Joined April 2024
- Tweets 1,219
- Following 2,009
- Followers 5,498
- Likes 3,647
286 Photos and videos
Jun 13
What zero egress fees means in practice?
→ Your data team runs analytics without checking the billing dashboard first
→ Engineers replicate across regions for resilience, not cost justification
→ DR runbook says test weekly, and you actually do
→ Training runs use the full dataset, not the affordable subset
Pricing isn't just economics. It's architecture.
1
4
163
Jun 12
If it speaks S3, it already works with Akave.
No code changes.
No migration project.
No parallel-run budget.
Your pipelines stay identical. Your egress bill doesn't.
One change. Every experiment after it is fully yours.
x.com/akavenetwork/status/20…
4
113
Jun 12
The 3 vectors of cloud vendor lock-in, and most migration plans only solve one:
1. Economic, egress fees make leaving cost-prohibitive
2. Operational, your toolchain is API-coupled to their stack
3. Legal, contract data portability clauses favor the vendor
Most teams fix #2. Discover #1 and #3 at renewal.
1
63
Jun 10
The infrastructure decision that breaks AI teams at scale isn't compute.
It's not even the model.
It's the storage bill that arrives 6 months after you went from 'experimenting' to 'production.' The number is always a surprise. It shouldn't be.
1
5
129
Jun 9
S3 launched in 2006.
Designed for: small datasets, rare experiments, no AI pipelines, two-PowerPoint compliance.
In 2026, we're running petabyte-scale AI training on 20-year-old storage assumptions.
That's the infrastructure debt nobody's pricing into their roadmap.
5
95
Jun 9
Data in the EU ≠ data under EU jurisdiction.
That distinction matters to your legal team.
Europe has the world's strictest data regulation.
And roughly 4% of global cloud capacity is European-owned.
The other 96%? Subject to the CLOUD Act, which allows US authorities to compel data access regardless of where the servers sit.
2
5
143
Jun 8
Under the EU AI Act, high-risk AI systems must maintain verifiable records of training data provenance.
Verifiable is the operative word.
If your training data lives in S3, and AWS can modify the metadata, your compliance documentation is a claim, not evidence.
Your auditor will notice. Opposing counsel will too.
2
5
110
Jun 8
If your AI model's biggest security risk isn't the model.
It's the training data stored in infrastructure where the provider holds root access.
You encrypted the weights.
The dataset that produced them lives in someone else's basement. Think about that.
1
3
95
Jun 6
The data sovereignty debate has been asking the wrong question.
Everyone argues about geography. Is your data in the EU?
The right question is custody,
Who can read it, modify it, or delete it, and can you prove they haven't?
Location is legal. Cryptographic proof is evidence.
1
1
5
136
Jun 5
The cloud providers that charge egress aren't pricing storage.
They're pricing exit.
You don't pay to keep data.
You pay to leave with it.
That's not infrastructure economics.
That's a retention mechanism with a bill attached.
1
5
125
Jun 4
The infrastructure checklist before you sign (save this)...
If answers aren't in writing: that's your answer.
7
115
Jun 3
The AI infrastructure stack in 2026:
The teams winning aren't the ones with the best GPU access.
They're the ones with reproducible, sovereign, cost-predictable data infrastructure that compounds over every experiment.
2
7
137
Jun 2
Most sky events pass overhead with no verified record.
No tamper-proof timestamp.
No audit trail for the observation data.
It's undocumented because no one built the infrastructure to capture it continuously and verify the record.
@Skymapperspace built SkySphere - AI-powered all-sky observatory network that monitors and captures sky events continuously.
We are backing it.
Our piece:
Providing the immutable storage infrastructure that ensures every observation is preserved.
If this matters to you, back it. Link in comments 👇
3
1
9
500
Jun 1
S3 compatibility is the entrance exam, not the job interview.
Every storage vendor passes it.
None of them pass what comes next: cost under iteration, throughput under concurrency, proof under audit.
7
107