trading Takeaways:
Tech Sector Volatility:
Continued pressure on NVIDIA, Broadcom, and AI-driven growth stocks. Consider hedging exposure to high-growth tech.
Opportunities in Infrastructure:
Cloud providers (AWS, Azure) and hardware players (AMD, TSMC) may see demand from open-source AI adoption.
AI Disruption Watch:
DeepSeek’s open-source model signals a shift away from high-cost proprietary platforms. Monitor mid-cap tech companies that can capitalize on the open-source wave.
Geopolitical Risks:
Heightened U.S.-China tensions may impact global tech trade. Investors should monitor regulatory changes or sanctions that could affect the AI and semiconductor sectors.
DeepSeek’s rise is a wake-up call for markets, governments, and tech incumbents. With AI barriers lowering, the industry could be entering a new era of open-source disruption, intensifying competition and market volatility. Keep an eye on NVIDIA and other Big Tech players as the fallout continues.
Start learning AI in 2025
Everyone talks about AI, but no one has the time to learn it. So, we found the easiest way to learn AI in as little time as possible: The Rundown AI.
It's a free AI newsletter that keeps you up-to-date on the latest AI news, and teaches you how to apply it in just 5 minutes a day.
Plus, complete the quiz after signing up and they’ll recommend the best AI tools, guides, and courses – tailored to your needs.
Premium Subscriber Section
You’ll need to upgrade your subscription to view our portfolio and get our real-time trade alerts. You can upgrade for $3/month or $14.99/year.
Trades, Watchlist & Live Portfolio
(paywall only)
Most Recently Revealed Trade:
Wednesday April 17 2024: We bought more Solana at $131 and added Solana’s top memecoin WIF at $2.36 on the heels of a leverage wipeout dip after the WW3 scare.
Here’s the link to The BRRR Technical Analysis Chatbot - let me know what you think!
Portfolio Notes
Monday November 4 2024: We haven’t updated the portfolio below, but we’re buying AI memecoin GOAT at its current $520m valuation as the fastest horse in a broad crypto rally post-election.
June 12: These assets all look great for continuation higher.
We are considering moving on from Tesla as it has lagged the rest of our portfolio badly and doesn’t have an obvious catalyst. We’ll monitor and let you know if we decide to move on.
Older Notes
Wednesday, April 3, 2024: We haven’t deployed the cash yet, but are eyeing exposure to a few assets including META and PLTR.
Monday, March 11, 2024: We sold Apple this morning. The newsletter held the stock from inception a year ago for a meager 12% gain.
The company has lost its magic evident by complacent iPhone releases, lack of a coherent vision for AI integration and punitive & anti-competitive App Store policies.
We believe the stock will move in-line with the broader Nasdaq going forward.
We’ll sit on the cash for now, but plan to redeploy it quickly.
Watchlist
$META: Sleeper in AI race and ad biz is proving resilient
How was today's email? |
Got feedback? Follow the writer on Twitter @frank_locascio and send a message.
The BRRR is meant for informational purposes only. It is not investment advice. Please consult with your investment, tax, or legal advisor before making any investment decisions.
Reply