Sensitive data lives across different repositories, which may be known or unknown. As a result, oftentimes, misconfigurations go unnoticed, and security alerts from different platforms rarely intersect with any data-centric risks. DSPM solutions fill this void by seamlessly integrating with different security platforms.

DSPM vs. CSPM

DSPM solutions provide data-centric security. It protects sensitive data across hybrid, multicloud environments. It can automatically discover and mitigate vulnerabilities associated with privacy and compliance. Cloud security posture management (CSPM), on the contrary, helps organizations secure their cloud infrastructures, such as VMs, IAM roles, and buckets, from risks like misconfigurations and general compliance issues. DSPM can help organizations to prioritize the alerts generated by CSPM tools based on sensitive data.

DSPM vs. CNAPP

Cloud-Native Application Protection Platforms (CNAPP) are known for unifying multiple security capabilities to provide a comprehensive approach to cloud application security. For instance, a CNAPP platform would offer CSPM, DSPM, runtime security, and workload protection. DSPM can help CNAPP narrow the lens to the data itself. For instance, a CNAPP platform identifies a container as vulnerable. DSPM will see if the container contains sensitive data or not. If it contains sensitive data, DSPM will help prioritize it by first spotting if the data is encrypted or not, and lock it for automated remediation.

DSPM vs. DLP

DSPM solutions often come with DLP-like capabilities, including but not limited to encryption and access controls. In fact, the solution may also go beyond those capabilities to offer functionalities like data breach analysis, mapping configurations, and remediation of misconfigurations across multicloud environments. Typically, DLP usually focuses on the prevention of unauthorized data sharing and exfiltration attacks in a single environment. Integrated DSPM solutions go beyond the identification of data leakage issues, offering insights into data residency, access insights, and automated remediation.

DSPM vs. SIEM

Security Information and Event Management (SIEM) analyzes event data to discover anomalies and suspicious behavior. These tools specialize in log collection, correlation, and threat detection. DSPM helps protect sensitive data at the source by ensuring robust access controls and policies, preventing threats before they materialize.

The simplest thing to outperform current AWS Inferentia 2 is to implement neural model with weigths in metal (see https://blog.azolotarev.com/billion-dollar-idea-ai)

Nowadays, Taalas is beating nVidia's 230 tokens per second with 16,960 tks/s generation speed by implementing Llama 3.1 8B model in metal.

By securing $219M in investment and current valuation of $700M.

Good job!

• For models training
• For response production (Inference)
  

With AI getting momentum and being widespread, the inference chips demand is going to peak.

To win the race, let's find the biggest bottleneck.

Building a data center is 1-2 years challenge. Current network capacity is good enough for text generation, don't expect a spike here. Power plant to supply a data center with energy is a venture for 3-6 years depends on type of a plant.

Therefore, is there is a chipset solving for power - it will be the next winner in the race.

Amazon and X is moving towards the right direction.

The simplest thing to outperform current AWS Inferentia 2 is to implement neural model with weigths in metal. Downside is that you cannot retrain it without physical rewiring.

However if you implement a model in bare metal, it will x100 the perfomance with the same TDP.

Currently there are FPGA (Field Programmable Gate Arrays) being extensively developed, which is middle grounds between the bare metal approach and existing AI specialised chipsets. Not the simplest solution though. But it is the most perspective direction is we focus on solving for Performance-per-Watt.

P.S. the version of this post for fellow colleagues on the board is here: https://blog.azolotarev.com/billion-dollar-idea-ai-board-version

The AI revolution is currently colliding with the laws of physics. While software scales instantly, the infrastructure required does not.

• The Infrastructure Lag: Building a data center takes 1–2 years, but securing the power plant to run it takes 3–6 years. Your growth is currently limited by the local utility company’s construction schedule.
• The EBITDA Drain: Specialized chips like the NVIDIA Blackwell consume between 700–1,000W per unit. In a high-volume inference environment, this energy "tax" becomes a permanent drag on operational margins.
• The Performance Gap: Standard architectures are designed for flexibility (training), making them inherently inefficient for the repetitive task of response production (inference).

