A recent discussion at ARK Invest points out that while Google's Gemini 3 has successfully returned to the ranks of top models, the breakthrough of this update is not in language capabilities, but rather in the new image and information generation model, "Nano Banana Pro." The ARK team believes that this update signifies Google's AI technology is back at the forefront, but it still faces challenges in user adoption and commercialization strategies.
In response to criticisms regarding the Gemini 3, Google emphasized that the law of AI expansion remains valid.
Fangzhou pointed out that Google's newly released Gemini 3 ranked highly in multiple benchmark tests, thus refuting external doubts about "diminishing returns for large models." Google's engineering team stated that scaling still brings visible benefits, showing that training volume can be significantly increased.
In terms of practical experience, Downing, Research Director at Ark AI, stated that Gemini 3 is on par with ChatGPT (5.1 Thinking) in processing most tasks, with each having its own advantages. He believes that Google's integration of memory and personalization features into Gemini is the key to the significant improvement in product maturity.
Nano Banana Pro is the real breakthrough.
Compared to language models, Ark believes that Nano Banana Pro's performance is more revolutionary. This model can condense large amounts of text into structured information such as images, presentations, and flowcharts, and can even correctly output text-to-image conversions, a hurdle that most models have struggled to overcome in the past.
ARK points out that this capability is highly valuable for content creation, marketing materials, and visual information work, demonstrating Google's clear lead in image generation and visual understanding.
Google has a first-mover advantage in integrating language and imagery into its AI architecture.
Winton, chief futurist at Ark, believes that the future architecture of AI will integrate language reasoning, image generation, and long-term memory systems, and Google has made moves in all three technologies, including the previously released Titans memory architecture.
The Gemini 3 and Nano Banana Pro are seen as key components of Google's next-generation AI architecture.
Google and OpenAI clash head-on, memory technology becomes the new battleground.
Fangzhou points out that cross-chat memory is becoming a key factor in platform user retention. ChatGPT's cross-chat memory feature, once launched, can remember user preferences and context, significantly improving user engagement.
Google's simultaneous adoption of a similar design in Gemini indicates that both companies see memory as the next competitive focus. However, Fangzhou believes that forgetting long conversations and the inconvenience of transferring historical content remain challenges that the entire industry needs to overcome together.
TPU vs. GB200: Both models choose NVIDIA as the winner.
It is worth noting that when Ark asked Gemini 3 and ChatGPT about the "performance/power consumption difference between Google TPU v7 and NVIDIA GB200", both gave the same conclusion: NVIDIA still has the upper hand in terms of performance/power consumption.
Fangzhou points out that while Google has a cost advantage in capital expenditure due to its self-developed TPU, AI training and inference are rapidly being limited by "electricity." As electricity becomes a new bottleneck for generative AI, the efficiency/power consumption ratio will directly determine the number of tokens that a model can output per watt of power, and will also affect overall operational efficiency and revenue ceiling.
If YouTube Premium were to be bundled with Gemini, Google might rewrite the game.
Fangzhou believes that if Google were to make YouTube Premium and Gemini Pro a single subscription plan (at $19.99 per month), it would put significant pressure on the market.
AI companies like OpenAI rely on subscription revenue and cannot withstand price wars, while Google has diverse revenue streams such as search, advertising, and cloud services, and can leverage its content ecosystem to significantly expand its competitiveness on the consumer side.
Google's biggest weakness lies in user adoption; its adoption rate is lower than ChatGPT's.
Despite the recognition of Google's technological advancement, Fangzhou also admitted that its biggest weakness remains on the user side. Fangzhou cited US app usage data, pointing out that ChatGPT leads by a landslide with 99%, while Gemini only has 1%.
Even on a global scale, Gemini's usage lags significantly behind, not only failing to catch up with ChatGPT but also being noticeably ahead of xAI's Grok in terms of usage minutes. Fangzhou points out that this demonstrates that competition among AI platforms isn't solely about model capabilities; product accessibility, marketing efforts, and user habits are the key variables determining long-term adoption rates.
This article, "Ark Investment: Google Nano Banana Pro is quite excellent, but its adoption rate is still lower than ChatGPT," first appeared on ABMedia .
