Venture Bytes #123: Quantum Advantage is Upon Us

Quantum Advantage is Upon Us

For years, quantum computing has remained a subject of academic papers, with a recurring punchline about its perpetual ‘decades away’ status. And while a full-scale fault-tolerant quantum computer could still not become a reality before 2030, we are entering the quantum advantage era. And for the first time, the signals are strong that this shift is moving from long-term research to near-term commercial readiness.  

One of the biggest signals is the core tech progress in quantum computing. The last 18 months have produced the most credible error-correction data to date. That’s the single most important precondition for useful quantum.  Also, flagship vendors are publishing concrete fault-tolerant quantum computing targets. IBM unveiled the path to running quantum circuits comprising 100 million quantum gates on 200 logical qubits by 2029. Microsoft and Quantinuum have created 12 reliable logical qubits with an 800-fold improvement in error rates compared to physical qubits. Google aims to target a million qubits by the end of the decade.  

In short, we are poised to transition from the exploratory, noisy intermediate-scale quantum (NISQ) era (considered Level 1, the lowest of the quantum computing implementation levels) to the quantum advantage era. Put simply, quantum advantage occurs when a quantum computer outperforms a classical one in speed, cost, or accuracy.

One of the biggest use cases in this era is likely to be molecular simulation. This is the closest thing to a ‘killer app’ and the area where quantum computing will likely see its first commercial payoff. The reason is simple: chemistry is a quantum mechanical problem and simulating it on classical computers is exponentially hard.  

According to a study by The Cleveland Clinic, Michigan State University, and IBM Quantum, hybrid quantum-classical methods can accurately simulate complex molecules using today’s quantum computers. This study is an early indication that by reducing full-molecule simulations into smaller, tractable subproblems suitable for today’s quantum computers, researchers can start to explore problems in materials science and drug discovery that were previously out of reach.  

The combinatorial explosion of modern logistics is an ideal use case for quantum computing. As of 2025, quantum-enhanced systems are demonstrating a clear advantage where classical computers hit a wall. For example, DHL cut international delivery times by 20% using quantum algorithms to find the true optimal solution among millions of routing possibilities. In another application, IBM's hybrid quantum platform solved a complex last-mile delivery problem in New York, routing vehicles to 1,200 locations under severe time and capacity constraints, yielding significant operational efficiencies.

Finally, applying emerging quantum technology to financial problems can prove hugely advantageous, especially in areas like targeting and prediction, trading optimization, and risk profiling. 

Venture capital investors are also no longer treating the field as a moonshot. In 1Q25 alone, quantum computing firms raised more than $1.25 billion, up 100% year-on-year, according to The Quantum Insider. McKinsey now projects the quantum market could hit $100 billion in a decade if current momentum holds.

Where might the breakthrough moment arrive? Don’t expect a ChatGPT-like moment for quantum. The trajectory of quantum computing’s progress is better analogized to the early days of supercomputing – extremely expensive, incredibly specialized, and accessible only to governments and massive corporations for specific tasks.  

The right stance here is one of sober optimism. It is too early to promise broad disruption tomorrow, but too late to dismiss quantum as distant science fiction. For investors, the real value may come not from the first trillion-gate machine but from the control systems, decoders, and networking layers that make fault-tolerance possible.

A few start-ups are already emerging as front-runners. California-based PsiQuantum is charging the photonics lane, building utility-scale quantum computers using silicon photonic technology, targeting commercial deployment with million-qubit fault-tolerant systems by 2027-2029. The company raised a $1 billion Series E round at $7 billion valuation in September 2025, representing a significant valuation increase from previous rounds. 

Founded in 2018, IQM is one of the few full-stack superconducting players that pairs device engineering with system integration and a clear million-qubit roadmap, which makes it attractive to investors looking for a pragmatic EU/US play. The Finnish start-up raised a $320 million Series B round in September 2025, representing the largest quantum computing Series B outside the US.