2. The Solution: Performance-per-Watt - The "Efficiency First" Architecture

To win the inference race, we must pivot from "Maximum Power" to "Maximum Efficiency."

• The New Metric: The winner will be the chipset with the highest Performance-per-Watt.
• Current Leaders: As seen in the data, AWS Inferentia2 ($4.67–7.0$ TOPS/W) and Groq LPU ($2.67–3.33$ TOPS/W) are already outperforming traditional GPUs by focusing on specialized inference paths.
• The Efficiency Moonshot: Moving toward "Weights in Metal" — implementing neural models directly into bare metal. By hard-wiring the model into the silicon, we can achieve a theoretical $100\times$ performance increase while maintaining the same Thermal Design Power (TDP).
• The Practical Step: Moving toward FPGA (Field Programmable Gate Arrays) — implementing neural models in chips which might be reconfigured. By soft-wiring the model into the silicon, we can achieve a theoretical $10\times$ performance increase while maintaining the same Thermal Design Power (TDP).

3. Implementation Strategy: Hard-Wired Advantage

We shift from being a "Compute Consumer" to a "Compute Architect."

• Identify Static Logic: Isolate high-volume AI tasks that do not require weekly retraining (e.g., base-level translation, sentiment analysis, or security filtering).
• Deploy in FPGA: For these static tasks, move away from programmable GPUs and implement configurable neural models.
• Bypass the Grid Bottleneck: Because these chips require significantly less power for the same output, you can deploy $10\times$ the compute capacity within your existing power footprint, effectively finding capacity that doesn't exist for your competitors.

4. The Moat: Structural Margin Superiority

This is not just a technical upgrade; it is a financial fortification.

• Unmatchable Cost-to-Serve: Ultimately by achieving $100\times$ efficiency, your cost per inference becomes a fraction of a competitor's. They cannot compete on price without destroying their own EBITDA.
• Infrastructure Lock-in: Once your models are "Weights in Metal," your hardware is optimized perfectly for your software. This creates a high barrier to entry for competitors who rely on general-purpose high-TDP hardware.
• Grid Resilience: While competitors wait years for new power plants to come online to scale their energy-hungry clusters, you scale horizontally within the power limits you already have.

1. The Problem

The current EV industry is trapped in the "Monolithic Battery" paradigm. By building cars around a single, massive 1,000lb battery pack, manufacturers have created a massive technical deficit:

• The Charging Bottleneck: Even with "Fast Charging," 20–40 minutes is the best user can get.
• The Stranded Asset Risk: If a single cell-group fails in a monolithic pack, the entire $15k–$20k battery pack is compromised.
• Infrastructure Failure: There is simply not enough Fast Charging infrastructure. Slow refill doesn't work for people living in apartment buildings.

2. The Solution: The Energy Blade

We need to stop viewing the battery as a "part" and start viewing it as a "Standardized Unit of Energy." The Product: A modular, hot-swappable "Energy Blade" unit.

• The Logic: Instead of parallel drainage of one massive pack, the vehicle utilizes energy unit-by-unit.
• The Ecosystem:
• 1 Unit Blade: Powers e-bikes and electric scooter.
• 2 Unit Blades: Mounted under each seat and 2 items in the trunk.
• 4 Unit Blades: 3 items of 4-unit size mounted under the car.

This setup provides incredible UX benefits with six 2-unit-blades available for a hot swap. With current technological level, the energy density per each 2-unit-blade is approximately 50 km. Therefore, each car has 300 km available for a hot swap and 300 km reserve mounted under the car.

The biggest tech shift is to enable energy utilisation unit-by-unit.

The city use case scenario: You just drove to a school, office and a market with total mileage of 40 km. In the evening, you're pikcing up a 2-unit-blade upstairs and plug it to regular outlet. Next morning, your car is at 100% again.

