Modal CEO: Who Wins the Inference Market | Has Open-Source LLM Inference Peaked? | What Happens to the Cost of Compute | The Real Limits of Post-Training and RL

Erik Bernhardsson is the Co-Founder and CEO of Modal, the AI infrastructure platform for training models, running inference and powering AI agents. Modal has raised over $1.2 billion, including a recent $750 million Series D led by Accel at a $15.75BN valuation. The company has surpassed $1BN in ARR, growing more than 15x year-on-year. Previously, Erik built Spotify's original music recommendation system.

In Today's Episode:

• Why wait until your mid-30s to start a company?

• Is founding a startup really like "eating glass"?

• How to Raise $1BN with No Pitch Deck

• The danger of raising too much venture capital

• How much cash is Modal burning on GPUs?

• Do early-stage startups actually have moats?

• Who wins the inference market in the next 5 years?

• Are 90% software gross margins dead in AI?

• The "open-air drug market" for GPUs

• Will GPU hardware prices ever collapse?

• Are we about to overbuild data centers?

• Can custom silicon dethrone NVIDIA's monopoly?

• Is CUDA really the moat everyone claims it is?

• Are Neo Cloud providers partners or mortal enemies?

• Nebius vs. CoreWeave: who built the better business?

• Open-source vs. frontier models: which takes the lead?

• Can Europe ever build a competitive frontier model?

• The bull case for a $100B verticalized model company

• Modal's biggest resource allocation mistake

• Has open-source LLM inference already peaked?

• The real limits of post-training and RL

• How much does Modal spend every month on coding agents?

• The massive gold rush in selling data environments to labs

• Should enterprises own their own intelligence?

• Is model routing a defensible standalone business?

• $20M+ compensation packages for exceptional new grads

• The #1 portfolio company Erik would bet on

• What Erik changed his mind on most this year