The AI landscape continues to broaden beyond Nvidia as alternative accelerators become more competitive, hyperscaler spending approaches physical limits, and cheaper open-weight models capture a growing share of usage.

Google’s TPUs Challenge Nvidia on Cost

For the first time, independent benchmarks show Google’s TPUv7 “Ironwood” competing directly with Nvidia’s latest hardware on inference economics. SemiAnalysis found Ironwood serving the same open-weight model at $0.181 per million tokens, compared with $0.222 on Nvidia’s B200 and $0.276 on the B300 - up to a 50% advantage in performance per dollar.

Google is also rewriting its TPU software to run natively in PyTorch, potentially reducing one of Nvidia’s biggest advantages: software ecosystem and developer lock-in. Nvidia still leads in some workloads, including FP4, but the competitive landscape for AI compute is expanding.

AI Infrastructure Faces a 2028 Ceiling

AI infrastructure spending is still accelerating, but the pace may begin to slow as physical constraints become harder to overcome. Morgan Stanley forecasts hyperscaler capital spending reaching $1.47 trillion in 2027, up roughly 60%,before growing another 12% in 2028.

The bank expects AI compute capacity to increase from roughly 35GW in 2025 to 145GW by 2028, with chips, power, labor and materials limiting how quickly new capacity can come online. That could shift attention further up the stack toward companies benefiting from the growing volume of AI inference.

AMD Leads a Broader Compute Rally

AMD was the best-performing S&P500¹ technology stock of the week, gaining 13.15%, after Piper Sandler initiated coverage with an Overweight rating and a $600 price target.The firm also rated Nvidia, Broadcom and Arm Overweight, highlighting the broader opportunity across AI infrastructure. (Source: Seeking Alpha, 10-11 Sep 2026)

AMD also said its Helios rack and MI450 accelerator are beginning to ship this quarter, marking a move toward selling complete AI systems rather than individual chips.

Open-Weight Models Continue to GainGround

DeepSeek’s latest V4.1-Flash highlights the growing price advantage of open-weight models.The 552-billion-parameter model is priced at $0.60 per million output tokens, compared with $20 for OpenAI’s GPT-5.6 Sol and $25 for Anthropic’s Claude Opus5.

That price gap is translating into usage. Chinese open models accounted for more than 60% of tokens consumed across the top 50 models by August, up from less than 20% in late 2025. Open-weight models now represent 61.5% of routed tokens in the latest smoothed week.

AI Pricing Is Shifting - but Not Always for the Reasons It Appears

GPU pricing remains relatively firm overall, with the latest decline concentrated in GCP A100-80GB pricing. The average across tracked GPUs is down just 1.7% since August 3, while AWS and RunPod A100 quotes have actually increased.

Model pricing tells a similar story: median prices can move simply because the mix of available models changes. Surviving frontier and small-model quotes were largely unchanged between comparison dates, suggesting the rebound in category medians does not necessarily mean existing workloads are becoming more expensive.

Meanwhile, recent open-model usage growth has been concentrated in just a few model versions, highlighting how routing decisions and cheap or free access can quickly shift token volumes.

Looking Ahead

AI adoption continues to expand, but the competitive landscape is becoming more complex. Alternative accelerators, infrastructure constraints, and lower-cost open-weight models are all reshaping the economics of AI compute and inference.

----------------------------------------------------------------------------------------

As AI continues to reshape computing, software, and the broader economy, identifying the companies positioned to benefit from these long-term trends is becoming increasingly important. LRNZ seeks to provide concentrated exposure to companies at the forefront of AI and deep learning, focusing on businesses where advanced AI applications may create a distinct competitive advantage.

Learn more about the TrueShares Technology, AI & Deep Learning ETF(LRNZ)

Please click on the link to view the funds top ten holdings. Fund holdings and allocations are subject to change and should not be considered recommendations to buy or sell any security.

1. TheS&P 500 (Standard & Poor's 500) is a stock market index that tracks the performance of 500 of the largest publicly traded companies in the United States

TRUE792

Please click on the link to view the funds top ten holdings. Fund holdings and allocations are subject to change and should not be considered recommendations to buy or sell any security.

Before investing, carefully consider theTrueShares ETFs investment objectives, risks, charges and expenses. Specific information about TrueShares is contained in the prospectus and a summary prospectus, copies of which may be obtained by visiting www.www.true-shares.com. Read the prospectus carefully before you invest.

The fund is distributed by Paralel Distributors LLC,Member FINRA. Paralel is not affiliated with TrueMark Investments, LLC and BlackHill Capital Management.

For informational purposes only. The information provided does not constitute investment advice or a recommendation to buy, sell, or hold any security. References to specific securities are provided solely for illustrative purposes.  Opinions and forecasts are current only as of the date indicated, are subject to change without notice, and are not guarantees of future results. Forward-looking statements involve risks and uncertainties, and actual results may differ materially.

‍