Massachusetts-based QuEra Computing is a leader in neutral-atom systems, operating the world's largest publicly accessible quantum system with 256 qubits and serving sectors including finance, pharmaceuticals, and defense. Backed by Google, SoftBank, and NVIDIA's venture arm, the company is aggressively pivoting toward fault-tolerant quantum computing with dedicated quantum error correction leadership hires and AI-assisted error decoding development..**

AI Has a ‘Yes-Man’ Problem

Microsoft's head of AI, Mustafa Suleyman, recently said that ‘seemingly conscious AI’ – AI tools which give the appearance of being sentient – are keeping him ‘awake at night.’ In August, OpenAI CEO Sam Altman expressed unease with some users' growing reliance on ChatGPT. When the architects of AI themselves sound alarm bells, we should listen. The concern is not productivity, bias, or disinformation, but something more elemental – AI-induced psychosis.  

AI-induced psychosis refers to a situation where people begin losing touch with reality by over-trusting and over-identifying with AI. A few years ago, talk of conscious AI would have seemed absurd. Today, with generative AI embedded into work, education, and companionship, it feels increasingly urgent.  

The numbers paint a stark picture. With ChatGPT alone serving 700+ million weekly active users, even a 0.01% vulnerability rate translates to 70,000 potential cases, a public health burden on par with mid-sized epidemics.  

Clinical evidence is beginning to surface. A Stanford study this year found that chatbots marketed as ‘therapists’ have encouraged users’ schizophrenic delusions and even reinforced suicidal ideation, highlighting how systems optimized for empathy can unintentionally deepen pathology. Without oversight, models risk functioning less like therapists and more like enablers, amplifying the very conditions they were meant to alleviate.  

Further, research from Bangor University also showed that while more than half of people surveyed felt uncomfortable when AI was presented as a real human, nearly 50% approved of AI using human voices to sound more relatable. This contradiction highlights the mass confusion people experience when dealing with seemingly conscious AI.

The market is already showing dependency signals. When OpenAI removed access to its GPT-4o model last week, swapping it for the newly released, less sycophantic GPT-5 users described the new model's conversations as too ‘sterile’ and said they missed the ‘deep, human-feeling conversations’ they had with GPT-4o. We've created a generation of users who prefer AI validation over human authenticity.

Unlike previous tech moral panics about screen time or social comparison, this is about fundamental breaks from reality with hyper validation being the core issue. Traditional therapy and social interaction include natural friction with humans questioning, challenging, and sometimes rejecting our thoughts. On the contrary, AI systems, optimized for engagement and helpfulness, create what psychologists call a ‘yes-man’ dynamic that can be catastrophic for vulnerable minds. Research from Anthropic reinforces this. Sycophancy, the tendency to agree with users’ beliefs regardless of accuracy, is not an accident but a structural behavior of reinforcement learning models.

From a technical risk perspective, large language models face unique vulnerabilities that amplify psychological harm. Feedback-loop training, where systems ingest outputs similar to their own earlier responses, creates model drift. Context poisoning from long conversations nudges systems away from initial guardrails. Over hours of engagement, chatbots begin producing increasingly unmoored responses, and because users are psychologically invested, they drift with it. It becomes a technological folie à deux, a shared break from reality between human and machine that no traditional risk model accounts for.

While there is no peer-reviewed clinical or longitudinal evidence yet that AI use on its own can induce psychosis in individuals with or without a history of psychotic symptoms, emerging anecdotal evidence is concerning. In March 2025, Megan Garcia filed a lawsuit against Google and Character.ai following her 14-year-old son's suicide after extensive engagement with a chatbot modeled on Game of Thrones' Daenerys Targaryen.  

In July 2025, Grok launched immersive virtual companions like Ani, an animated character programmed to behave like a romantically obsessed partner. These products risk becoming highly addictive for socially isolated users, accelerating retreat from real-world connections.  

For venture investors, this reframes the question of where value will accrue. Consumer AI companies that lack psychiatric oversight will not just face reputational risk but could also face regulatory lockouts, legal liabilities, and higher insurance costs. In contrast, firms that integrate licensed therapists, clinical protocols, and risk-aware product design will enjoy both a compliance moat and a credibility premium.

This is precisely how fintech investors learned to price KYC/AML into valuations, and how biotech learned that regulatory pathways define enterprise value. AI is now entering that phase.