Weekly Investment Update (08/21/2026)

Two themes are increasingly shaping the market debate: who ultimately captures the economics of AI, and what a higher-for-longer rate environment means for asset prices. On AI, the latest Chinese models are closing the performance gap with leading U.S. systems while doing so at materially lower cost. That should be good for adoption and usage, but it also challenges the assumption that the largest share of value will necessarily accrue to the companies that own the most advanced proprietary models. The opportunity remains enormous, but the market may have to become more discriminating about where durable pricing power and returns on capital ultimately reside. Nvidia’s earnings next week will be an important test of how strong infrastructure demand remains as that debate evolves.

At the same time, the rise in long-term Treasury yields deserves attention, but we do not view it as an inherently bearish signal. Part of the move reflects a higher term premium tied to fiscal uncertainty and geopolitical risk, but part also reflects stronger nominal growth, rising capital investment, and the sheer scale of the AI buildout itself. Equity markets have so far absorbed that move remarkably well, which suggests investors are distinguishing between higher yields driven by deteriorating fundamentals and higher yields driven by stronger growth and investment demand. The next catalysts will be Federal Reserve (Fed) Chair Kevin Warsh at Jackson Hole, as markets look for clues on how the Fed views that balance, and the administration’s expected new economic measures against Iran. Both could create volatility, but neither changes the broader point: the market remains supported by strong fundamentals, even as the sources of risk, and opportunity, continue to shift.

China's AI Models Are Catching up to Leading U.S. Models

What is happening: The competitive gap between China’s leading open-weight AI models and the best proprietary U.S. systems has narrowed significantly. Moonshot AI’s Kimi K3, Alibaba’s Qwen 3.8, and DeepSeek’s V4 represent a new generation of very large and increasingly efficient models designed for reasoning, coding, and autonomous agentic work. According to Livebench.ai, the overall benchmark scores of Kimi K3 and Qwen 3.8 are now within four points of Claude Fable 5, the highest-scoring U.S. model. While four points is not an insignificant margin, these two Chinese models now have higher overall scores than the best models of several U.S. companies, including Grok 4.6 (SpaceX), Gemini 3.7 (Alphabet), and Muse Spark 1.2 (Meta). They are even outperforming Claude Fable 5 in certain subcategories, such as Kimi K3 for reasoning and Qwen 3.8 for agentic coding.

Equally important, the competition is shifting from who has the smartest model toward who has the best cost efficiency, and that’s an area where the Chinese models have a big advantage. Kimi K3 and Qwen 3.8 cost approximately $0.30 per successful task, compared to about $1.50 for Claude Fable 5 and $0.50 for GPT-5.6 Sol. This has coincided with U.S. companies recently reducing prices on some of their non-leading models. For example, OpenAI cut GPT-5.6 Luna’s prices by 80%, and Anthropic cut Claude Opus’ prices by 67%. The U.S. companies attributed these changes partly to their own efficiency improvements and product architecture, but price competition from cheaper Chinese rivals likely influenced the magnitude of the price cuts. 

Why it matters: A clear implication of recent Chinese AI model advancements is that the model itself is becoming a more competitive and less economically scarce layer of the AI value chain. If comparable intelligence can increasingly be obtained from several different models, pricing power and excess returns at standalone model developers may be harder to sustain. At the same time, cheaper intelligence should be highly favorable for companies that successfully incorporate AI, potentially improving margins and making previously uneconomic projects more viable.

Major U.S. hyperscalers, such as Alphabet, Amazon, Meta and Microsoft, have continued to expand infrastructure aggressively in response to both strong customer demand and current capacity constraints. If Chinese and other open-weight models can achieve similar performance with less compute or substantially lower prices, the revenue generated by each incremental dollar of frontier-training infrastructure could be lower than the hyperscalers originally anticipated. 

However, that does not necessarily mean the massive capex will be wasted. Greater model efficiency reduces the amount of compute required for any individual task. Therefore, the ultimate demand for GPUs, high-bandwidth memory, and other AI infrastructure components depends on whether usage grows faster than price per task falls. Chinese competition could accelerate that process by pushing the entire industry down the cost curve, likely resulting in lower margins but not necessarily lower total revenues. At the same time, it’s important to note that benchmark score parity does not automatically mean parity in reliability and security. Regulatory and data security concerns may also limit Chinese model adoption in sensitive Western workloads and preserve a premium for trusted U.S. providers. 

