Advertisement tailored for each single consumer.
Advertisement is heavily reliant on distribution optimization. Customer targeting is now very sophisticated; the ads one consumes are a function of that person's preference, socio-economic level, and culture, to list a few. However, an iPhone ad should illustrate how great the camera quality is to a photographer while the processor speed is the better pitch to a gamer. The ad sellers already have all the crucial personal data for one-to-one ad personalization, but are bottlenecked by the static nature of ads. With the help of generative AI and powerful inference chips, it is now possible to morph visuals in real time. The Daily Prophet rewrites itself for every reader. This not only opens up a new frontier for ad giants like Google and Meta and chipmakers like Apple and Nvidia, but also affects small ad businesses and all the customers who actually need the biggest adjustment.
This grandiose proposal is not a future dream but is happening in real time, as the ad industry is evolving fastest in its history. Neural Processing Units (NPUs) are already in all of our pockets, as Apple's Neural Engine, Qualcomm's Hexagon, and Google's TPUs provide the inference needed to run the generative ads. A big pretrained model sits in the hardware. Low-Rank Adapters (LoRA) carry the specialized model weights for specialized content generation. Real-time local text and image generation is already here. However, the real breakthrough will happen as soon as online video generation lands. As you finish watching a reel, the social media agent decides which ad to show you; the ad's LoRA augments the pretrained model inside the user's hardware, takes inputs like user's mood and the weather, and generates a video shaped to the user's needs and preferences.
It also opens the door for combinational ads, where promoting more than one product in a single frame becomes a reality. Netflix can combine a few different movie trailers generated specifically for the target consumer. The density of information per ad will go up, adjusted to consumer's attention capacity. This also opens up a new frontier for small businesses that cannot afford a full-length spot but are open to sharing space with others. Cowboy Mike, wearing boots from Thursday Boot Co., walks into the local bar the Desert Flow and orders a Lone Star beer. The current issue is that static combinational ads will fall flat for most of the users, as a specific combination only appeals to a small slice of the demographic. However, online generation of a personalized ad bypasses this bottleneck completely. A local coffee shop can bid on the coffee cup in a blockbuster movie, but only when that movie is streamed by users within a 5-mile radius of the shop.
This has a few interesting consequences, for instance the GPUfication of every electronic device that has information-output capabilities. The death of the "dumb" display. The Billboard insurance lady reminds us of tire chains in snowy weather while warning us to slow down during heavy rain, branding the company as the most reliable advisor. That requires information as input, neural chips as processors, and memory to hold the models. These technologies already exist; therefore, justifying the cost might be the only barrier right now. Joi from Blade Runner 2049 is going to debut very soon in real life.
What do consumers do under this constant barrage of personalized ads? To resist this much personalization is really very difficult. The savior might be artificial agents that negotiate, on the consumer's behalf, how ads get presented to them. Cloud-based agentic tools such as Perplexity already allow agentic shopping; but the user's personal preference data becomes vulnerable. The solution is owning the personal agent locally to intercept incoming outreach. The personal agent lives on the chip, so it is not controlled or manipulated by the ad sellers, as that would defeat the purpose. Online shopping is already facing agentic waves, though generative ads will make agentic shopping mandatory.
As the big software companies themselves are selling ads, it's likely that chipmakers or new wave cybersecurity companies will produce these personal agents, confined in a single device. The idea is to increase the gap between data and the ad sellers for maximum protection. Consequently, the ultimate winners of this paradigm shift won't just be the creative agencies or the ad giants, given they already own the lion's share of the world's ad business; the big winners will be the chipmakers who can provide the ironclad, local security architectures necessary to keep our personal preference data locked in silicon.