Long haul scenario: You just drove 300 km, make a stop at a petrol station and swap every replaceable unit with a fresh now. 5 mins and you are back on the road.

This solves the range problem not by adding "more capacity," but by enabling instant unit by unit swapability. A car can be "recharged" in 60 seconds by swapping two depleted blades from under the seat at a kiosk.

3. The Implementation Strategy

This is not an idea for a startup; it is a play for a Vertical Champion. Champions like Tesla or BYD have the manufacturing gravity to enforce a new standard.

• Vertical Integration: To win, the manufacturer must control the car, the blade, and the supply chain for swap-kiosks.
• Manufacturing Efficiency: Standardizing one "Blade" size across the entire product line (from the Model 2 to the Semi) creates massive economies of scale that competitors with fragmented battery sizes can never match.

4. The Moat

The "Energy Blade" isn't just hardware; it is an Interoperable Energy Standard.

• Standardization as a Barrier: Once a "Blade" size is standardized, the manufacturer becomes the de facto utility provider for all mobility.
• Residual Value: Used car valuations currently tank because of battery degradation. With "Blades," the battery is no longer a permanent part of the car—it is a liquid asset that is constantly cycled and refreshed in the network.
• The P&L Win: By moving battery costs from CapEx (buying the whole battery) to OpEx (subscribing to a blade-swap network), you lower the entry price of EVs, capturing the massive "budget-conscious" segment of the market.

Whenever challenge is presented, tech person is focused on solving it with data we have.

Business person is focused on the data we need.

1. The Critical Failure (The Problem)

The aviation industry is currently operating on "Security Theater" regarding Lithium-ion (Li-ion) fires.

• The Technical Deficit: Handheld Halon extinguishers are useless against Thermal Runaway. Li-ion fires are self-oxygenating; you can't "smother" them.
• The Scale of Danger: While power banks over 27,000 mAh (100Wh) are restricted, the ones under that limit still carry enough energy to cause a catastrophic event.
• The Financial Ghost: A single battery fire can cause an emergency diversion (Cost: $50k–$250k) or, in the worst case, a total hull loss. Look at UPS Flight 6—a billion-dollar disaster caused by battery ignition.
• The "Tesla" Metric: It takes roughly 10x to 40x more water to extinguish a lithium fire than a gasoline fire. On an aircraft, water is not a resource.

2. The Solution: The Vulcan Sleeve

We aren't building a better extinguisher; we are building a containment moat.

Product: A vacuum-packed, flexible, foldable containment box made of high-grade vermiculite-coated fiberglass and proprietary fire-suppressant layers.

• The Innovation: Unlike existing "burn bags" which are bulky, the Vulcan Sleeve is vacuum-sealed to the size of a thin laptop sleeve.
• The Physics: It uses a "starve and isolate" approach. When a device smokes, it is dropped into the sleeve and sealed. The material withstands 1,100°C+ and contains the off-gassing, preventing the fire from spreading to the airframe or cabin oxygen.
• Feasibility: The materials exist (Zetex or Vermiculite-coated fiberglass); the engineering challenge is the "form factor" for cockpit and cabin integration.

3. The Market Opportunity (TAM)

The aviation industry doesn't buy "safety gadgets"—it buys Risk Mitigation.

• Total Addressable Market (TAM): There are currently 35,550 commercial aircraft globally, projected to hit 50,000 by 2044.
• The Unit Economics: * Fleet-wide mandate: 2 units per cockpit, 4 per cabin = 6 units per aircraft.
• Total initial unit demand: ~213,000 units.
• The "Insurance" Angle: If a Vulcan Sleeve prevents just one emergency diversion, it pays for an entire airline’s fleet-wide implementation 10 times over.

4. The Moat

• Certification as a Barrier: The "secret sauce" isn't just the fabric; it’s the FAA/EASA certification. Once a product is "baked into" airline safety protocols, the switching costs for competitors are massive.
• Data Play: Each sleeve has a "shelf-life" and compliance, moving from a hardware sale to a Safety-as-a-Service recurring revenue model.