From an investment standpoint, this development supports diversification across the AI value chain rather than assuming the largest share of economics will accrue to whichever company owns the leading model. This is consistent with Bessemer’s current portfolio positioning. We are bullish on semiconductors and power infrastructure because we believe these areas will continue to be key bottlenecks in the expansion of AI adoption. While we maintain sizable positions in the hyperscalers, we believe lower costs should increase aggregate demand and also have broad exposure to networking, memory, storage, and electricity companies. 

What’s Behind the Backup in Bond Yields

What is happening: Longer-dated bond yields are rising, with 30-year Treasury yields rising by more than 40 basis points since late June to 5.3%, the highest levels since 2007. Over the same period,10-year Treasury yields are up more than 30 basis points to 4.7%, which is just below recent highs recorded in 2023. The rise prompted a response from the U.S. Treasury, which doubled the size of buybacks of 10- to 30-year debt. This is not a durable solution, but it does send a message to bond markets that intervention remains a tool and may signal Treasury’s readiness to shift new issuance to shorter tenors. 

The global nature of the backup, with similar increases in the U.K., Europe, and Japan, as well as the underperformance of the 30-year sector, suggests that fiscal uncertainty is a factor. We also find that a dominant theme is the AI boom. AI credit supply has increased sharply this year, up 27% year over year to $1.23 trillion in the first half of 2026. Hyperscaler capex is expected to increase into 2027 and beyond, further ramping up external funding needs. While longer-dated investment-grade (IG) credit does not compete directly with Treasuries due to ratings differences, it can be relatively attractive given fiscal debt outlooks. 

The AI boom is also driving a global cyclical upturn, echoed by the rise in manufacturing purchasing managers’ indexes (PMIs), where in the U.S. the index has seen its highest rate of change since 2017 outside of the pandemic. AI also underpins long-term growth expectations to the extent that the capex boom raises productivity. Notably, U.S. nominal GDP accelerated to a 6.5% year over year pace in Q2 on broadening business investment. A repricing in long-term growth prospects can help explain rising yields since the start of the year. 

Digging further, the rise in nominal yields also reflects a higher term premium — the additional yield investors require to hold longer-dated bonds. Factors affecting term premium include inflation uncertainty, central bank credibility, and debt issuance. Elevated oil prices and ongoing geopolitical uncertainty are driving concerns over recurring supply-related price shocks while Fed Chair Warsh’s embrace of limited guidance and support for a smaller balance sheet introduces a new layer of uncertainty over both inflation and central bank purchases. While these factors contribute to inflation uncertainty, a loss in Fed credibility over the dual mandate does not appear to be at stake, as longer-term inflation expectations would have also increased, which is not the case.

Why it matters: The pace and extent of the recent rise in long-term yields is notable but still well short of the bond market selloffs in 2022 and 2023 and has been met with new records in equities, outperformance of long-duration assets like small caps, tight corporate credit spreads, and subdued market volatility. What drives an overall tightening in financial conditions tends to be the pace of the increase in yields rather than the absolute levels of yields. 

A further disorderly increase in yields would be unwelcome but is not our base case. While debt-to-GDP ratios in the U.S. and other major developed countries remain on a rising trend, the trajectory has not changed significantly. The U.S. Treasury already leans heavily on bill issuance and may be open to cutting auction sizes in the long end. 

Cyclical drivers are just as important, if not more so. While private demand was remarkably resilient in Q2, growth is set to moderate as the temporary boost from OBBBA tax relief and the World Cup wanes and subdued real income growth caps spending. Slowing inflation through year-end will also help lower inflation anxiety. 

Structurally, however, there is a case for long-term yields to remain elevated. Many of the factors that previously kept term premiums subdued have now reversed. These include adverse supply shocks, geopolitical uncertainty, falling sources of demand from central banks and pensions, and debt issuance. Furthermore, nominal GDP growth is stronger amid procyclical fiscal policy, AI, and capex. 

Bessemer portfolios remain overweight equities relative to fixed income, which has contributed positively to performance year-to-date. A further or sharp move higher in bond yields could tighten financial conditions but also offer tactical opportunities in our portfolios, which we will watch for closely.