The Machine Gaze: How Trevor Paglen Maps the Erosion of Visual Reality
Artist Trevor Paglen examines how computer vision and generative media have transformed images into operational tools of power and control.

As the race to harness artificial intelligence reaches a fever pitch, the eye-watering costs associated with training models of ever-expanding complexity have become a significant hurdle for many in the industry.
Enter Phison, a company that has long been a stalwart in the realm of portable SSDs, now poised to revolutionize the landscape of AI training with a bold new approach that promises to slash these costs dramatically.
AI models have been ballooning in size at a pace that seems to mirror the fantastical leaps of technology in science fiction.
To put it in perspective, Llama 2 boasted 69 billion parameters just last year, and the ante has been upped with Llama 3.1’s 405 billion parameters.
The imminent arrival of DeepSeek R3’s 671 billion parameters hints at a trillion-parameter model on the horizon, possibly even before 2026.
This exponential growth demands unprecedented computational muscle, traditionally provided by GPUs, but at a steep price.
Phison’s groundbreaking strategy pivots on shifting a substantial portion of the processing workload from GPUs to SSDs using their aiDAPTIV+ software.
This ingenious move could potentially reduce the training costs of these mammoth models from an astronomical $3 million to a mere $100,000—a staggering 97% savings that could democratize access to cutting-edge AI.
How does this magic happen?
By integrating high-performance SSDs with NVIDIA’s GH200 Superchip, Phison’s solution efficiently reallocates tasks to SSDs that were once the exclusive domain of GPUs.
This shift not only trims the number of GPUs required but also leverages the rapid data access speeds of SSDs to streamline operations.
It’s a mix of technological wizardry and economic pragmatism that could redefine the AI training playbook.
The implications of such a cost reduction are profound.
With potentially more players able to enter the arena, we might witness an explosion of innovation and diversity in AI applications.
Smaller companies, previously sidelined by prohibitive expenses, can now dream of developing models that rival those from more financially robust competitors.
It’s a democratization of AI that could foster a more dynamic and inclusive tech ecosystem.
Phison’s collaboration with Maingear to launch AI workstations powered by Intel Xeon W7-3455 CPUs underscores its commitment to reshaping AI hardware.
This partnership is a harbinger of what could be a broader industry shift, where SSDs take center stage in AI processing, leaving traditional HDDs for long-term data storage.
While DeepSeek made waves earlier this year by unveiling a model trained with 95% fewer chips for $6 million, Phison’s approach suggests an even more radical reimagining of AI training economics.
As the competition heats up, Phison’s SSD strategy might just be the ace up its sleeve, offering a glimpse into a future where AI’s potential is limited not by cost, but by imagination.
In this unfolding narrative of technological evolution, Phison stands not just as an innovator but as a potential catalyst for change in the AI industry.
As we edge closer to the trillion-parameter milestone, the question is no longer about if we’ll get there, but rather who will lead the charge and how accessible this frontier will be to all eager to explore it.
Artist Trevor Paglen examines how computer vision and generative media have transformed images into operational tools of power and control.
Abhishek Saxena’s work at Sentient targets the gap where open-source AI keeps losing: not capability, but economics—and his answer is infrastructure that automatically pays builders, maintainers, and evaluators every time their artifact is used, enforced by smart contracts rather than legal goodwill. By combining cryptographic fingerprinting, on-chain attribution, and grant funding with no equity attached, Sentient is building the coordination layer that would make open-source development financially rational enough to compete with a corporate salary.
As AI-generated content floods the digital landscape, Substack is empowering readers to verify the human origins of the newsletters they consume.