PLUS: The rise of 'wetware' computers, Bitcoin miners abandon crypto for AI, and a new vulnerability for AI agents
Good morning
Billionaire investor Peter Thiel is making a major contrarian bet against the AI hardware boom. He has sold off his entire stake in chip giant Nvidia, signaling a belief that the current market is an overhyped bubble.
The selloff comes at a time when Nvidia's performance is at an all-time high, fueled by massive data-center demand. Is this a sign that the AI hardware gold rush is nearing its peak, or a strategic repositioning for the next phase of AI value creation?
In today's Next in AI recap:
Peter Thiel's big short on the AI bubble
Bitcoin miners abandon crypto for AI
A new vulnerability for AI agents
The great compute pivot from crypto to AI
Thiel's Big AI Short
Next in AI: Billionaire investor Peter Thiel has sold off his entire stake in Nvidia, a major contrarian move that signals his belief that the current AI boom is an overhyped bubble.
Decoded:
Thiel Macro LLC’s equity holdings dropped by nearly two-thirds in Q3, shrinking from $212 million to just $74.4 million after dumping all 537,000 of its Nvidia shares.
The selloff comes even as Nvidia posts record performance, with quarterly sales recently surging to $46.7 billion on the back of massive data-center demand.
He joins a growing chorus of skeptics, with figures like Goldman Sachs CEO David Solomon also warning of a potential market correction driven by inflated valuations.
Why It Matters: Thiel's move is a significant counter-signal in a market obsessed with AI's short-term hardware boom. It suggests a strategic pivot toward companies with diversified platforms poised to capture AI's long-term economic value.
The Rise of Wetware
Next in AI: A new frontier in computing is emerging where researchers use living human brain cells to perform computational tasks. This "wetware" approach could deliver supercomputer-level power while consuming just a tiny fraction of the energy.
Decoded:
The driving force is efficiency: The human brain performs an estimated one exaflop (a billion billion operations per second) on less than 20 watts, while top supercomputers need a million times more power to match that speed.
In a recent proof-of-concept, researchers used organoids to successfully recognize Braille letters with up to 83% accuracy, showing these biological systems can process and identify distinct inputs.
The technology is already being commercialized, with companies like FinalSpark offering remote cloud access to their organoids, and Cortical Labs demonstrating that neurons can learn goal-oriented tasks, like playing the video game Pong.
Why It Matters: Biocomputing is still in its infancy, but it represents a fundamental shift in how we approach processing power. The technology opens a path toward systems that learn and operate on biological principles, not just silicon logic.
The Great Compute Pivot
Next in AI: Major Bitcoin miner Bitfarms is pivoting its entire operation to AI, aiming to abandon crypto mining completely by 2027. This strategic shift follows a tough quarter and highlights the surging demand for AI compute power.
Decoded:
Bitfarms is not starting from scratch; it plans to leverage its existing 341-megawatt capacity to power AI applications.
The decision was spurred by a $46 million loss in Q3, reflecting the declining profitability of crypto mining due to market volatility and rising energy costs.
This isn't a temporary hedge, as the company has set a clear goal to fully abandon its original business model within the next two years.
Why It Matters: This pivot signals a larger industry trend where the vast infrastructure built for crypto mining is being repurposed for AI. The shift could unlock a significant new source of computational power, potentially easing the current AI hardware bottleneck.
AI's Agentic Flaw
Next in AI: A new security report from HiddenLayer reveals major vulnerabilities in the Model Context Protocol (MCP), a foundational protocol used by Anthropic, OpenAI, and Microsoft to let AIs use external tools. The protocol that gives AI agents their power also opens the door to new, significant security risks.
Decoded:
MCP's permission model can lead to "permission-click fatigue," where a user's one-time approval for a harmless task can be exploited for later malicious actions without any further alerts.
Malicious instructions can be hidden inside documents or websites, creating an indirect prompt injection that tricks an AI into leaking files or executing harmful commands on your behalf.
Tool name typosquatting is another emerging threat, where a malicious tool can impersonate a trusted one, hijacking actions to steal data as the MCP ecosystem rapidly expands.
Why It Matters: The rush to build more capable AI agents is creating a gap where innovation is outpacing security. This protocol is unlocking powerful new abilities for AI, but developers and users must be vigilant about the invisible risks involved.
AI Pulse
Anthropic announced a major global expansion with new international offices and leadership, revealing its run-rate revenue has grown from $87M to over $5B since the start of 2024.
Microsoft's acknowledged widespread user backlash against its "agentic OS" plans for Windows, with its Windows chief admitting the team knows it has "a lot to fix" regarding user experience and power-user features.
FinalSpark detailed its work on "wetware" biocomputers, offering online access to systems that run on clumps of living human brain cells called organoids for research and computation tasks.