With over 70% accuracy! Theory Ventures' 12 ultimate predictions for 2026: AI, IPOs, and stablecoins.

Original text by Tomasz Tunguz , founder of Theory Ventures.

Compiled by: Yuliya, PANews

Every year at the end of the year, Tomasz Tunguz, founder of Theory Ventures, systematically reviews his predictions for the past year and makes new judgments for the coming year based on them.

In his review of 2025, Tunguz gave himself a high score of 7.85 out of 10 on his ten predictions (one point each). That year saw profound structural changes in areas such as artificial intelligence, capital markets, data infrastructure, and crypto finance. These changes not only validated most of the predictions but also laid a clear foundation for development trends in 2026. This article reviews the market events of 2025 and looks ahead to twelve major trends for 2026. PANews has compiled and edited these two articles.

A Review of the Top 10 Predictions for 2025

Prediction 1: The IPO market will experience a boom. Score: 0.6

Overall, the IPO market saw a significant recovery in 2025. A total of 46 software companies went public, raising a total of $12.3 billion, significantly higher than the 21 companies and $3.8 billion in 2024, but still far below the scale of the peak period for technology IPOs in 2021.

CoreWeave and Circle successfully went public, demonstrating strong market capitalization and post-IPO trading performance. Meanwhile, companies like Figma and Chime saw their share prices fall below their previous private equity valuations, indicating a more rational valuation approach in the public market. Some high-profile companies, such as SpaceX, Stripe, and Databricks, did not complete their IPOs in 2025, but they have accumulated significant potential momentum for the years to come.

Prediction 2: Google will continue to expand in the field of artificial intelligence. Score: 1

Google has reclaimed its leading position in artificial intelligence, ranking among the top in almost all major AI categories. Its Gemini3 models have achieved fundamental leaps in pre-training efficiency and multimodal integration. Gemini3 Flash has redefined the industry frontier in performance and latency, becoming the default engine for high-frequency agent workflows. In the open-source space, the Gemma series models continue to hold the top position in their respective weight levels, delivering 70B-level inference capability with a parameter scale of 27B. Even in the creative media field, Google's video models rank among the top three globally, with their priority on temporal consistency and role stability making them ideal for enterprise applications.

Prediction 3: Voice will become a crucial entry point for human-AI interaction. Score: 1

According to an OpenAI report, by October 2025, voice chat will account for 19% of ChatGPT's user interactions. The number of voice assistants worldwide has reached 8.4 billion, with 153 million users in the United States. It is estimated that 80% of enterprises plan to integrate AI-driven voice functionality into their operations by 2026. Using Whisper and WisprFlow for voice input, and conversing with intelligent agents such as Gemini Live, has become commonplace.

Prediction 4: Total US venture capital investment will remain between $210 billion and $230 billion, but venture capital fund raising will increase by 20%. Score: 0.5

In terms of venture capital, total VC investment in the US for the year was approximately $220 billion, in line with forecasts, primarily driven by mega-AI funding rounds. However, fundraising failed to grow, declining by about 20% year-on-year, with total fundraising reaching approximately $65 billion. Although the number of deals rebounded somewhat at the beginning of the year, persistent liquidity constraints and a sluggish exit environment kept LPs cautious.

Prediction 5: Integration is a key theme of the modern data stack. Score: 1

2025 will be a record year for data infrastructure mergers and acquisitions. The “modern data stack” will transform from a collection of “best-in-class” tools into a race to build vertically integrated platforms.

This wave of consolidation extends downstream, demonstrating that the core of current competition lies in computing power, computing resources, and integrated software. Of particular note is CoreWeave's series of acquisitions, which signify the rise of "full-stack hyperscale cloud service providers"—vendors possessing a complete technology stack from GPUs to MLOps.

Prediction 6: The first company with annual recurring revenue (ARR) of $100 million and fewer than 30 employees will emerge. Score: 1

AI-native teams have redefined our understanding of business efficiency. Cursor achieved $100 million in ARR by January 2025 with only 12 employees, and Midjourney achieved $500 million in ARR with a team of approximately 100 people. This efficiency far surpasses that of traditional SaaS companies, demonstrating the significant advantage of intelligent agent software in terms of capital efficiency. In comparison, Slack had 650 employees when it reached $100 million in ARR, Ramp had 275, and Wiz had 400.

Prediction 7: After years of decline, the number of Web3 engineers in the US has grown by 25% as governments embrace cryptocurrency and Web3. Score: 1

In 2025, the number of Web3 jobs in the United States grew by 26% to 21,600. The significant shift in the regulatory environment has sparked a surge in institutional adoption and spurred the emergence of new consumer applications based on decentralized technology stacks.

Prediction 8: The AI race drives GPU demand, with hyperscale service providers spending over $125 billion annually in data centers. Broadcom becomes the hottest semiconductor stock of the year. Score: 0.75

Investment in computing infrastructure far exceeded expectations. By 2025, hyperscale cloud vendors' capital expenditures are projected to reach $315 billion to $350 billion, with Amazon accounting for approximately $100 billion, Microsoft around $80 billion, and Google around $75 billion. In the semiconductor sector, Broadcom's stock price surged due to demand for AI network construction, even outperforming Nvidia in the second half of the year. Although its annual growth ranked third in the industry, behind Micron and Google, its performance was still impressive.

Prediction 9: Stablecoin supply will grow by 50% to $300 billion, with transaction volume more than three times that of Visa. Score: 1

Stablecoins have become an important part of global payments. As of December 2025, the total supply of stablecoins reached $310 billion, with annual on-chain transaction volume exceeding $46 trillion, nearly three times the transaction volume of Visa. The adoption of stablecoins in B2B payments is accelerating as businesses seek faster and cheaper cross-border settlement methods.

Prediction 10: Observability, SIEM, and business intelligence will begin using the same data lake. Score: 0

This prediction failed to materialize. While usage-based pricing models did drive demand for a single data lake, and the dominance of data lake architectures across all workloads became increasingly apparent, the vision of observability, security information and event management (SIEM), and business intelligence (BI) sharing the same data lake did not materialize by 2025.

12 major predictions for 2026

Based on observations from 2025, the predictions for 2026 are more systematic and in-depth. The core logic is that artificial intelligence is evolving from an auxiliary tool to an autonomous system, and from cutting-edge experimentation to core infrastructure. 2026 will be a year in which enterprises massively invest AI in practical production applications.

First, for the first time, companies will spend more on AI agents than on human labor. This phenomenon has already emerged on the consumer side; for example, Waymo's average cost for self-driving cars is higher than Uber's, yet demand continues to grow. After considering the overall costs of recruitment, training, and management, companies will accept premium pricing for agents in repetitive tasks.

Second, 2026 will be a year of record liquidity. Companies such as SpaceX, OpenAI, Anthropic, Stripe, and Databricks are expected to go public, with SpaceX and OpenAI's IPOs potentially ranking among the top ten in history. Meanwhile, facing the disruptive threat posed by AI, traditional enterprises will launch a defensive M&A wave exceeding $25 billion, choosing to "buy" rather than "build" AI capabilities.

Third, vector databases will once again become a key infrastructure in the AI technology stack. Multimodal models and world models place new demands on data structures, and vector databases, as the central hub connecting basic models and enterprise data, will experience explosive revenue growth.

Fourth, AI models will be able to autonomously execute tasks for more than a full workday. According to METR data, the duration of AI tasks doubles every seven months. Following this trend, by the end of 2026, AI agents will be able to continuously complete workflows of more than eight hours, fundamentally changing the way projects are configured.

Fifth, AI budgets are under systematic scrutiny for the first time. Boards of directors and procurement committees are beginning to question AI spending, and small models and open-source solutions are gaining popularity due to their cost advantages. Research teams are enabling task specialization to achieve or even surpass the performance of cutting-edge models at low cost, allowing developers to achieve orders-of-magnitude cost reductions.

Sixth, Google has further widened the gap with its competitors by leveraging the breadth and depth of its AI strategy. Breakthroughs in multiple areas, including cutting-edge models, edge inference, video generation, open-source weighting, and search integration, have forced companies like OpenAI, Anthropic, and xAI to focus on niche areas, ending the era of all-encompassing competition.

Seventh, agent observability has become the most competitive layer in the inference stack. As AI agents become more deeply integrated into enterprise operations, traditional engineering, security, and data observability will merge into a unified discipline. Enterprises need to conduct integrated, end-to-end monitoring of AI agent code execution, security threats, and data lineage. The convergence of the three major observability domains predicted for 2025 will find its starting point here.

Eighth, by December, 30% of international payments will be completed through stablecoins. The efficiency advantage of stablecoins in cross-border settlements is too significant. As regulations in major markets become clearer, stablecoins are moving from the periphery of crypto to the core of global trade finance, replacing some B2B transactions under the SWIFT system.

Ninth, the data access patterns of AI agents will overwhelm existing database architectures. Their query counts and concurrency demands are at least an order of magnitude greater than those of humans, forcing transactional and analytical databases to undergo a complete architectural restructuring to cope with the ever-increasing demands for automation.

Tenth, data center construction will reach 3.5% of US GDP by 2026. To support the exponential computing power demands of AI, data center construction will reach an unprecedented scale, with investment comparable to the historical expansion of railroads. The only thing that might slow this process is the potential risks in the credit market, particularly the rising default rates in the private lending sector, which could become a bottleneck for these ultra-capital-intensive projects.

Eleventh, the internet will shift towards an "agent-first" design. In the future, a large amount of developer documentation and websites will be designed and optimized primarily for AI agents rather than humans. This is because AI agents will be responsible for the initial information gathering and comparative analysis in more and more business purchasing decisions. This means that the "front door" of a website needs to be open to robots, while a "side door" will be reserved for human visitors.

Twelfth, Cloudflare will become a key hub for proxy payments. The x402 protocol reactivates the long-dormant 402 "Payment Required" status code in HTTP, enabling AI proxies to pay for API access in real time. Given its position in network infrastructure, Cloudflare naturally becomes the gateway to this new business model, thus triggering a new round of debate about openness versus centralization.